KR102528286B1 - Automatic feeding system and method for fish farms based on AI - Google Patents

Abstract

The present invention relates to an automatic feeding technology for an AI-based fish farm that can objectively analyze information obtained while operating a farm, determine an optimal feed supply amount, and supply feed in the farm based on this.

Description

Automatic feeding system and method for fish farms based on AI}

The present invention relates to an automatic feeding system and method for an AI-based fish farm, and an AI-based fish farm capable of determining an optimal feed supply amount according to water quality information of the farm and movement state information of fish in the farm using an AI model. It relates to an automatic feeding system and its method.

Usually, in order to farm fish, it is limited to supplying a fixed amount of feed at a fixed time or additional feeding of feed by visually checking the movement of the fish in light of the farmer's experience.

Naturally, this method is highly likely to supply insufficient or excessive feed because the optimal feeding amount for each fish species is not supplied. may also be the main cause of

In this regard, Korean Registered Patent No. 10-0821882 ("Automated Feed Quantitative Feed Supply System for Tank-type Fish Farm") discloses a technology capable of quantitatively supplying feed for tank-type fish farms through remote control using a communication network.

Korean Patent Registration No. 10-0821882 (Registration date 2008.04.07.)

Therefore, the present invention has been made to solve the above problems, and an object of the present invention is to objectively analyze the information obtained while operating the farm, and to automatically feed the AI-based farm, which can determine the optimal feed supply amount. system and its method.

In particular, rather than setting a time for feeding, AI analyzes the condition of fish in the tank in real time and feeds in both directions according to the fish's response, rather than just feeding unilaterally. It relates to an automatic feeding system and method for a farm based farm.

An AI-based fish farm automatic feeding system according to an embodiment of the present invention to solve the above problems is a data input unit 100 that receives at least one of water quality related information and image related information about the farm, and stored The AI analysis unit 200 extracts information on the amount of feed supplied to the farm by inputting the information by the data input unit 100 using the AI model, and the amount of feed supplied by the AI analysis unit 200 and a supply control unit 300 generating a control signal for controlling the feed supply unit in the farm, based on It is desirable to carry out feeding.

Furthermore, the automatic feeding system of the AI-based farm collects in advance water quality-related information according to the amount of feed in the farm and information related to images of fish movements according to the feed supply in the farm, and transmits them to the pre-stored AI network. It is preferable to include an AI processing unit 10 that performs learning processing by

Furthermore, the AI processing unit 10 includes a first learning data collection unit 11 that collects water quality related information including water temperature information and dissolved oxygen information according to a plurality of preset feed supply amount information for each stage, and pre-stored first and a first learning processing unit 12 for performing learning processing by inputting information from the first learning data collection unit 11 to the AI network, and It is preferable to store 1 AI model in the AI analysis unit 200.

Furthermore, the AI processing unit 10 includes image information photographing the movement of fish in the water and on the surface of the water at predetermined times for each of a plurality of stages in a time series after the feed is supplied in the farm. The second learning data collection unit 13 that collects image-related information and the second learning processing unit that inputs the information by the second learning data collection unit 13 to the pre-stored second AI network and performs learning processing ( 14), and it is preferable to store the second AI model based on the learning result of the second learning processing unit 14 in the AI analysis unit 200.

Furthermore, the data input unit 100 includes a sensing means installed inside the water tank of the farm, and a water quality input unit that receives water quality related information including water temperature information and dissolved oxygen information from the sensing means in units of predetermined time. It is preferable to include (110).

Furthermore, the AI analysis unit 200 inputs the water quality related information from the water quality input unit 110 to the first AI model, and outputs information on the stage of feeding amount of feed in the farm predicted according to the water quality related information. It is preferable to include a first analysis unit 210 to do.

Furthermore, the data input unit 100 includes photographing means respectively installed inside and outside the water tank of the farm, and is related to an image including image information obtained by photographing the movement of fish in units of predetermined time from each photographing means. It is preferable to further include an image input unit 120 for receiving information.

Furthermore, the AI analysis unit 200 inputs image-related information from the image input unit 120 to the second AI model and outputs fish movement stage information predicted according to the image-related information. It is preferable to further include an analysis unit 220 .

Furthermore, the AI analyzer 200 uses the feed feeding amount stage information predicted by the first analyzer 210 and the fish movement stage information predicted by the second analyzer 220. It is preferable to further include a final analysis unit 230 that performs integrated analysis and extracts information on the final feed amount to be fed into the farm.

Furthermore, the final analyzer 230 analyzes the sharpness of the image quality of the image-related information by the image input unit 120, and based on the analyzed sharpness of the image quality, the fish predicted by the second analyzer 220 It is preferable to extract the final feeding amount information by controlling the weight of the motion step information.

An automatic feeding method for an AI-based fish farm by an AI-based fish farm automatic feeding system in which each step is performed by an arithmetic processing means according to an embodiment of the present invention, wherein one of water quality related information and image related information for the farm A data input step (S100) of receiving an abnormality, an AI analysis step (S200) of extracting feed amount information from the farm by analyzing the information from the data input step (S100) using the stored AI model, A feeding control step (S300) of generating a control signal for controlling the feed supply means in the farm based on the feed amount information by the AI analysis step (S200) and the feeding control step (S300) In accordance with the control signal by the control signal, it includes a feeding step (S400) of performing feed supply in the farm, and prior to performing the AI analysis step (S200), information related to water quality according to the amount of feed in the farm and farm It is preferable to include an AI processing step (S10) of collecting image-related information obtained by photographing fish movements according to my feeding, performing learning processing by a pre-stored AI network, and storing an AI model.

