A control method and system for an underwater luminaire based on artificial intelligence

Abstract

The application provides a control method and system of an underwater illuminator based on artificial intelligence, and relates to the technical field of underwater illuminator control. The method comprises the following steps: acquiring an initial monitoring video of the spatial distribution of fish groups in a net cage, a light intensity tomographic image of a breeding water area, and installation information of each fish gathering and growth promoting regulation lamp; determining the optimal output light power of each fish gathering and growth promoting regulation lamp based on the light intensity tomographic image of the breeding water area, the installation information of each fish gathering and growth promoting regulation lamp, fish group behavior response videos corresponding to each survey regulation lamp in the net cage, and fish group behavior response videos corresponding to each supplementary survey regulation lamp in the net cage; and controlling each fish gathering and growth promoting regulation lamp to emit light to the fish groups in the deep-sea net cage based on the optimal output light power of each fish gathering and growth promoting regulation lamp. The method can efficiently and accurately determine the optimal output light power of each fish gathering and growth promoting regulation lamp in the deep-sea net cage and realize multi-lamp linkage regulation.

Description

Technical Field

[0001] This invention relates to the field of underwater lighting control technology, and more specifically to a control method and system for an underwater lighting device based on artificial intelligence. Background Technology

[0002] Deep-sea cage aquaculture is a modern aquaculture model that cultivates fish on a large scale in a controlled deep-sea environment. In deep-sea cage aquaculture, light regulation is a key environmental factor influencing fish schooling behavior and growth promotion. Fish-attracting and growth-promoting lights guide fish to gather spatially and stimulate their feeding and metabolic activities by outputting specific power of light into the cages, which is an important means to improve aquaculture yield and efficiency. However, existing underwater lighting control methods have significant shortcomings. The light field distribution within deep-sea cages is complex, and fish behavior responses are highly variable. Existing methods heavily rely on the experience of aquaculture technicians to judge and dynamically adjust the output light power of each light, which is highly subjective and incurs high trial-and-error costs, making it difficult to guarantee the objectivity and accuracy of control decisions. If unreasonable lighting configurations are not detected and corrected in time, they may have a sustained negative impact on fish school aggregation and growth promotion, and may even trigger acute stress and mortality in the fish. Existing control methods cannot perform refined, multi-stage feedback surveys of fish behavior responses in cages at different locations, thus restricting the level of refined management and overall production efficiency in deep-sea cage aquaculture.

[0003] Therefore, how to efficiently and accurately determine the optimal output light power of each fish-attracting and growth-promoting control lamp in deep-sea cages and achieve multi-lamp linkage control is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem solved by this invention is how to efficiently and accurately determine the optimal output light power of each fish-attracting and growth-promoting control lamp in a deep-sea cage and achieve multi-lamp linkage control.

[0005] According to a first aspect, the present invention provides a control method for an underwater lighting device based on artificial intelligence, comprising: acquiring an initial monitoring video of the spatial distribution of fish in a net cage, a tomographic image of the light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; determining multiple survey control lamps and the initial output light power of each survey control lamp based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic image of the light intensity in the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp; controlling each survey control lamp to emit light according to the corresponding initial output light power, and acquiring a video of the fish behavior response in the net cage corresponding to each survey control lamp; and based on the video of the fish behavior response in the net cage corresponding to each survey control lamp... The process involves determining multiple supplementary survey and control lights and the supplementary output light power of each light; controlling each light to emit light according to its corresponding supplementary output light power, and acquiring video footage of fish behavior responses within the net cages corresponding to each light; determining the optimal output light power of each light based on the tomographic scan image of the aquaculture area, the installation information of each fish-attracting and growth-promoting control light, the video footage of fish behavior responses within the net cages corresponding to each survey and control light, and the video footage of fish behavior responses within the net cages corresponding to each supplementary survey and control light; and controlling each light to emit light to the fish in the deep-sea net cages based on its optimal output light power.

[0006] In one possible implementation, determining the optimal output light power of each fish-gathering and growth-promoting control lamp based on the tomographic scan image of the light intensity of the aquaculture area, the installation information of each fish-gathering and growth-promoting control lamp, the video of fish behavior response in the net cage corresponding to each survey control lamp, and the video of fish behavior response in the net cage corresponding to each supplementary survey control lamp includes: constructing a control lamp map, wherein the control lamp map includes multiple nodes and multiple edges between the multiple nodes, the multiple nodes include multiple survey control lamp nodes and multiple supplementary survey control lamp nodes, and the node characteristics of each survey control lamp node are the video of fish behavior response in the net cage corresponding to each survey control lamp, the installation information of each fish-gathering and growth-promoting control lamp, the node characteristics of each supplementary survey control lamp node are the video of fish behavior response in the net cage corresponding to each supplementary survey control lamp, and the video of fish behavior response in the net cage corresponding to each supplementary survey control lamp. The installation information of the fish-attracting and growth-promoting control lights is provided, and the edges represent the positional relationship information between nodes. Based on the control scheme determination model, the control light spectrum is processed to determine multiple preliminary linkage control schemes for the fish-attracting and growth-promoting control lights in the net cage. Each preliminary linkage control scheme includes a preliminary output optical power for each fish-attracting and growth-promoting control light. Based on the fish behavior response video in the net cage corresponding to each survey control light, the fish behavior response video in the net cage corresponding to each supplementary survey control light, the tomographic scan image of the light intensity in the aquaculture water area, and the multiple preliminary linkage control schemes for the fish-attracting and growth-promoting control lights in the net cage, a simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme is generated. Based on the simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme, the optimal output optical power of each fish-attracting and growth-promoting control light is determined.

