JPH01263555A - Apparatus and method for monitoring image of fish - Google Patents

Abstract

PURPOSE:To judge abnormal water quality with good accuracy, by a method wherein a certain set value is provided in the depth-of-water direction of a breeding water tank and the frequency of fish floating in a state exceeding said set value is statistically processed on the basis of the binary image thereof. CONSTITUTION:The water specimen is supplied to an environment control tank 1 through a water distributing pipe 2A and a pump 2B and the temp. thereof is detected by a temp. sensor 4 and the concn. of dissolved oxygen thereof is detected by a sensor 6. These detection values are processed by a system processor 43, and the temp. and the concn. of dissolved oxygen are controlled by a radiator 5A and an air diffuser 7A while the controlled water specimen is supplied to a water tank 10 through a supply pipe 11 and a water supply pump 12. Fish 14 (14A-14C) is bred in said water tank 10 and the image of a breeding space 19 is taken by an imaging device 20. A certain set value is provided in the depth-of-water direction of the water tank 10 and the signal of the fish 14 photographed by the device 20 is guided to an image processor 40 to recognize an image. Statistical processing is performed on the basis of the binary image thereof by a processor 42 and the behavior of the fish 14 is monitored and abnormal water quality is judged precisely.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image monitoring device for fish, and in particular, for monitoring fish in inflow water and treated water of water purification plants and sewage treatment plants, river water, etc. The present invention relates to a fish image monitoring device and monitoring method that recognizes and determines the presence or absence of poisonous substances.

[Conventional technology]

At water treatment plants, in order to determine whether or not toxic substances have been mixed into the raw water, a portion of the raw water is channeled into an aquarium where fish such as crucian carp, carp, dace, tanager, haya, and oysterfish are raised in this aquarium. . That is, when a poisonous substance is mixed into the raw water, the phenomenon that fish behave abnormally or die is used to monitor the inflow of the poisonous substance into the raw water. In addition, sewage treatment plants need to know whether legally prohibited toxic substances have entered the sewage water, so they rely on intermittent manual water quality analysis. Similarly, treated water from water treatment plants and sewage treatment plants, as well as river water, needs to be monitored for the presence of toxic substances.

As described above, monitoring of toxic substances in water currently relies on human visual inspection and complicated manual analysis. For this reason, there was a drawback that continuous monitoring and early detection were not possible.

To monitor fish, there is a method of detecting fish in an aquarium from the top of the tank using an industrial television camera (ITV) and processing the image (Reference:
36th Kinkuni Waterworks Research Presentation, 3J Collection p464-46
6) has been devised. In this method, a method is shown in which a fish is illuminated from the top of the aquarium and an image is similarly taken from the top. However, since this method cannot detect fish unless they rise to the surface of the water, it is difficult to detect abnormalities in fish behavior at an early stage.

Japanese Unexamined Patent Publication No. 61-46294 discloses a method of monitoring behavioral patterns of a plurality of living things in order to monitor water quality. The disclosed technology describes that a water quality sensor measures temperature and water quality, and the measured values are used as a reference to determine whether a behavioral pattern is normal or abnormal. The behavioral pattern is
It is stated that speed and position are also included. However, even if we simply analyze behavioral patterns using image technology,
Although such a concept is conventionally known, it is difficult to implement it because it does not disclose how to specifically detect or evaluate the speed and position of multiple living things.

[Problem to be solved by the invention]

In the above-mentioned conventional technology, fish in a breeding tank are detected using an industrial camera, and when the fish float to the surface, it is determined that there is an abnormality, but it is unclear what kind of change is caused.

An object of the present invention is to image-monitor the swimming behavior of fish.

It is an object of the present invention to provide a practical fish image monitoring device and monitoring method that can statistically process the results and determine abnormal water quality with high accuracy based on the results.

[Means to solve the problem]

The present invention provides a certain setting value in the water depth direction of the breeding tank,
This is an image monitoring device for fish that determines abnormal water quality by statistically processing the frequency of fish rising above the set value based on the binary image. Furthermore, comparison standard values for determining abnormal water quality were appropriately set.