Furthermore, the AI processing step (S10) collects water quality-related information including water temperature information and dissolved oxygen information according to a plurality of preset feed supply amount information for each step, and inputs the collected information to the first pre-stored AI network. and a first learning step (S11) of performing the learning process by doing so, and it is preferable to store the first AI model based on the learning result of the first learning step (S11).

Furthermore, the AI processing step (S10) includes image information photographing the movement of fish in the water and on the surface of the water at predetermined times for each of a plurality of steps in a time series after the feed is supplied in the farm. and a second learning step (S12) of collecting image-related information to be stored and inputting the collected information to a pre-stored second AI network to perform learning processing, and by the learning result of the second learning step (S12) It is preferred to store the second AI model.

Furthermore, the data input step (S100) is a water quality input step (S110) of receiving water quality related information including water temperature information and dissolved oxygen information in predetermined time units using a sensing means installed inside the water tank of the farm. ) and an image input step of receiving image-related information including image information of the movement of fish by using the photographing means installed inside and outside the tank in the farm, in units of predetermined time pre-stored from each photographing means. It is preferable to include (S120).

Furthermore, the AI analysis step (S200) uses the first AI model to enter the water quality-related information from the water quality input step (S110), and the feed feeding amount of the farm is predicted according to the water quality-related information. The first analysis step (S210) of outputting step information and the fish movement step predicted according to the image-related information by inputting the image-related information through the image input step (S120) using the second AI model It is preferable to include a second analysis step (S220) of outputting information.

Furthermore, the AI analysis step (S200) includes the predicted feed feeding amount stage information of the farm by the first analysis step (S210) and the predicted fish movement stage information by the second analysis step (S220). It is preferable to further include a final analysis step (S230) of integrally analyzing and extracting the final feed amount to be fed into the farm.

Furthermore, the final analysis step (S230) analyzes the image quality sharpness of the image-related information by the image input step (S120), and fish predicted by the second analysis step (S220) based on the analyzed image quality sharpness. It is preferable to extract the final feeding amount by controlling the weight of the motion step information of .

Furthermore, in the first analysis step (S210), for the water quality-related information input by the water quality input step (S110), based on a preset maximum value or minimum value, the water quality-related information exceeding the maximum value is set to the maximum value. , and mapping information related to water quality that is less than the minimum value to the minimum value, and inputting it to the first AI model.

An automatic feeding system and method for an AI-based fish farm according to an embodiment of the present invention objectively analyzes information acquired while operating a farm (water quality status information of the farm and movement status information of fish in the farm), It has the advantage of being able to determine the amount of feed supply.

In addition, there is an advantage in that optimal growth can be derived by continuously analyzing the necessary feed amount information even after the feed is supplied, that is, by performing the feed according to the reaction of the fish after the feed is supplied.

1 is an exemplary configuration diagram showing an automatic feeding system for an AI-based farm according to an embodiment of the present invention.
2 is an exemplary sequence diagram illustrating an automatic feeding method for an AI-based farm according to an embodiment of the present invention.

Objects, features and advantages of the present invention described above will become more apparent through the following examples in conjunction with the accompanying drawings. The following specific structural or functional descriptions are only illustrated for the purpose of explaining embodiments according to the concept of the present invention, and embodiments according to the concept of the present invention can be implemented in various forms and are described in this specification or application. It should not be construed as being limited to the examples. Embodiments according to the concept of the present invention can apply various changes and can have various forms, so specific embodiments are illustrated in the drawings and described in detail in this specification or application. However, this is not intended to limit the embodiments according to the concept of the present invention to a specific form disclosed, and should be understood to include all modifications, equivalents, or substitutes included in the spirit and scope of the present invention. Terms such as first and/or second may be used to describe various components, but the components are not limited to the above terms. The above terms are used only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the present invention, a first component may be named a second component, and similar Happily, the second component may also be referred to as the first component. It should be understood that when a component is referred to as being connected or connected to another component, it may be directly connected or connected to the other component, but other components may exist in the middle. On the other hand, when it is mentioned that a certain component is directly connected to another component or is directly connected, it should be understood that no other component exists in the middle. Other expressions for explaining the relationship between components, such as 'between' and 'directly between' or 'adjacent to' and 'directly adjacent to' should be interpreted similarly. Terms used in this specification are only used to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms include or have are intended to indicate that the described features, numbers, steps, operations, components, parts, or combinations thereof exist, but one or more other features, numbers, steps, operations, It should be understood that it does not preclude the presence or possibility of addition of components, parts, or combinations thereof. Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this specification, it should not be interpreted in an ideal or excessively formal meaning. don't Hereinafter, the present invention will be described in detail by describing preferred embodiments of the present invention with reference to the accompanying drawings. Like reference numerals in each figure indicate like elements.

In addition, a system refers to a set of components including devices, mechanisms, and means that are organized and regularly interact to perform necessary functions.

1 is a diagram showing an automatic feeding system for an AI-based farm according to an embodiment of the present invention. Referring to FIG. 1 , the automatic feeding system of an AI-based farm according to an embodiment of the present invention will be described in detail.

As shown in FIG. 1, the automatic feeding system of an AI-based fish farm according to an embodiment of the present invention preferably includes a data input unit 100, an AI analysis unit 200, and a supply control unit 300. do. It is preferable that each component performs an operation by being individually or integrally included in at least one arithmetic processing unit including a computer or the like.