[0007] In one possible implementation, determining multiple supplementary survey and control lights and the supplementary output optical power of each supplementary survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light includes: determining multiple response fish movement state information corresponding to each survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light; clustering the multiple response fish movement state information corresponding to each survey and control light to obtain multiple behavioral feature clusters corresponding to each survey and control light; determining multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light based on the multiple behavioral feature clusters corresponding to each survey and control light; and determining multiple supplementary survey and control lights and the supplementary output optical power of each supplementary survey and control light based on the installation information of each fish gathering and growth promotion control light and the multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light.

[0008] In one possible implementation, the control scheme is determined by a graph neural network model.

[0009] According to a second aspect, the present invention provides a control system for an underwater lighting device based on artificial intelligence, comprising: an acquisition module for acquiring initial monitoring video of the spatial distribution of fish in a net cage, a tomographic image of the light intensity of the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; an initial parameter determination module for determining multiple survey control lamps and the initial output light power of each survey control lamp based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic image of the light intensity of the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp; a survey illumination control module for controlling each survey control lamp to emit light according to the corresponding initial output light power, and acquiring a video of the fish behavior response in the net cage corresponding to each survey control lamp; and a supplementary parameter determination module for determining the fish behavior in the net cage corresponding to each survey control lamp. The system responds to video signals to determine multiple supplementary survey and control lights and the supplementary output light power of each light. A supplementary lighting control module controls each light source to emit light according to its corresponding supplementary output light power and acquires video signals of fish behavior within the net cage corresponding to each light source. An optimal parameter determination module determines the optimal output light power of each fish-attracting and growth-promoting light based on the tomographic scan image of the aquaculture area, the installation information of each fish-attracting and growth-promoting light, the video signals of fish behavior within the net cage corresponding to each survey and control light, and the video signals of fish behavior within the net cage corresponding to each supplementary survey and control light. An optimal lighting control module controls each fish-attracting and growth-promoting light to emit light to the fish within the deep-sea net cage based on its optimal output light power.

[0010] In one possible implementation, the optimal parameter determination module is further configured to: construct a control light map, the control light map including multiple nodes and multiple edges between the multiple nodes, the multiple nodes including multiple survey control light nodes and multiple supplementary survey control light nodes, the node characteristics of each survey control light node being a video of fish behavior response in a net cage corresponding to each survey control light and the installation information of each fish gathering and growth promoting control light, the node characteristics of each supplementary survey control light node being a video of fish behavior response in a net cage corresponding to each supplementary survey control light and the installation information of each fish gathering and growth promoting control light, and the edges being the positional relationship information between the nodes; and perform control light map analysis based on the control scheme determination model. The process involves determining multiple preliminary linkage control schemes for the fish-attracting and growth-promoting lights within the net cages. Each preliminary linkage control scheme includes a preliminary output optical power for each fish-attracting and growth-promoting light. Based on the fish behavior response videos within the net cages corresponding to each survey control light, the fish behavior response videos within the net cages corresponding to each supplementary survey control light, the tomographic scan image of the light intensity in the aquaculture area, and the multiple preliminary linkage control schemes for the fish-attracting and growth-promoting lights within the net cages, simulation videos of the fish behavior response within the net cages under each preliminary linkage control scheme are generated. Based on the simulation videos of the fish behavior response within the net cages under each preliminary linkage control scheme, the optimal output optical power for each fish-attracting and growth-promoting light is determined.

[0011] In one possible implementation, the supplementary parameter determination module is further configured to: determine multiple response fish movement state information corresponding to each survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light; cluster the multiple response fish movement state information corresponding to each survey and control light to obtain multiple behavioral feature clusters corresponding to each survey and control light; determine multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light based on the multiple behavioral feature clusters corresponding to each survey and control light; and determine multiple supplementary survey and control lights and the supplementary output optical power of each supplementary survey and control light based on the installation information of each fish gathering and growth promotion control light, the multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light.

[0012] In one possible implementation, the control scheme is determined by a graph neural network model.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method described above, the method including: acquiring initial monitoring video of the spatial distribution of fish in a net cage, tomographic images of light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; determining multiple survey control lamps and the initial output light power of each survey control lamp based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic images of light intensity in the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp; controlling each survey control lamp to emit light according to the corresponding initial output light power, and acquiring the fish behavior response in the net cage corresponding to each survey control lamp. Based on the video of fish behavior response in the net cage corresponding to each survey and control light, determine multiple supplementary survey and control lights and the supplementary output light power of each supplementary survey and control light; control each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and acquire the video of fish behavior response in the net cage corresponding to each supplementary survey and control light; based on the tomographic scan image of the light intensity of the aquaculture water area, the installation information of each fish gathering and growth promotion control light, the video of fish behavior response in the net cage corresponding to each survey and control light, and the video of fish behavior response in the net cage corresponding to each supplementary survey and control light, determine the optimal output light power of each fish gathering and growth promotion control light; control each fish gathering and growth promotion control light to emit light to the fish in the deep-sea net cage based on the optimal output light power of each fish gathering and growth promotion control light.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned control method for an artificial intelligence-based underwater lighting device. The method includes: acquiring initial monitoring video of the spatial distribution of fish in a net cage, tomographic images of light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; determining multiple survey control lamps and the initial output light power of each survey control lamp based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic images of light intensity in the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp; controlling each survey control lamp to emit light according to the corresponding initial output light power, and acquiring a video of the fish behavior response in the net cage corresponding to each survey control lamp; based on the... The system describes the behavior response video of fish in the net cage corresponding to each survey and control light, determines multiple supplementary survey and control lights, and the supplementary output light power of each supplementary survey and control light; controls each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and acquires the behavior response video of fish in the net cage corresponding to each supplementary survey and control light; based on the light intensity tomographic scan image of the aquaculture water area, the installation information of each fish gathering and growth promotion control light, the behavior response video of fish in the net cage corresponding to each survey and control light, and the behavior response video of fish in the net cage corresponding to each supplementary survey and control light, determines the optimal output light power of each fish gathering and growth promotion control light; based on the optimal output light power of each fish gathering and growth promotion control light, controls each fish gathering and growth promotion control light to emit light to the fish in the deep-sea net cage.