[Effect]

According to the fish image monitoring device of the present invention, a certain set value is set in the water depth direction of the breeding tank, and the frequency of fish rising above this set value is statistically processed based on the binary image. Since the water quality is determined, water quality abnormality detection can be performed with high accuracy.

〔Example〕

The present invention will be described in detail below using the drawings.

The configuration and operation of the embodiment will be briefly explained using FIG.

Test water is supplied to the environmental control tank 1 by a water pipe 2A and a pump 2B. Excess water is drained through drain pipe 2C. The stirrer 3 stirs the test water using stirring blades 3A. A temperature sensor 4 such as a thermistor detects the temperature of the water to be tested. What is the detected temperature?

It is processed by the system processor, and the temperature is controlled to be constant by adjusting the radiator output device 5 based on the signal from the system processor and controlling the amount of heat radiated from the radiator 5A. As a temperature control method, known techniques such as an on/off control method and a PID control method can be easily used.

The target value of the temperature to be adjusted is set to a temperature suitable for the fish kept in the aquarium 10 to be active. The dissolved oxygen concentration of the test water is detected by the sensor 6. The detected value is processed by the system processor, and based on the signal from there, the air pump 7 supplies air bubbles from the air diffuser 7A to adjust the dissolved oxygen concentration. Air pump 7 supplies air. if,
If dissolved oxygen in the test water is supersaturated, the dissolved oxygen concentration is reduced by supplying air.

The test water whose temperature and dissolved oxygen concentration are suitable for raising fish is transferred to the aquarium 10 by the supply pipe 11 and the water supply pump 12.
supplied to The water introduced into the water tank 10 is drained through a drain pipe 13. Inside the aquarium lo, a breeding space 1 is partitioned by partition plates 18A and 18B such as wire mesh or perforated plates.
9, and fish 14A, 14B, and 14C are raised here. Although the present embodiment will be described here for the case where the fish have two legs, it goes without saying that the present embodiment can be similarly applied to a case where there are many more fish. The lighting device 15 illuminates the fish 14 in the aquarium 10. A semitransparent plate 16 made of a semitransparent substance such as ground glass or paper is provided between the lighting device 15 and the aquarium 10.

Receiving the light from the illumination device 15, the transparent plate 16 scatters the light, and the light emitted from the entire semi-transparent plate 16 illuminates the aquarium 10. An imaging device 20 such as an industrial television camera (ITV) is placed on the opposite side of the water tank 1° as viewed from the lighting device 15.

That is, the imaging device 20 images the light emitted from the illumination device 15 and passed through the semi-transparent plate 16. Here, the imaging device 20
images the rearing space 19.

The signal from the imaging device 2o is guided to the image processing device 40, which recognizes the fish as an image and calculates the center of gravity and speed of the fish. The center of gravity and velocity signals are input to the system processor 42, which calculates statistical patterns of center of gravity and velocity to monitor the behavior of the fish 14. The alarm device 48 issues an alarm in the event of an abnormality based on the monitoring results obtained by the system processor 42.

The monitor television 50 displays the captured image.

The image monitor 46 receives signals from the image processing device 4o and the system processor 42, and displays image recognition results and monitoring results such as fish position distribution and speed distribution. The keyboard 44 is used to input calculation conditions for the image processing device 40 and system processor 42, as well as information for controlling the display on the image monitor 46.

Next, the configuration of image processing after the imaging device 20 will be described.

FIG. 2 is a diagram showing the configuration of the processing device, and the processing device is connected by an image processing device 40, a system processor 42, and a system system bus 52.