For a detailed look at each component,

It is preferable that the data input unit 100 receives at least one of water quality related information and image related information about the farm.

In detail, the data input unit 100 includes a water quality input unit 110 and an image input unit 120 as shown in FIG. 1 .

The water quality input unit 110 preferably includes at least one sensing unit installed inside the water tank of the farm.

The water quality input unit 110 receives water quality related information including water temperature information and dissolved oxygen information in units of a predetermined time. At this time, in addition to water temperature information and dissolved oxygen information, pH concentration, salinity, ammonia, and nitrite information may be further included.

It is preferable that the image input unit 120 includes a photographing means installed inside and outside the tank of the farm, respectively.

The image input unit 120 receives image-related information including image information obtained by photographing the movement of fish in units of a preset predetermined time from each photographing means. That is, information for photographing the movement of fish in the water and information for photographing the movement of fish on the surface of the water are respectively input. Here, the motion photographing information of the fish on the water surface does not mean only the motion photographing information of the fish seen from outside the water, but also captures a motion of moving up to the surface of the water and opening its mouth.

The AI analysis unit 200 uses a pre-stored AI model, inputs the information from the data input unit 100, and extracts feed supply amount information of the farm.

At this time, in the AI-based fish farm automatic feeding system according to an embodiment of the present invention, it is preferable to generate and store an AI model by the AI processor 10 prior to analysis by the AI analyzer 200. do.

To this end, the AI processing unit 10 collects information related to water quality according to the amount of feed supplied in the farm and information related to images of fish movements according to the supply of feed in the farm in advance, and learns and processes the pre-stored AI network. will perform

In detail, as shown in FIG. 1, the AI processing unit 10 includes a first learning data collection unit 11, a first learning processing unit 12, a second learning data collection unit 13, and a second learning data collection unit 11. It will include a processing unit (14).

The first learning data collection unit 11 collects water quality related information including water temperature information and dissolved oxygen information according to a plurality of previously set feed supply amount information for each stage.

In other words, the first learning data collection unit 11 collects the water quality-related information for each step (different feed supply amount) by differentiating the feed supply amount for each of a plurality of preset stages. In this case, the water quality-related information may further include pH concentration, salinity, ammonia, and nitrite information in addition to water temperature information and dissolved oxygen information.

Here, the automatic feeding system of the AI-based fish farm according to an embodiment of the present invention sets 10 steps as preset steps, but this is only an embodiment of the present invention, and is not necessarily limited to 10 steps. , the size of the farm, the type of fish, etc., are freely set.

Most preferably, after receiving input from the outside (aquaculture operator, aquaculture expert, etc.) on the maximum feed supply amount and minimum feed supply amount that can be paid, taking into account the size of the farm, the type of fish, the number of fish, etc. It is preferable to divide by the number uniformly. That is, it is preferable to divide the feed supply amount by the set number of steps so as to reach from the minimum feed supply amount to the maximum feed supply amount or from the maximum feed supply amount to the minimum feed supply amount.

Through this, the first learning data collection unit 11 senses the water quality state (water temperature information, dissolved oxygen information, pH concentration, salinity, ammonia, nitrite information) according to the feed supply amount set differently for each step, and then feed It is desirable to collect supply amount data-water quality state sensing data.

Of course, it is preferable to process the collected data as learning data for labeling by matching water quality condition information for each feed supply step (feed supply amount step).

Through the learning process of such learning data, an AI model for predicting the amount of currently supplied feed is generated using real-time input or water quality state information input at a desired time.

In detail, the first learning processing unit 12 inputs information by the first learning data collection unit 11 to the first AI network stored in advance to perform learning processing.

Here, the pre-stored first AI network is preferably an inception time algorithm that uses multivariate time series data as an input, but this is also only one embodiment of the present invention, and multivariate Any AI network that can receive time series data and output probability values for each stage can be applied.

The first learning processing unit 12 receives water quality-related information at a predetermined number of stages (each feeding stage), which is learning data by the first learning data collection unit 11, and performs learning, thereby performing learning at a specific point in time. When water quality related information is input, a first AI model capable of predicting/outputting the amount of feed supply (feed supply stage) at the corresponding time is generated.

Through this, the AI analysis unit 200 stores the first AI model based on the learning result of the first learning processing unit 12 .

More specifically, when water quality-related information at a specific point in time is input to the first AI model by the first learning processing unit 12, a probability value corresponding to each step is output based on each learned step, The step with the highest probability is predicted as the feed supply amount (feed supply step) at the corresponding time and output.

In addition, the second learning data collection unit 13 captures image information obtained by photographing the movements of fish in the water and on the surface of the water at predetermined times for each of a plurality of steps set in advance in a time series order after the feed is supplied in the farm. It collects video-related information.

In other words, after feeding the fish, the second learning data collection unit 13 acquires the movement of the fish in the water/on the water surface for each preset time unit.

Here, the automatic feeding system of the AI-based fish farm according to an embodiment of the present invention sets 5 steps as preset steps, but this is only an embodiment of the present invention, and is not necessarily limited to 5 steps. , the size of the farm, the type of fish, etc., are freely set.

However, the second learning data collection unit 13 analyzes the degree of movement of the fish from immediately after receiving the feed (satisfied state) until it feels hungry over time, in other words, at a predetermined interval (hunger state). Data is collected to create an AI model for predicting the current hunger state using the movement information of the fish at every time interval that can divide the degree of hunger.