[0015] This invention provides a control method and system for an underwater lighting device based on artificial intelligence. The method includes acquiring initial monitoring video of the spatial distribution of fish in a net cage, tomographic images of light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic images of light intensity in the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp, determining multiple survey control lamps and the initial output light power of each survey control lamp; controlling each survey control lamp to emit light according to its corresponding initial output light power, and acquiring a video of the fish behavior response in the net cage corresponding to each survey control lamp; based on the video of the fish behavior response in the net cage corresponding to each survey control lamp, determining multiple supplementary survey control lamps and the initial output light power of each supplementary survey control lamp. The method involves adjusting the supplementary output light power of the lights; controlling each supplementary survey and control light to emit light according to its corresponding supplementary output light power, and acquiring video of fish behavior response in the net cage corresponding to each supplementary survey and control light; determining the optimal output light power of each fish gathering and growth promoting control light based on the tomographic scan image of the aquaculture water area, the installation information of each fish gathering and growth promoting control light, the video of fish behavior response in the net cage corresponding to each survey and control light, and the video of fish behavior response in the net cage corresponding to each supplementary survey and control light; and controlling each fish gathering and growth promoting control light to emit light to the fish in the deep-sea net cage based on the optimal output light power of each fish gathering and growth promoting control light. This method can efficiently and accurately determine the optimal output light power of each fish gathering and growth promoting control light in the deep-sea net cage and realize multi-light linkage control. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a control method for an underwater lighting device based on artificial intelligence, provided in an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a deep-sea cage provided in an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of a fish-attracting and growth-promoting control lamp provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of a process for determining multiple adjustment survey control lamps and the supplementary output optical power of each adjustment survey control lamp, provided as an embodiment of the present invention.

[0020] Figure 5 A schematic diagram of a process for determining the optimal output light power of each fish-attracting and growth-promoting control lamp, provided in an embodiment of the present invention;

[0021] Figure 6 This is a schematic diagram of a control system for an artificial intelligence-based underwater lighting device provided as an embodiment of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0023] In this embodiment of the invention, the following are provided: Figure 1 The above describes a control method for an artificial intelligence-based underwater illuminator, comprising steps S1 to S7:

[0024] Step S1: Obtain initial monitoring video of the spatial distribution of fish in the net cage, tomographic scan image of light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp.

[0025] Deep-sea cages refer to large-scale, enclosed, three-dimensional aquaculture facilities placed in deep-sea areas for large-scale aquaculture. Figure 2 This is a schematic diagram of a deep-sea cage provided in an embodiment of the present invention.

[0026] Fish attraction and growth-promoting control lights are underwater lighting devices deployed in deep-sea cages that use adjustable power light to induce fish to gather at the light source and regulate their growth physiological rhythms, thereby controlling the spatial distribution of fish schools and increasing fish production. Figure 3 This is a schematic diagram of a fish-attracting and growth-promoting control lamp provided in an embodiment of the present invention.

[0027] The initial monitoring video of the spatial distribution of fish within the net cage is obtained by continuously filming the fish within the cage using video acquisition equipment installed inside the deep-sea net cage. This initial monitoring video of the spatial distribution of fish within the net cage can show the temporal changes in the fish's location, aggregation density, and movement within the three-dimensional space of the deep-sea net cage.

[0028] The tomographic images of light intensity in aquaculture waters are obtained by performing tomographic scanning of the aquaculture waters using light intensity detection equipment deployed at different water depths. Each location in the tomographic images of light intensity in aquaculture waters is labeled with the corresponding light intensity value for that cross-section.

[0029] Tomographic images of light intensity in aquaculture waters can reflect the spatial distribution of light intensity at different water depths.

[0030] The installation information for each fish attraction and growth promotion control light describes its actual installation status within the deep-sea cage. This information includes the installation coordinates, installation depth, illumination direction angle, rated output light power, and illumination coverage radius of each light.

[0031] Step S2: Based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic scan image of the light intensity in the aquaculture area, and the installation information of each fish gathering and growth-promoting control lamp, determine the multiple survey control lamps and the initial output light power of each survey control lamp.

[0032] In some embodiments, a survey and control lamp determination model can be used to determine multiple survey and control lamps and the initial output optical power of each lamp. The survey and control lamp determination model is a gated loop unit. The inputs to the survey and control lamp determination model are the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic image of the light intensity in the aquaculture area, and the installation information of each fish-attracting and growth-promoting control lamp. The outputs of the survey and control lamp determination model are the multiple survey and control lamps and the initial output optical power of each lamp.

[0033] A Gated Recurrent Unit (GRU) is a recurrent neural network architecture for processing time-series data. GRUs control the flow of information between time steps through update and reset gates. The update gate determines how much of the previous state is retained in the current time step, while the reset gate determines how new input information is combined with previous memories. GRUs effectively alleviate the vanishing gradient problem present in traditional recurrent neural networks and can reduce computational complexity while preserving important memories of long sequences.