External storage device 54 stores data processed by system processor 42. An input/output port 56 is coupled to the system processor 42 so that signals from sensors 4 and 6 are routed to the system processor 42 via regulators 58 and 6o;
The control signal obtained by the processing operation of the actuator 5
and sent to 7. The image processing device 40 includes an image processing unit 401 having a histogram processing function, a labeling processing function, a feature extraction function, a convolution function, and other image processing functions, an image storage unit 403,
Consists of 405. Image storage unit (gradation image memory)
4o3 is a memory for the grayscale image obtained by the camera 20, and the image storage unit (binary image memory) 405 is
This is a memory for binary images processed by the image processing unit 401. One image obtained by the camera 20 is 25
Assuming that it is composed of 6 pixels x 256 pixels, the grayscale image storage unit 403 has a storage capacity of 256 pixels x 256 pixels x 8 bits, and the binary image storage unit 405 has a storage capacity of 256 pixels x 256 pixels x 1 bit. be. The image processing apparatus used by the inventors has four grayscale image storage units 403 and four binary image storage units 405.

Although the signal obtained by the camera is converted by an A/D converter, the A/D converter is not shown in FIG. 2 in order to simply represent FIG. 1.

FIG. 3 shows the processing steps for the aforementioned processing bag @ (consisting of 40 and 42). This flowchart shows the image processing device 4.
0 and system processor 42 processing operations. First, the number of repetitions m is set to 0 in step 502. m is increased by one in step 504. next,
In step 506, a grayscale image is read from the grayscale image memory 403. Next, in step 508, the grayscale image is converted to a binary image. The converted binary image is stored in binary image memory 405. Step 510 projects the binary image in the horizontal direction and integrates the number of pixels. This measures the position of the fish in the aquarium, that is, calculates the frequency of the fish's center of gravity in the horizontal direction as a position distribution. In step 512, the one-position distribution is standardized by the cumulative total number of pixels and stored in the system processor 42.

In step 512, if the time is shorter than the set one-time measurement time, the process returns to step 504 and the measurement is repeated.

In step 518, based on the calculation results in step 512, the nose-up action right m is used to determine which fish have surfaced near the water surface.
Calculation of Rw (Yc, t). In step 520, in a nose-up behavior.

Calculate the frequency distribution Fm of fish surfacing near the water surface. In step 522, a calculation is performed to obtain a water quality abnormality determination reference value Rw (Yc, t) using the nose-up behavior index. In step 524, Rw (Yc, t) and Rw” (Yc
, j), and if Rw)Rw is found, the water quality is abnormal, and an alarm is issued in step 526, and Rw>Rw.
If it is a skewer, the water quality is normal and you can return to the start.

Next, the configuration of image processing will be explained in detail. It receives the image signal output from the imaging device (camera) 20 and converts it from an analog value to a digital value.

Store the image in the grayscale image memory. Recognize the fish body from this image and use the histogram binarization method.

By setting a certain threshold, pixels brighter than this threshold are set to “0”.
” level, and conversely, dark pixels were binarized with the “1” level. Figure 4 shows an example of a grayscale image. Figure 4 is an image with multilevel luminance, and the luminance distribution is determined in step 508. Calculate. The calculation results are shown in Fig. 5. Next, the method of setting the threshold value will be explained.

With the illumination method of the present invention, the fish school 14 can always be imaged as a dark object, so as shown in FIG. Set the first threshold value.

Since the area f varies depending on the state, the minimum area is set. This threshold setting method is particularly effective when the water becomes cloudy.

However, if the water is not cloudy, it is better to use the second threshold. In FIG. 5, the peak Pf represents the fish body, the peak pb represents the background, and the portion represented by Pe represents the gills and outline of the fish. In order to extract only the fish body, a second threshold value 2 is set at the boundary between Pf and Pe. As shown in Fig. 5, set the threshold value to at least 11 in brightness in advance, search each frequency in the direction of increasing brightness, and if there is a boundary (minimum value) between Pf and Pe, add I2 to this brightness. choose. Next, the specific operation of the image processing unit 401 will be explained. In step 508, the brightness G of step 506 is
(ITJ), all pixels brighter than the threshold are
OTT level, and conversely, all pixels darker than the threshold are set to 1.
This signal is always stored in the image storage unit 403 as the 1111 level except for the first time. Letting this set of binarized signals be B(II j), the binarization calculation is expressed by the following equation.