Unlike the first learning data collection unit 11, which can relatively objectively divide the amount of feed and set the level for labeling, since the degree of fish hunger is set in a relatively subjective level, immediately after feeding is set as step 1, and a certain level is higher than that. The point in time when the interval has passed is set as step 2, and motion status information is sequentially collected/obtained in up to 5 steps by repeating this process, and then labeled and processed as learning data.

At this time, for a series of processes for acquiring movement state information of fish sequentially at predetermined intervals from immediately after feeding, at least once after supplying feed according to instructions from an outside (farm operator, aquaculture expert, etc.) , It is desirable to collect data by photographing the movement state of fish at predetermined time intervals set by the input.

After that, learning data is generated through a labeling process for this.

Here, the movement state information of the fish includes how many fishes have passed through a specific area set in advance for a predetermined time, or at least how many fishes have moved in the specific area (the number of movements of fins), or a predetermined time period. During this process, how many fish came up to the surface of the water and bitten (bite) is photographed, and the movement itself is captured as a feature to create learning data.

The second learning processing unit 14 inputs the information by the second learning data collection unit 13 to the pre-stored second AI network and performs learning processing.

Here, the pre-stored second AI network is preferably a convolution LSTM algorithm, but this is also only an embodiment of the present invention, and any AI network used for classification of time-series behavioral changes can be applied.

The second learning processing unit 14 performs learning by receiving information related to images of motions of fish taken at a plurality of preset steps, which are learning data by the second learning data collection unit 13, and performs learning. When motion-related information is input, a second AI model capable of predicting/outputting the fish's hunger stage (time interval stage after feeding) at the corresponding time point is created.

Through this, the AI analysis unit 200 stores the second AI model based on the learning result of the second learning processing unit 14 .

More specifically, the second AI model by the second learning processing unit 14 outputs a probability value corresponding to each step based on each learned step when information related to an image at a specific point in time is input, The stage with the highest probability is predicted as the fish's hunger stage (time interval stage after feeding) and output.

In this way, using the generated first AI model and second AI model, the AI analysis unit 200 inputs the information from the data input unit 100 and extracts information on the feeding amount of feed in the farm. .

As shown in FIG. 1 , the AI analysis unit 200 includes a first analysis unit 210 , a second analysis unit 220 and a final analysis unit 230 .

The first analysis unit 210 inputs the water quality related information from the water quality input unit 110 to the first AI model, and outputs the feed supply amount stage information of the farm predicted according to the water quality related information. .

As described above, when water quality-related information at a specific time point is input, the first AI model can predict/output the amount of feed supply (feed supply step) at that point in time, and thus water quality by the water quality input unit 110. Relevant information is input, and the feed supply amount time point (feed supply amount stage information) predicted according to the corresponding water quality related information is output.

At this time, the output value is output by predicting the step having the highest probability among a plurality of steps as the feed supply amount (feed supply step) at the corresponding time.

In addition, it is preferable that the first analysis unit 210 pre-processes the water quality related information by the water quality input unit 110 and then inputs the information to the first AI model.

That is, the pre-processing is performed in consideration of the fact that an error in the sensing unit may occur.

In detail, the water quality input by the water quality input unit 110 based on the maximum value (for example, the maximum temperature of the water, the maximum amount of dissolved oxygen, etc.) and the minimum value of the water quality-related information set in advance for each fish in the tank. If the relevant information exceeds the maximum value (exceeds the maximum value), it is mapped to the maximum value, and if it exceeds the minimum value (below the minimum value), it is mapped to the minimum value.

At this time, if water quality-related information that exceeds the maximum value or minimum value is input continuously for more than a predetermined number of times, it is judged not to be an error of the sensing means, but to an abnormality in the water quality condition in the tank, and an external manager (farmer, etc.) An alarm related to water quality abnormalities is sent to the user.

In addition, the first analyzer 210 does not set the maximum and minimum values of water quality-related information for each fish in the tank in advance, but recognizes the fish in the tank by using a photographing means installed inside the tank of the farm. In addition, it is possible to automatically set the maximum and minimum values of water quality-related information for each fish in the tank and utilize them. In this case, there is an advantage in minimizing human errors that may occur in the process of setting fish.

In addition, in this way, when the fish in the tank is automatically recognized or the fish in the tank is input and set in advance, various fish, not a single fish, may exist according to the farm situation.

In this case, the first analyzer 210 may set the minimum value among preset maximum values for each fish as the representative maximum value, and set the maximum value among preset minimum values as the representative minimum value.

The second analysis unit 220 inputs image-related information from the image input unit 120 to the second AI model, and outputs fish movement stage information predicted according to the image-related information.

Again, when movement-related information at a specific time point is input, the second AI model can predict/output the fish's hunger stage (time interval stage after feeding) at that time point, so that the video input unit 120 It receives image-related information by and outputs the movement stage of the fish (time interval stage after feeding/hunger stage) predicted according to the corresponding image-related information.

At this time, the output value is output by predicting the step having the highest probability among a plurality of steps as the movement step of the fish (time interval step after feeding/hunger step) at the corresponding time point.

The final analysis unit 230 integrates and analyzes the predicted feed amount stage information of the farm by the first analysis unit 210 and the movement stage information of the fish predicted by the second analysis unit 220. In this case, information on the final feed amount to be fed into the farm is extracted. That is, the most optimal feed amount information is extracted.

To this end, the final analysis unit 230 performs integrated analysis in consideration of Equation 1 below.