[0034] Multiple survey and control lights are specific fish-attracting and growth-promoting control lights that are preferentially used in the net cage to emit probe light to obtain the initial response of the fish school, as determined by the survey and control light determination model.

[0035] The initial output optical power of each survey control lamp is a power parameter used to control the illumination detection of each survey control lamp.

[0036] The initial monitoring video of the spatial distribution of fish in the net cages objectively revealed the key areas where the fish densely gathered and the main direction of their swimming. The tomographic image of the light intensity in the aquaculture area visually reflected the degree of attenuation of the background light. The installation information of each fish-attracting and growth-promoting lamp defined the specific location range of equipment that could be used to emit light stimuli. By analyzing the temporal changes in the location distribution, aggregation density, and movement status of the fish in the initial monitoring video of the spatial distribution of fish in the net cages, the model can accurately identify the sensitive areas where the fish may respond to external stimuli. Furthermore, by using the light attenuation presented in the tomographic image of the light intensity in the aquaculture area, the model measures the minimum additional light intensity required to trigger a fish response, and then uses the installation information of each fish-attracting and growth-promoting lamp to select the optimal light-emitting equipment in the spatial location.

[0037] The gated loop unit possesses a powerful ability to process time-series evolution data. The update gate within the model filters and retains the temporal segment features of continuously increasing fish density in the initial monitoring video of the fish spatial distribution within the cage, and propagates these features forward along the time-series link. Next, the reset gate removes irrelevant interference information such as the occasional scattered swimming of fish in the initial monitoring video, thereby resolving the coordinates of the core aggregation area where the fish stably exist within the cage. After determining the coordinates of the core aggregation area, the processing mechanism incorporates the ambient light attenuation values ​​from the corresponding depth section of the tomographic image of the aquaculture water area to perform spatial light field comparison of the core aggregation area, and calculates the background light barrier threshold required to induce phototaxis in the core aggregation area based on the ambient light attenuation values. Subsequently, the model can calculate the straight-line distance between each luminous device within the cage and the core aggregation area, following the topological distribution of the location coordinates contained in the installation information of each fish-attracting and growth-promoting control lamp. The model uses a nonlinear activation function to perform cross-fitting between the linear distance of the luminescent devices and the background illumination barrier threshold. The computational unit can accurately identify specific luminescent devices that are spatially closest to the core aggregation area and whose illumination coverage radius can effectively encompass the fish school. These specific luminescent devices are then output as multiple survey and control lights. For each of these multiple survey and control lights, the model can further extract the rated output light power from the installation information of each fish-attracting and growth-promoting control light as a reference upper limit. Based on the spatial span of the core aggregation area from the multiple survey and control lights and the current activity level of the fish school as shown in the initial monitoring video of the fish school's spatial distribution within the cage, the gated loop unit can dynamically calculate the minimum stimulus energy required to induce the initial phototactic response of the fish school. The model can continuously calibrate the stimulus energy value to ensure that the energy value can penetrate the water layer to attract the attention of the fish school without causing them to scatter due to excessive light intensity. Finally, the model can accurately lock the calibrated stimulus energy value and output it as the initial output light power of each survey and control light.

[0038] In some embodiments, determining multiple survey and control lights and the initial output optical power of each survey and control light light based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic scan image of the light intensity of the aquaculture water area, and the installation information of each fish gathering and growth promoting control light includes steps S21 to S23:

[0039] Step S21: Based on the initial monitoring video of the spatial distribution of fish in the net cage and the tomographic image of the light intensity in the aquaculture water area, determine multiple high-density fish areas, multiple medium- and low-density fish areas in the deep-sea net cage, and the phototactic stress sensitivity of each area.

[0040] In some embodiments, a gated loop unit can be used to determine multiple high-density fish population areas, multiple medium- and low-density fish population areas within a deep-sea cage, and the phototactic stress sensitivity of each population area.

[0041] Multiple high-density fish schools within deep-sea cages represent three-dimensional water areas where fish are spatially concentrated within the cages.

[0042] Multiple low- to medium-density fish populations within deep-sea cages represent three-dimensional water areas where fish are spatially dispersed within the cages.

[0043] The phototactic sensitivity of each area is a numerical indicator that quantifies the swimming activity of fish in each corresponding high-density or medium-low-density fish population area after being stimulated by external light.

[0044] The gated loop unit can analyze the continuous temporal fish location data carried by the initial monitoring video of the fish spatial distribution in the net cage frame by frame, and extract the real-time fish distribution points under the three-dimensional coordinates of different water layers frame by frame. By comparing the differences in the density of fish points in the same water space within consecutive video frames, the model can distinguish between multiple high-density fish areas in deep-sea net cages where the fish are spatially concentrated and multiple medium- and low-density fish areas in deep-sea net cages where the fish are spatially dispersed. The gated loop unit can simultaneously read the fixed baseline illumination values ​​of each water layer stored in the tomographic image of the illumination intensity of the aquaculture water area, and use the amplitude of fish swimming fluctuations in different areas in the video sequence as the core calculation basis. Then, it combines the influence of the native illumination conditions of the corresponding water area on the fish activity state for quantitative conversion, thereby calculating the phototactic stress sensitivity of each area that can characterize the intensity of the fish's phototactic response.

[0045] Step S22: Based on multiple high-density fish population areas within the deep-sea cage, multiple medium- and low-density fish population areas within the deep-sea cage, the phototactic stress sensitivity of each area, the tomographic image of the light intensity in the aquaculture water area, and the installation information of each fish-attracting and growth-promoting control lamp, determine multiple suitable candidate fish-attracting and growth-promoting control lamps for each area and the estimated light requirement value for each area.