If G(1#J)≧1, B (x + j) = O・
...(1) If G(1+j)<1, then B(j, j)=
1 ... (2) (1) By calculating the formula (2) for each pixel, the background can be set to the 110" level and the fish school 14 to the "1" level. Binarize Fig. 4 The results are shown in Figure 6.

Step 510 receives the signal from step 508 and moves to fish school 1.
Calculate the position distribution of 4. The position distribution in FIG. 6 is defined by the distribution obtained by horizontally projecting the image in FIG. 4, as shown in FIG. That is, the position distribution shown in FIG. 7 represents the depth at which the school of fish was located in the water depth direction. In other words,
The obtained position distribution is a distribution representative of the positions of fish schools.

Step 512 receives the result of step 510 and adds up the position distributions measured at each time interval so that an average position distribution can be calculated. This repetition is based on step 514, and the above-described series of processes is performed a predetermined number of times m.

As a result of step 516, an example of the standardized position distribution becomes a distribution with peaks as shown in FIG.

Next, in step 518, a nose-up behavior index of the fish is calculated. Here, t is the number of times of monitoring.

The frequency at which fish are located at water depth Y+ is expressed as relative frequency (%) by FW (YI+t). Since this position distribution Fw (Ys, t) in the water depth direction is a vector quantity, the position of the fish cannot be directly evaluated from it.

Therefore, in order to quantitatively evaluate the position distribution in the water depth direction as a scalar quantity, as illustrated in the ninth example, the proportion Rw (Yc, t) of fish above a predetermined water depth Yc is expressed as (3)
Calculate by formula.

Rw (Ys, t): The proportion of fish that were between water depths Yl and yc in the second monitoring (nose raising behavior index)
Relative frequency (%) of fish located at Yc: Specific position in water depth direction From this, Yc is the coordinate near the water surface, and Rw
A large value of (Yc, t) indicates that fish are frequently present near the water surface. for example.

In the case of nose-raising behavior, Rw (Ycst) takes a large value.

Next, since the frequency with which fish surface near the water surface can be considered a stochastic event, the certainty of this frequency distribution depends on the size of the group. For example, the maximum value of Rw (Yc, t) when water quality is normal varies depending on the length of the measurement period. Therefore,
In order to clarify these relationships, Rw (Yet t
) value frequency distribution is calculated in step 520. The results are illustrated in FIG. From this figure, Rw (Yc, t)
The inventors have discovered that the frequency distribution of values can be approximated by an exponential distribution. Therefore, a general frequency distribution of Rw (Y c t t ) values is assumed by equation (4), and data that deviates from this distribution is regarded as abnormal.

Ft=a 6 Rw (Yc, t,) +b
-(4) Here, a and b are constants that change depending on the type of fish, growth conditions, water quality conditions, etc.

In step 522, equation (4) is used to calculate a method for determining water quality abnormality. First, the relative frequency that can occur once when monitoring is performed n times is expressed by equation (5).

Fx=Qn (1/n)
-(5) Set the Rw (Yc, t) value at this time to the Rw umbrella (
Yc, t) and called the critical straight-up action index value.

To find this Rw'' (Yc, t) value, (
All you have to do is substitute equation 5) into equation (4) and solve for n, and the result is equation (6) %Formula %If Rw occurs only with a probability of - times in one month (this time can be set arbitrarily) The skewer (Ycv t) is regarded as abnormal, and this is used as the abnormality determination reference value.

Find an example of Rw table (Yc, t). If monitoring is continued every ten minutes, one month will be 4320 times (30 days/month x 24 hours/day x 6 times/hour = (4320 times in July), so n = 43
20 and calculate equation (6), Rw skewer (Y
c, t) values can be obtained. Incidentally, here, the description is given using a representative example of n in -month.