Here, T is the final feed amount information,

A is information on the feed supply amount stage of the farm predicted by the first analysis unit 210,

B is the movement stage information of the fish predicted by the second analysis unit 220,

W denotes weight setting information corresponding to the analyzed picture quality sharpness.

Regarding the weight setting information, the final analysis unit 230 analyzes the sharpness of the image quality of the image related information by the image input unit 120, and based on the analyzed sharpness of the image quality, the second analyzer 220 The weight for the motion stage information of the fish predicted by the method is controlled.

That is, as much as the movement of fish is captured inside and outside the tank, the clarity of the image varies depending on the water quality and the weather condition outside the tank.

Considering this point, since problems such as image quality can reduce the accuracy of analysis or improve the degree of error occurrence, the degree of sharpness of image quality (range: 0 to 1) is analyzed, and based on this, the second analysis It is preferable to control the degree of reflection of the fish motion stage information predicted by the unit 220 .

Based on the weight setting information and each preset step, the final feed amount information is calculated within 0.5 to 10.

The calculated value is set as the final feed amount information by extracting the matching feed amount information based on Table 1 below.

Feeding level top supply

108 < T

10 ∨ 6 < T

8 ∨ 4 < T

6 ∨ 2 < T

4 lowest supply 0.5

T

2

The AI-based fish farm automatic feeding system according to an embodiment of the present invention divides the final feed feeding amount information into 5 stages based on Table 1, but this is only an embodiment of the present invention, and the size of the farm , the type of fish, etc., and can be freely set.

The supply control unit 300 generates a control signal for controlling the feed supply means in the linked farm, based on the feed amount information from the AI analysis unit 200.

This is generated by matching with Table 1, and the feed supply means supplies feed to the farm.

The AI-based fish farm automatic feeding system according to an embodiment of the present invention does not stop at simply supplying feed, but continuously receives at least one of water quality related information and image related information about the farm after feeding. , additional feed may be provided by determining whether appropriate feed has been provided.

For example, based on the feed amount information by the AI analyzer 200, a control signal for controlling the feed supply means in the farm is generated, and even after feeding is performed, the second As a result of analyzing the video-related information input to the AI model, the hunger level/stage did not correspond to the movement immediately after feeding, and as a result of analyzing the water quality-related information input to the first AI model, the actual feed feeding If the stage matching the quantity information is not output, if the integrated analysis is performed, the feed quantity information above a certain level is extracted. As a result, additional feed is provided.

That is, in the conventional case, when a preset amount of feed is supplied at a preset time, no additional feed is supplied, but in the case of the present invention, even after the feed is supplied, the required feed amount information is continuously analyzed. By doing so, there is an advantage in that it is possible to derive optimal growth by performing feed supply according to the fish's response.

2 is an exemplary sequence diagram illustrating an automatic feeding method for an AI-based farm according to an embodiment of the present invention. Referring to FIG. 2 , the automatic feeding method of the AI-based farm according to an embodiment of the present invention will be described in detail.

As shown in FIG. 2, the automatic feeding method of an AI-based farm according to an embodiment of the present invention includes a data input step (S100), an AI analysis step (S200), a feeding control step (S300), and a feed supply step. It is preferable to include (S400), and each step is operated using an automatic feeding system of an AI-based farm in which each step is performed by an arithmetic processing unit.

For a detailed look at each step,

In the data input step (S100), at least one of water quality related information and image related information about the farm is received from the data input unit 100.

As shown in FIG. 2, the data input step (S100) includes a water quality input step (S110) and an image input step (S120).

In the water quality input step (S110), water quality related information including water temperature information and dissolved oxygen information is received in units of predetermined time by using at least one sensing unit installed inside the water tank of the farm. At this time, in addition to water temperature information and dissolved oxygen information, pH concentration, salinity, ammonia, and nitrite information may be further included.

The image input step (S120) is image-related information including image information obtained by photographing the movement of the fish in units of a preset time from each photographing means using photographing means respectively installed inside and outside the tank of the farm. will be input. That is, information for photographing the movement of fish in the water and information for photographing the movement of fish on the surface of the water are respectively input. Here, the motion photographing information of the fish on the water surface does not mean only the motion photographing information of the fish seen from outside the water, but also captures a motion of moving up to the surface of the water and opening its mouth.

In the AI analysis step (S200), the AI analysis unit 200 uses the stored AI model to analyze the information obtained in the data input step (S100) to extract feed supply amount information of the farm.

At this time, in order to store the AI model for use in the AI analysis step (S200), as shown in FIG. 2, an AI processing step (S10) is further included.

In the AI processing step (S10), in advance, the AI processing unit 10 collects information related to water quality according to the amount of feed supplied in the farm and information related to images of fish movements according to the supply of feed in the farm, and pre-stored Learning processing by the AI network is performed to save the AI model.

As shown in FIG. 2, the AI processing step (S10) includes a first learning step (S11) and a second learning step (S12).

In the first learning step (S11), water quality related information including water temperature information and dissolved oxygen information according to feed supply amount information for each of a plurality of predetermined steps is collected.

In other words, the water quality related information is collected for each step (different feed supply amount) by making the feed supply amount different for each of a plurality of preset stages. In this case, the water quality-related information may further include pH concentration, salinity, ammonia, and nitrite information in addition to water temperature information and dissolved oxygen information.

Here, 10 steps are set as preset steps, but this is only one embodiment of the present invention, and is not necessarily limited to 10 steps, and is freely set in consideration of the size of the farm, the type of fish, etc.