[0046] In some embodiments, a deep neural network can be used to determine multiple suitable candidate fish growth control lights for each area and the estimated light demand value for each area.

[0047] A deep neural network is a multilayer perceptron model consisting of an input layer, multiple hidden layers, and an output layer. Deep neural networks utilize non-linear activation functions and fully connected weights between multiple neurons to abstract and extract information from high-dimensional input features layer by layer. Deep neural networks possess powerful data fitting capabilities, enabling them to uncover deep mapping relationships behind complex data.

[0048] Each area has multiple matching candidate fish-attracting and growth-promoting control lights, whose illumination range can fully cover the corresponding high-density fish population area or medium-low density fish population area.

[0049] The estimated light requirements for each area are the basic light power values ​​required to maintain a stable fish population in the corresponding area.

[0050] Deep neural networks can establish a spatial mapping relationship between the illumination space of lighting fixtures and the activity areas of fish schools. They can overlap and verify the three-dimensional illumination boundary of a single fish-attracting and growth-promoting light with the water boundaries of each high-density fish school area and medium-to-low-density fish school area. This identifies lights whose illumination range can completely encompass the corresponding fish activity area. These lights are then grouped and organized to determine multiple suitable candidate fish-attracting and growth-promoting lights for each area. Deep neural networks rely on multi-layered nonlinear computing units to perform illumination numerical extrapolation. Using the native illumination of the water layer where the area is located as the calculation benchmark, the computational load is adjusted layer by layer by incorporating the fish school's light response characteristics reflected by the area's phototactic stress sensitivity. This avoids the problems of excessive light intensity causing fish to scatter and escape, and insufficient light intensity preventing fish from gathering. Deep neural networks can accurately calculate the estimated illumination requirements for each area that can continuously and stably maintain the fish school's aggregation state.

[0051] Step S23: Based on the multiple matching candidate fish-gathering and growth-promoting control lights for each area and the estimated illumination requirements for each area, determine the multiple survey control lights and the initial output light power of each survey control light.

[0052] In some embodiments, a deep neural network can be used to determine a plurality of survey control lights and the initial output optical power of each survey control light.

[0053] Deep neural networks can perform a full-domain traversal screening of suitable candidate fish-gathering and growth-promoting lights for all areas, comparing the types of fish areas covered by each candidate light and their deployment locations. Lights with highly overlapping coverage areas or covering only a single type of fish area are eliminated layer by layer according to the standard of balanced coverage of fish areas across the entire net cage. The model can retain lights that can accommodate high-density, medium-, and low-density fish activity areas, integrating them into multiple survey and control lights. The deep neural network can pinpoint the corresponding illumination requirements for each survey and control light in the fish area, then perform numerical correction calculations based on the light's adjustable power range, and offset numerical deviations caused by native water illumination and the phototactic stress of the fish. This allows for the allocation of suitable power parameters to each survey and control light, ultimately outputting a set of survey lights suitable for the first round of underwater illumination testing, along with the initial output light power of each matched survey and control light.

[0054] Step S3: Control each survey and control light to emit light according to the corresponding initial output light power, and acquire the fish behavior response video in the net cage corresponding to each survey and control light.

[0055] The video of fish behavior response in the net cage corresponding to each survey and control light is captured in real time by video acquisition equipment deployed in the deep-sea net cage during the independent test period corresponding to each survey and control light, when the survey and control light emits light according to the corresponding initial output light power.

[0056] The video of fish behavior response in the net cage corresponding to each survey and control light is used to independently present the phototaxis and aggregation process of the fish, the changes in the swimming direction of the fish, and the temporal changes in the density of the fish during the period when the survey and control light is the only one turned on in the entire net cage.

[0057] Step S4: Based on the video of fish behavior response in the net cage corresponding to each survey and control light, determine the multiple supplementary survey and control lights and the supplementary output light power of each supplementary survey and control light.

[0058] In some embodiments, Figure 4 This is a schematic flowchart illustrating the process of determining multiple adjustment survey control lamps and the supplementary output optical power of each lamp, as provided in an embodiment of the present invention. The determination of the multiple adjustment survey control lamps and the supplementary output optical power of each lamp includes steps S41-S44:

[0059] Step S41: Based on the fish behavior response video in the net cage corresponding to each survey and control light, determine the multiple response fish movement status information corresponding to each survey and control light.

[0060] In some embodiments, a state information determination model can be used to determine the multiple responsive fish movement state information corresponding to each survey and control light. The state information determination model is a gated loop unit. The input to the state information determination model is the video of the fish behavior response within the net cage corresponding to each survey and control light, and the output of the state information determination model is the multiple responsive fish movement state information corresponding to each survey and control light.

[0061] The movement status information of multiple responding fish groups corresponding to each survey and control light is determined by the state information determination model. During the illumination emission of light by each survey and control light according to the corresponding initial output light power, the information on the specific movement performance of multiple independent and separate fish groups under light stimulation is obtained.

[0062] Each response fish swarm movement status information includes the centroid position coordinate curve of each independent individual fish in the swarm, the overall swimming direction angle sequence, the temporal spatial distribution matrix sequence of the swarm aggregation density, and the numerical sequence of the fish swarm movement speed.

[0063] The centroid position coordinate curve of each independent fish group refers to the continuous geometric curve of the trajectory of the overall spatial center position of each independent fish group over time.

[0064] The overall swimming direction angle sequence refers to the data sequence in which the overall swimming direction angle value of each independent individual fish in a school dynamically evolves over time.