The n value is not particularly limited, and the sensitivity of anomaly detection can be tuned by adjusting the n value setting.If the n value is small, the probability of abnormality detection increases, but the detection sensitivity decreases. However, for this reason, it is desirable to update Rw(Yet t ) sequentially. In step 524, Ry(Yet t))Rw*(Yc,
t), the water quality is abnormal and an alarm is issued in step 526, and Rw (Y c Ht ) < Rw fist (Yc
, t), the water quality is normal and the process returns to the start.

An embodiment using the image processing described above will be described next.

Using the fish species Tanago and Carp, the nose-up behavior index RW (
YC, M) was calculated using equation (3). The results are shown in Figure 12. In FIG. 12, the horizontal axis is the measurement time, the vertical axis is the nose-up behavior index Rw (Yc, t), and the specific position Yc:3 in the water depth direction. The measurement time was 7 days for the Japanese carp (a), 3 days for the carp (b), and 16 hours for the Japanese carp (c). In each case, harmful substances were injected into the water one hour before the end of the measurement. As a result, when water quality is normal, Rw (Ya, t)
The value does not exceed 0.2, but when looking at changes over time when water quality is abnormal due to the injection of harmful substances, Rw (Yc, t)
The values for each of (a), (b), and (C) rise steeply, which clearly indicates an abnormal state. Next, based on the results shown in FIG. 12, the frequency distribution of the nose-up behavior index was calculated using equation (4). The results are shown in FIG.

Figure 13 shows the frequency distribution when the water quality is normal and when it is abnormal.

Fm on the vertical axis is a logarithmic representation of relative frequency. Values when water quality is abnormal are indicated by diagonal lines. As a result, Fig. 13(a)
, The frequency distribution when the water quality is normal in (b) is expressed as a straight line,
The distribution during abnormal conditions deviates from this. In addition, Fig. 13(e)
In this case, because the measurement time when the water quality was normal was short, the frequency distribution of Rw was displayed only in one class (left end), and the frequency distribution could not be expressed as a straight line. As described above, it is clear that in order to obtain the frequency distribution of the nose-up behavior index, it is necessary to lengthen the measurement time when the water quality is normal to some extent.

The relative frequencies F in FIGS. 13(a) and 13(b) are expressed by the following equations.

Tanago (a) −F t = (30,3) Rw
1.358 − (7) Carp (b) −F m =
(43, 4) Rw O, 227-(8) Also, the critical straight-up action index value Rm umbrella (Yc, t) is calculated as the abnormality determination reference value. When monitoring was continued every ten minutes for one month, Rw* (Yc, t) was obtained from n=4320: Tanago (a)...0.19 Carp (b)...0.16. Therefore, Rw(Yc, t)>Rm books(Yc
.

t) can be determined as water quality abnormality.

Furthermore, we also investigated other fish species, crucian carp and Japanese dace, and obtained the same results as above. Also, a mixture of tanager, carp, crucian carp, and dace.

Similar results were obtained.

<Loneliness of Loss> Here, the influence of dissolved oxygen concentration (Do) on fish behavior patterns, as a water quality environmental factor other than harmful substances, was analyzed using the image processing of the present invention. The fish used in the experiment was a Japanese dace, and the water temperature was 15°C. The experimental results are shown in the 14th
Shown in the figure. The vertical axis in Figure 14 is the nose-raising behavior index Rw.
(Yc, t) Tori.

The value, the horizontal axis is the measurement time. The experiment was conducted at a water temperature of 15℃ in the aquarium.
The Japanese dace were kept at a saturated level of about 7 to 8 ppm for two days, and then argon gas was blown into the water to gradually lower the Do value in the water. 14th
The figure shows normal conditions up to measurement 60n+in, and the Do value decreased after 60m1n. Although the Rw (Yc-L) value may rise to around 0.1 when the water quality is normal, it is not a major change. On the other hand, a change occurs when the Do value is around 0.8 ppm, and Rm (Yc, t) at 0 and 5 ppm.
stood up suddenly. Although not shown here, similar results were obtained for other fish species such as carp, crucian carp, and tanago.