Most preferably, after receiving input from the outside (aquaculture operator, aquaculture expert, etc.) on the maximum feed supply amount and minimum feed supply amount that can be paid, taking into account the size of the farm, the type of fish, the number of fish, etc. It is preferable to divide by the number uniformly. That is, it is preferable to divide the feed supply amount by the set number of steps so as to reach from the minimum feed supply amount to the maximum feed supply amount or from the maximum feed supply amount to the minimum feed supply amount.

That is, after sensing the water quality condition (water temperature information, dissolved oxygen information, pH concentration, salinity, ammonia, nitrite information) according to the feed supply amount set differently for each stage, feed supply amount data-water quality state sensing data is collected.

Of course, it is preferable to process the collected data as learning data for labeling by matching water quality condition information for each feed supply step (feed supply amount step).

Through the learning process of such learning data, an AI model for predicting the currently supplied feed amount is generated using real-time input or water quality state information input at a desired time.

The collection information is input to the pre-stored first AI network to perform learning processing.

Here, the pre-stored first AI network is preferably an inception time algorithm that uses multivariate time series data as an input, but this is also only one embodiment of the present invention, and multivariate Any AI network that can receive time series data and output probability values for each stage can be applied.

That is, in the first learning step (S11), learning is performed by receiving water quality related information for each of a plurality of predetermined steps (for each feed supply step), which is learning data, so that when water quality related information at a specific point in time is input, A first AI model capable of predicting/outputting the feed supply amount (feed supply stage) at the time of feeding is generated. In the first AI model, when water quality-related information at a specific time point is input, a probability value corresponding to each stage is output based on each learned stage, and the feed supply amount at the time corresponding to the stage with the highest probability (feed supply stage) to predict and output.

The second learning step (S12) is an image including image information of the movement of fish in the water and on the water surface at predetermined times for each of a plurality of preset stages in a time series order after feed is supplied in the farm. You will collect relevant information.

In other words, after feeding the fish, the movement of the fish is acquired in the water/surface for each preset time unit.

Here, although 5 steps are set as preset steps, this is only one embodiment of the present invention, and is not necessarily limited to 5 steps, and is freely set in consideration of the size of the farm, the type of fish, etc.

However, in order to analyze the degree of movement from immediately after the fish is fed (full state) to when it feels hungry over time, in other words, at predetermined intervals (time intervals sufficient to divide the degree of hunger) Using fish movement information, data is collected to create an AI model to predict the current hunger state.

Unlike the learning data of the first learning step (S11), which can be labeled by relatively objectively dividing the amount of feed and setting the level, the degree of fish hunger is relatively subjective, so right after feeding is set as step 1, The point at which a predetermined interval has passed is set as step 2, and this is repeated to process learning data by labeling motion state information in a maximum of step 5.

At this time, for a series of processes for acquiring movement state information of fish sequentially at predetermined intervals from immediately after feeding, at least once after supplying feed according to instructions from an outside (farm operator, aquaculture expert, etc.) , It is desirable to collect data by photographing the movement state of fish at predetermined time intervals set by the input.

After that, learning data is generated through a labeling process for this.

Here, the movement state information of the fish includes how many fishes have passed through a specific area set in advance for a predetermined time, or at least how many fishes have moved in the specific area (the number of movements of fins), or a predetermined time period. During this process, how many fish came up to the surface of the water and bitten (bite) is photographed, and the movement itself is captured as a feature to create learning data.

In the second learning step (S12), the collection information is input to the pre-stored second AI network to perform learning processing.

Here, the pre-stored second AI network is preferably a convolution LSTM algorithm, but this is also only an embodiment of the present invention, and any AI network used for classification of time-series behavioral changes can be applied.

In the second learning step (S12), learning is performed by receiving information related to images of motions of fish for each of a plurality of preset steps, which are learning data. A second AI model capable of predicting/outputting the fish's hunger stage (time interval stage after feeding) is created.

More specifically, when image-related information at a specific point in time is input, the second AI model outputs a probability value corresponding to each step based on each learned step, and a time point corresponding to the step with the highest probability. It predicts and outputs the fish's hunger stage (time interval stage after feeding).

The AI analysis step (S200) uses the first AI model and the second AI model generated by the AI processing step (S10) to input the information from the data input step (S100) to supply feed to the farm. information is extracted.

As shown in FIG. 2 , the AI analysis step (S200) includes a first analysis step (S210), a second analysis step (S220) and a final analysis step (S230).

The first analysis step (S210) uses the first AI model to enter the water quality-related information from the water quality input step (S110), and the farm's feed supply amount step information predicted according to the water quality-related information will output

As described above, when water quality-related information at a specific time point is input, the first AI model can predict/output the amount of feed supply (feed supply step) at that point in time, so that the water quality input step (S110) Water quality-related information is input, and a feed supply amount time point (feed supply amount stage information) predicted according to the corresponding water quality-related information is output.

At this time, the output value is output by predicting the step having the highest probability among a plurality of steps as the feed supply amount (feed supply step) at the corresponding time.

In addition, in the first analysis step (S210), it is preferable to perform pre-processing on the water quality-related information by the water quality input step (S110) and then input the information to the first AI model. That is, the pre-processing is performed in consideration of the fact that an error in the sensing unit may occur.

In detail, based on the maximum value (for example, maximum temperature of water, maximum amount of dissolved oxygen, etc.) and minimum value of water quality-related information preset for each fish in the tank, the water quality input step (S110) inputs If the information related to water quality exceeds the maximum value (exceeds the maximum value), it is mapped to the maximum value, and if it is out of the minimum value (below the minimum value), it is mapped to the minimum value.