[0065] The temporal spatial distribution matrix sequence of population aggregation density refers to the data matrix sequence in which the aggregation density of each independent fish population at different spatial locations continuously evolves over time.

[0066] The temporal spatial distribution matrix sequence of population aggregation density consists of a set of three-dimensional spatial grid density matrices arranged in chronological order. Each time node corresponds to a three-dimensional spatial grid density matrix, and each element in the matrix corresponds to the population density value of the individual fish in the independent fish population within a three-dimensional coordinate grid region of the net cage at the current time node.

[0067] The temporal spatial distribution matrix sequence of population aggregation density can accurately and quantitatively describe the dynamic non-uniform evolution of the aggregation density of each independent fish population at different spatial locations within the cage over time during continuous light stimulation.

[0068] The numerical sequence of fish movement speed refers to the data sequence of the dynamic change of the overall movement speed of each independent individual fish in the school over time.

[0069] The video recording of fish behavior response within the cage corresponding to each survey and control light continuously captures the complete visual process of fish behavior changes during illumination by the corresponding survey and control light source at its initial output power. The video visually presents the evolutionary trajectory of the fish school from its initial discrete distribution to gradually converging towards the center of the light source, while also showcasing dynamic images of swimming direction reversals and localized increases in density. The gated loop unit can accurately extract the changes in physical motion parameters of the fish school before and after stimulation using the video recording of fish behavior response within the cage corresponding to each survey and control light.

[0070] The gated loop unit advances frame-by-frame along the timeline of the fish behavior response video within the net cage corresponding to each survey and control light. Each frame carries the spatial distribution of the fish within the net cage at the current moment. During forward processing, the update gate continuously tracks the evolution trend of the overall swimming direction angle of the fish, allowing the cumulative process of the fish gradually shifting from their initial scattered state towards the light position to be fully modeled in the temporal dimension. The reset gate can intervene when the fish rapidly disperse or suddenly group, clearing the memory of earlier, no longer representative aggregation states, allowing the model to focus on the actual movement pattern of the fish at the current moment. When processing each frame, the gated loop unit can simultaneously extract the fish density values ​​of each coordinate region in the 3D mesh, arrange them in chronological order into a temporal spatial distribution matrix sequence of group aggregation density, and continuously update the overall centroid position of the fish to output the centroid position coordinate curve. As the hidden state accumulates between time steps, the gated loop unit can also extract the rhythm of the overall movement amplitude of the fish in each frame over time, and then summarize the numerical sequence of fish movement speed. By performing the above process on multiple independent fish individuals in the video time series, the model finally identifies and outputs multiple response fish movement state information corresponding to each survey and control light, including the centroid position coordinate curve of each independent fish individual, the overall swimming direction angle sequence, the temporal spatial distribution matrix sequence of the group aggregation density, and the numerical sequence of fish movement speed.

[0071] Step S42: Cluster the multiple response fish movement state information corresponding to each survey and control light to obtain multiple behavioral feature clusters corresponding to each survey and control light.

[0072] In some embodiments, multiple behavioral feature clusters corresponding to each survey and control light can be obtained by using the K-means clustering algorithm based on the multiple response fish movement state information corresponding to each survey and control light.

[0073] K-means clustering is an unsupervised clustering algorithm that divides a dataset into a predetermined number of K clusters based on the distance between samples. K-means clustering iteratively updates the cluster centers to maximize the similarity of samples within the same cluster and minimize the similarity between samples in different clusters, ultimately assigning each data point to the cluster with the nearest cluster center. The value of K can be preset manually.

[0074] The multiple behavioral feature clusters corresponding to each survey and control light are formed by dividing the multiple response fish movement state information corresponding to each survey and control light during the initial output light power illumination emission period into multiple data groups based on the similarity of movement features.

[0075] Each survey control light independently corresponds to a set of multiple behavioral feature clusters during the illumination period of that survey control light. The number of multiple behavioral feature clusters corresponding to each survey control light is the K value in the K-means clustering algorithm.

[0076] Each behavioral feature cluster corresponding to each survey and control light contains multiple responses from fish with similar movement characteristics. There are significant differences in the responses from fish within different behavioral feature clusters.

[0077] As an example, specifically: The K-means clustering algorithm pre-determines K cluster centers within the dataset consisting of multiple response fish movement status information corresponding to each survey and control light. It transforms the centroid coordinate curve, overall swimming direction angle sequence, group aggregation density temporal spatial distribution matrix sequence, and fish movement speed numerical sequence of each response fish movement status information into high-dimensional feature points. Then, it calculates the distance between each high-dimensional feature point and each cluster center, assigning each response fish movement status information to the cluster containing the nearest cluster center. After one round of partitioning, the K-means clustering algorithm recalculates the feature average of all response fish movement status information within each cluster to update the cluster centers, and then re-partitions according to the updated cluster centers. The K-means clustering algorithm repeats the above iterative process until the cluster centers no longer move significantly and the clustering results tend to stabilize. Ultimately, multiple behavioral feature clusters are formed corresponding to each survey and control light. The response fish movement status information within each behavioral feature cluster is highly similar in movement characteristics, while the movement characteristics differ significantly between different behavioral feature clusters.

[0078] Clustering fish movement status information into multiple behavioral feature clusters allows for the grouping of fish movement status data with similar swimming patterns and light response behaviors into a single category, thus achieving structured grouping of dispersed movement data. This grouping method can clearly distinguish fish activity units with different phototactic behaviors and avoid indiscriminate analysis of all movement status information. Subsequently, based on the overall characteristics of the clusters, it can accurately delineate areas of excessive or sluggish behavior, improving the efficiency and accuracy of identifying abnormal activity areas in underwater fish schools.