<Loss of Fire> As a water quality environmental factor other than harmful substances, the influence of water temperature on fish behavior patterns was analyzed here using the image processing of the present invention. The results are shown in FIG. The experiment used Japanese dace, and the Do value was kept at saturation. The figure shows the mixture of harmful substances at each set water temperature (in this experiment, potassium cyanide was 0)
, 1 mg/Q) showed the image recognition area of abnormal behavior patterns of fish. For example, at a water temperature of 20°C, it takes about 5 minutes for image recognition to be abnormal, but at a water temperature of 10°C, it is approximately sufficient, and at a water temperature of 5°C, it is more than enough. Therefore,
By keeping the water temperature around 20 degrees Celsius, except for cold-water fish such as rainbow trout, it is possible to monitor abnormal fish behavior early.

Note that in the present invention, the adjustment of Do and water temperature has been described, but in other embodiments, adjustment of PH + chlorine concentration and turbidity (
(not shown).

[Effects of the Invention] According to the present invention, since the behavior of fish can be image-monitored, abnormalities in the behavior patterns of living things caused by the influx of harmful substances or other factors can be detected accurately and with high precision. .

[Brief explanation of the drawing]

Fig. 1 is a perspective view of an embodiment of the present invention, Fig. 2 is a block diagram of an image processing device, Fig. 3 is a flowchart of the processing steps of the image processing device, Fig. 4 is a fish imaging diagram, and Fig. 5 is a luminance diagram. Frequency distribution diagram, Figure 6 is a two-image capture diagram, Figure 7 is a horizontal distribution diagram of the image in Figure 6, Figure 8 is a distribution diagram of fish locations in the water depth direction, and Figure 9 is a diagram of fish locations. Explanation of the nose-raising behavior index Figures 10 and 11 are frequency diagrams of the nose-raising behavior index, and Figure 12 is a diagram showing an example of experimental results summarized by the nose-raising behavior index.
Figure 13 is a frequency distribution diagram of the nose-up row index, and Figure 14 is the DO
FIG. 15 is a diagram showing an example of the experimental results of the image recognition area due to temperature change. 1...Environment control tank, 4...Temperature sensor, 6...D
o Sesa, 10... Aquarium, 14... Fish, 20... Imaging device (mera), 40... Image processing device, 42...
System processor, 44...keyboard, 46...image monitor 48...alarm device, 50...monitor television.

Claims (1)

[Scope of Claims] 1. An environment control tank that maintains constant environmental characteristics and water quality of test water, an aquarium through which the test water led from the environment control tank flows, and fish reared in the aquarium. , an imaging device that converts image information of the fish into electrical signals at regular time intervals, a lighting device that illuminates the fish, an image processing device that recognizes the fish from the imaging device, and an image processing device that uses the image processing device. a system processor that monitors the behavior of fish from images obtained by the system; and an alarm device that notifies the inflow of harmful substances and abnormalities caused by water quality environmental factors other than harmful substances based on the monitoring results of the system processor. A fish image viewing device characterized by: 2. An image monitoring method for fish, comprising means for calculating a frequency distribution based on a nose-up behavior index calculated based on an image processing device and a system processor. 3. Measure the frequency distribution of the nose-up behavior index when water quality is normal,
3. The image monitoring method for fish according to claim 2, wherein the frequency distribution is used as a reference value for determining water quality abnormality. 4. The image monitoring method for fish according to claim 2, wherein the frequency distribution of the nose-up behavior index is approximated by an exponential distribution. 5. The image monitoring method for fish according to claim 4, wherein the determination reference value is determined by measuring the nose-up behavior index during normal water quality for 1 to 10 hours. 6. The image monitoring method for fish according to claim 4, wherein the determination reference value is updated every 1 to 3 days. 7. The fish image viewing device according to claim 1, which controls the dissolved oxygen concentration in the test water so as not to fall below 0.8 ppm. 8. The fish image monitoring device according to claim 1, wherein the water temperature of the water to be tested is controlled within a range of 15 to 25°C.