At this time, if water quality-related information that exceeds the maximum value or minimum value is input continuously for more than a predetermined number of times, it is judged not to be an error of the sensing means, but to an abnormality in the water quality condition in the tank, and an external manager (farmer, etc.) An alarm related to water quality abnormalities is sent to the user.

In addition, instead of setting the maximum and minimum values of water quality-related information for each fish in the tank in advance, the fish in the tank is recognized using a photographic means installed inside the tank at the farm, and the water quality is automatically related to each fish in the tank. It is also possible to set and utilize the maximum and minimum values of information. In this case, there is an advantage in minimizing human errors that may occur in the process of setting fish.

In addition, in this way, when the fish in the tank is automatically recognized or the fish in the tank is input and set in advance, various fish, not a single fish, may exist according to the farm situation.

In this case, the minimum value among preset maximum values for each fish may be set as the representative maximum value, and the maximum value among preset minimum values may be set as the representative minimum value and utilized.

The second analysis step (S220) inputs the image-related information from the image input step (S120) using the second AI model, and outputs fish movement stage information predicted according to the image-related information. do.

In detail, when motion-related information at a specific time point is input, the fish's hunger stage (time interval stage after feeding) at the corresponding time point can be predicted/output, so that the image related by the image input step (S120) can be predicted. Information is input, and the fish's movement stage (time interval stage after feeding/hunger stage) predicted according to the corresponding image-related information is output.

At this time, the output value is output by predicting the step having the highest probability among a plurality of steps as the movement step of the fish (time interval step after feeding/hunger step) at the corresponding time point.

The final analysis step (S230) integrates and analyzes the predicted feed amount stage information of the farm by the first analysis step (S210) and the fish movement stage information predicted by the second analysis step (S220). Thus, the final feed amount to be fed into the farm is extracted. That is, the most optimal feed amount information is extracted.

To this end, in the final analysis step (S230), an integrated analysis is performed in consideration of Equation 1 above.

In addition, the final analysis step (S230) analyzes the image quality sharpness of the image-related information by the image input step (S120), and based on the analyzed image quality sharpness, the fish predicted by the second analysis step (S220). A weight for motion step information is controlled.

That is, as much as the movement of fish is captured inside and outside the tank, the clarity of the image varies depending on the water quality and the weather condition outside the tank.

Considering this point, since problems such as image quality can reduce the accuracy of analysis or improve the degree of error occurrence, the degree of sharpness of image quality (range: 0 to 1) is analyzed, and based on this, the second analysis It is preferable to control the reflectivity of the fish movement stage information predicted by step S220.

Based on the weight setting information and each preset step, the final feed amount information is calculated within 0.5 to 10.

The calculated value is set as the final feed amount information by extracting the matching feed amount information based on Table 1 above.

In the feeding control step (S200), the supply control unit 300 transmits a control signal for controlling the feed supply means in the farm based on the feed amount information by the AI analysis step (S200). will create

In the feed supply step (S400), the feed supply unit performs feed supply in the farm according to a control signal from the feed control step (S300).

Here, the automatic feeding method of an AI-based fish farm according to an embodiment of the present invention does not stop at simply supplying feed, but after supplying feed, continuously at least one of water quality related information and image related information for the farm Upon receiving the input, it is determined whether appropriate feed supply has been made, and additional feed supply may be made.

For example, based on the feed amount information by the AI analysis step (S200), a control signal for controlling the feed supply means in the farm is generated, and even after feeding is performed, the second As a result of analyzing the video-related information input to the AI model, the hunger level/stage did not correspond to the movement immediately after feeding, and as a result of analyzing the water quality-related information input to the first AI model, the actual feed feeding If the stage matching the quantity information is not output, if the integrated analysis is performed, the feed quantity information above a certain level is extracted. As a result, additional feed is provided.

Although the preferred embodiments of the present invention have been described above, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but are only for explanation. Therefore, the technical spirit of the present invention includes not only each disclosed embodiment, but also a combination of the disclosed embodiments, and furthermore, the scope of the technical spirit of the present invention is not limited by these embodiments. In addition, many changes and modifications to the present invention can be made by those skilled in the art without departing from the spirit and scope of the appended claims, and all such appropriate changes and modifications are equivalent. As water, it should be considered within the scope of this invention.

10: AI processing unit
11: first learning data collection unit 12: first learning processing unit
13: second learning data collection unit 14: second learning processing unit
100: data input unit
110: water quality input unit 120: video input unit
200: AI analysis unit
210: first analysis unit 220: second analysis unit
230: final analysis unit
300: supply control unit

Claims (18)