[0079] Step S43: Based on the multiple behavioral feature clusters corresponding to each survey and control light, determine multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light.

[0080] In some embodiments, a behavior region determination model can be used to determine multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey control light. The behavior region determination model is a deep neural network. The input to the behavior region determination model is multiple behavior feature clusters corresponding to each survey control light, and the output of the behavior region determination model is multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey control light.

[0081] Each survey and control light corresponds to multiple overactive zones, which refer to multiple spatial geometric areas in which the fish in the net cage exhibit highly active swimming behavior during the illumination period when the corresponding survey and control light emits light according to the corresponding initial output light power.

[0082] The multiple behavioral sluggish zones corresponding to each survey and control light refer to multiple spatial geometric areas in which the fish in the net cage exhibit a low-activity swimming state during the period when the corresponding survey and control light emits light according to its corresponding initial output light power.

[0083] Each survey and control light corresponds to multiple behavioral feature clusters, which group and classify the movement patterns of fish during light stimulation according to their similarity. The common movement features represented by each behavioral feature cluster can reflect the concentration trend of the activity level of individuals of this type of fish under specific light conditions, enabling the model to identify the spatial distribution areas of high and low activity based on the changes in movement speed and aggregation density.

[0084] The deep neural network performs layer-by-layer nonlinear feature extraction on the response fish movement state information contained in multiple behavioral feature clusters corresponding to each survey and control light through multiple hidden layers. Shallow neurons extract the spatial clustering range features of the response fish movement state information within each behavioral feature cluster on the centroid position coordinate curve. The intermediate layers further integrate the amplitude level of the fish movement speed numerical sequence with the density peak information of the group aggregation density temporal spatial distribution matrix sequence, thus forming a distinguishable distribution pattern between high-activity and low-activity behavioral clusters in the feature space. Subsequent layers of the deep neural network can back-project the three-dimensional spatial grid coordinates corresponding to the aggregation density peak of each behavioral feature cluster to locate continuous geometric regions where density maxima continuously appear. The spatial range where the fish movement speed numerical sequence remains high for a long time under continuous light stimulation and density extremes frequently occur in the group aggregation density temporal spatial distribution matrix sequence is identified as the overactive behavior region. Conversely, the spatial range where the fish movement speed numerical sequence remains low for a long time and density values ​​are continuously sparse in the group aggregation density temporal spatial distribution matrix sequence is identified as the sluggish behavior region. The output layer defines the region boundaries based on the density of the activity feature distribution in each continuous geometric region, and finally outputs the spatial geometric range of multiple overactive and multiple sluggish behavior regions for each survey and control lamp.

[0085] Step S44: Based on the installation information of each fish-attracting and growth-promoting control light, the multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey control light, determine multiple supplementary survey control lights and the supplementary output light power of each supplementary survey control light.

[0086] In some embodiments, a supplementary adjustment determination model can be used to determine multiple supplementary adjustment survey and control lights and the supplementary output optical power of each supplementary adjustment survey and control light. The supplementary adjustment determination model is a deep neural network. The inputs to the supplementary adjustment determination model are the installation information of each fish-attracting and growth-promoting control light, multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light, and the outputs of the supplementary adjustment determination model are the multiple supplementary adjustment survey and control lights and the supplementary output optical power of each supplementary adjustment survey and control light.

[0087] Multiple supplementary survey and control lights are determined by the supplementary adjustment model and are additional fish-attracting and growth-promoting control lights that need to be activated for further supplementary illumination detection.

[0088] The supplementary output optical power of each supplementary survey and control lamp is a power parameter used to control the illumination detection of each supplementary survey and control lamp.

[0089] Each survey and control light corresponds to multiple overactive zones, indicating which spatial areas within the net cage have been excessively stimulated by the initial output light power of that light source. The distribution of these overactive zones reflects locations where the current lighting configuration results in excessive illuminance. Multiple sluggish zones indicate areas where fish show almost no phototactic aggregation response, suggesting insufficient light stimulation in these areas and requiring additional detection lighting. The installation information for each fish-attracting and growth-promoting control light clarifies the spatial layout and coverage capacity of each light within the net cage, enabling the model to determine which fish-attracting and growth-promoting control lights, not yet selected as survey and control lights, are located near the sluggish zones.

[0090] A deep neural network can input the spatial geometric range of multiple overactive and sluggish behavior zones corresponding to each survey and control light into the hidden layer. It extracts the three-dimensional spatial coordinate boundaries of each overactive and sluggish zone and performs spatial arrangement analysis of all overactive and sluggish zones in a high-dimensional feature space, identifying the distance gap between sluggish zones and the coverage area of ​​existing survey and control lights. The hidden layer further projects the installation location coordinates, installation depth, illumination direction angle, and illumination coverage radius of each fish-attracting and growth-promoting control light into the same spatial coordinate system, calculating the degree of spatial overlap between the illumination coverage area of ​​each unselected fish-attracting and growth-promoting control light and the sluggish behavior zone. The model marks fish-attracting and growth-promoting control lights with high overlap as supplementary control candidates. Combining the rated output light power of each fish-attracting and growth-promoting control light with the lower limit of the response threshold reflected by the numerical sequence of fish movement speed in the corresponding sluggish behavior zone, it calculates the minimum light power required to evoke an observable phototactic response in the fish when the supplementary survey and control light illuminates the sluggish behavior zone, using this as the supplementary output light power. For survey and control lights covering a large area of ​​overly aggressive behavior, the model introduces additional suppression penalties when determining whether supplementary lights still need to be deployed near the overly aggressive area, avoiding redundant illumination in already overly aggressive areas. The output layer combines the spatial overlap score and power threshold calculation results to finally determine and output multiple supplementary survey and control lights and the supplementary output optical power of each supplementary survey and control light.