A water quality input unit 110 including a sensing means installed inside the fish farm's water tank and receiving water quality-related information about the farm in a predetermined time unit from the sensing means, and a water quality input unit 110 installed inside and outside the water tank of the fish farm, respectively A data input unit including a photographing means and including an image input unit 120 that receives information related to the movement of fish in the water and photographing the movement of fish on the surface of the water in units of a predetermined time from each photographing means as image-related information. (100);
an AI analysis unit 200 that extracts feed supply amount information of the farm by inputting information by the data input unit 100 using the stored first and second AI models; and
a supply control unit 300 generating a control signal for controlling feed supply means in the farm, based on the feed amount information from the AI analyzer 200;
Including,
The AI analysis unit 200
A first analysis unit 210 that outputs information on the feed supply amount stage of the farm predicted by inputting water quality related information by the water quality input unit 110 to the first AI model, and outputting the image to the second AI model A second analysis unit 220 that receives image-related information through the input unit 120 and outputs predicted fish movement stage information, and the predicted fish farm feed feeding amount stage information and predicted fish movement stage information It is composed of a final analysis unit 230 that comprehensively analyzes and calculates the final feed amount information to be fed into the farm,
The final analysis unit 230
The image quality sharpness of the image-related information is analyzed by the image input unit 120, and a weight corresponding to the analysis result of the image quality sharpness is set to the motion stage information of the fish predicted by the second analyzer 220. Calculate the final feed amount information,
The feed supply means
An automatic feeding system for an AI-based farm that supplies feed to the farm according to a control signal from the supply control unit 300.
According to claim 1,
The AI-based fish farm automatic feeding system
In advance, an AI processing unit 10 that collects information related to water quality according to the amount of feed supply in the farm and information related to images of fish movements according to the supply of feed in the farm, and performs learning processing by a pre-stored AI network;
Including, AI-based automatic feeding system of the fish farm.
According to claim 2,
The AI processing unit 10
A first learning data collection unit 11 that collects water quality related information including water temperature information and dissolved oxygen information according to a plurality of preset feed supply amount information for each stage; and
a first learning processing unit 12 that inputs information by the first learning data collection unit 11 to a pre-stored first AI network and performs learning processing;
Including,
An automatic feeding system for an AI-based fish farm that stores the first AI model based on the learning result of the first learning processing unit 12 in the AI analysis unit 200.
According to claim 3,
The AI processing unit 10
Second learning data that collects image-related information including image information of the movement of fish in the water and on the surface of the water at predetermined times for each of a plurality of stages in a time series after feeding is provided in the farm collection unit 13; and
a second learning processing unit 14 that inputs information by the second learning data collection unit 13 to a pre-stored second AI network and performs learning processing;
Including,
An automatic feeding system for an AI-based fish farm that stores the second AI model based on the learning result of the second learning processing unit 14 in the AI analysis unit 200.
According to claim 4
The water quality input unit 110
An automatic feeding system for an AI-based fish farm that includes water temperature information and dissolved oxygen information as the water quality related information.
delete delete delete delete delete An automatic feeding method for an AI-based fish farm by an automatic feeding system for an AI-based fish farm in which each step is performed by an arithmetic processing means,
A data input step of receiving water quality related information and image related information about the farm (S100);
An AI analysis step (S200) of extracting feed amount information of the farm by analyzing the information obtained in the data input step (S100) using the stored first and second AI models;
A feeding control step (S300) of generating a control signal for controlling a feed supply means in the farm based on the feed amount information from the AI analysis step (S200); and
Feed supply step (S400) of performing feed supply in the farm according to the control signal by the feed control step (S300);
Including,
Before performing the AI analysis step (S200),
In advance, AI that collects information related to water quality according to the amount of feed in the farm and video related to the movement of fish according to the supply of feed in the farm, performs learning processing by the pre-stored AI network, and saves the AI model. processing step (S10);
Including,
The data input step (S100) is
A water quality input step (S110) of receiving water quality related information including water temperature information and dissolved oxygen information in predetermined time units using a sensing means installed inside the water tank of the farm; and
Image-related information including information about the movement of fish in the water and information about the movement of fish on the surface of the water is captured in predetermined time units from each of the photographing means using the photographing means installed inside and outside the tank in the farm. An input image input step (S120);
Including more,
The AI analysis step (S200)
A first analysis step (S210) of inputting water quality-related information from the water quality input step (S110) and outputting information on a feed supply amount stage of the farm predicted by using the first AI model;
a second analysis step (S220) of inputting image-related information from the image input step (S120) and outputting predicted movement stage information of the fish using the second AI model; and
The feed feeding amount stage information predicted by the first analysis step (S210) and the movement stage information of the fish predicted by the second analysis step (S220) are integrated and analyzed, A final analysis step of calculating final feed amount information (S230);
Including more,
The final analysis step (S230)
The image quality sharpness of the image-related information is analyzed by the image input step (S120), and a weight corresponding to the analysis result of the image quality sharpness is set to the motion stage information of the fish predicted by the second analysis step (S220). An automatic feeding method for an AI-based farm that calculates the final feed amount information.
According to claim 11,
The AI processing step (S10)
A first learning step of collecting water quality-related information including water temperature information and dissolved oxygen information according to a plurality of preset feed supply amount information for each step, and performing learning processing by inputting the collected information to a pre-stored first AI network ( S11);
Including,
An automatic feeding method for an AI-based farm that stores the first AI model based on the learning result of the first learning step (S11).
According to claim 12,
The AI processing step (S10)
After the feed is supplied in the farm, image information including image information of the movement of fish in the water and on the surface of the water is collected at predetermined times for each of a plurality of predetermined steps in a time series order, and stored 2 A second learning step (S12) of performing learning processing by inputting the collected information to the AI network;
Including,
An automatic feeding method for an AI-based farm that stores a second AI model based on the learning result of the second learning step (S12).
delete delete delete delete According to claim 11,
The first analysis step (S210)
Regarding the input water quality related information in the water quality input step (S110), based on a preset maximum value or minimum value, water quality related information exceeding the maximum value is mapped to the maximum value, and water quality related information below the minimum value is mapped to the minimum value. An automatic feeding method for an AI-based farm that is mapped and input to the first AI model.

Images