[0091] Step S5: Control each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and acquire the fish behavior response video in the net cage corresponding to each supplementary survey and control light.

[0092] The video of fish behavior response in the net cage corresponding to each supplementary survey and control light is a single independent video captured in real time by video acquisition equipment deployed in the deep-sea net cage during the independent test period corresponding to each supplementary survey and control light, when the supplementary survey and control light emits light according to its corresponding supplementary output light power.

[0093] The video of fish behavior response in the net cage corresponding to each supplementary survey and control light is used to independently present the phototaxis and aggregation process of the fish, the changes in the swimming direction of the fish, and the temporal changes in the density of the fish during the period when the supplementary survey and control light is the only one turned on in the entire net cage.

[0094] Step S6: Based on the tomographic scan image of the light intensity of the aquaculture water area, the installation information of each fish gathering and growth promoting control lamp, the video of the fish behavior response in the net cage corresponding to each survey control lamp, and the video of the fish behavior response in the net cage corresponding to each supplementary survey control lamp, determine the optimal output light power of each fish gathering and growth promoting control lamp.

[0095] In some embodiments, Figure 5 This is a schematic flowchart illustrating the process of determining the optimal output light power of each fish-attracting and growth-promoting control lamp according to an embodiment of the present invention. The determination of the optimal output light power of each fish-attracting and growth-promoting control lamp includes steps S61 to S64:

[0096] Step S61: Construct a control light map. The control light map includes multiple nodes and multiple edges between the nodes. The multiple nodes include multiple survey control light nodes and multiple supplementary survey control light nodes. The node characteristics of each survey control light node are the fish behavior response video in the net cage corresponding to each survey control light and the installation information of each fish gathering and growth promoting control light. The node characteristics of each supplementary survey control light node are the fish behavior response video in the net cage corresponding to each supplementary survey control light and the installation information of each fish gathering and growth promoting control light. The edges represent the positional relationship information between the nodes.

[0097] The control light map is a map data composed of multiple survey control light nodes, multiple supplementary survey control light nodes, and multiple edges between nodes. The control light map can integrate the fish behavior response data and spatial location correlation of each survey control light and each supplementary survey control light.

[0098] Each survey and control light node corresponds to a survey and control light. The node characteristics of each survey and control light node include the fish behavior response video in the net cage corresponding to the survey and control light and the global installation information of the fish gathering and growth promotion control lights.

[0099] Each supplementary survey and control light node corresponds to a supplementary survey and control light. The node characteristics of each supplementary survey and control light node include the fish behavior response video in the net cage corresponding to the supplementary survey and control light and the global installation information of the fish gathering and growth promotion control lights.

[0100] Multiple edges between nodes are used to connect adjacent nodes in the control light map, and the edges represent the positional relationship information between the control lights corresponding to the nodes.

[0101] The location relationship information includes the relative orientation angle and spatial distance between adjacent control lights.

[0102] Step S62: Based on the control scheme determination model, process the control lamp spectrum to determine multiple preliminary control schemes for fish gathering and growth promotion control lamps in the net cage. Each preliminary control scheme includes a preliminary output optical power for each fish gathering and growth promotion control lamp.

[0103] In some embodiments, the control scheme determination model is a graph neural network model. The input of the control scheme determination model is the control light spectrum, and the output of the control scheme determination model is multiple preliminary control schemes for the fish gathering and growth promoting control lights in the net cage.

[0104] Graph Neural Network (GNN) models consist of a Graph Neural Network (GNN) and fully connected layers. A GNN is a deep learning architecture specifically designed for processing graph data. GNNs continuously update the feature representation of nodes by aggregating information from neighboring nodes to the target node through message passing mechanisms. GNNs can simultaneously utilize the topological connections in the graph and the node's own attributes for high-dimensional feature inference.

[0105] The multiple preliminary linkage control schemes for the fish-attracting and growth-promoting lights in the net cage are determined by the control scheme determination model. These schemes configure the linked output optical power for all fish-attracting and growth-promoting lights in the net cage. Each preliminary linkage control scheme includes a preliminary output optical power corresponding to each fish-attracting and growth-promoting light in the net cage.

[0106] The lighting control map uses a graphical structure to represent the lighting control relationships within deep-sea cages, visually displaying the physical and spatial topological relationships of each survey and supplementary survey and control light. Survey and supplementary survey and control light nodes respectively carry video feeds of fish behavior responses within their respective cages, as well as installation information for each fish-gathering and growth-promoting control light. Edges between nodes characterize the physical topological connections and spatial constraints between the control lights. This structured representation facilitates the understanding and processing of complex spatial interaction interference relationships by graph neural networks. Through node features and edge information, graph neural networks can better capture the similarities and differences in the phototactic behavior of fish swarms triggered by different control lights acting individually. For example, the video feeds of fish behavior responses within the cages in the node features provide the spatial response evolution levels of each control light under independent test conditions, such as fish aggregation density, swimming speed, and direction. The installation information clarifies its spatial physical coordinates and rated output capacity, while the edges reflect the transmission interactions of these phototactic response characteristics under specific spatial topologies, including cross-overlapping, synergistic gains, or mutual cancellation. The construction of the control light map effectively organizes the various control lights and their lighting response relationships, facilitating multi-source fusion processing by graph neural networks. This organization reduces the computational complexity of graph neural networks and improves the efficiency of model training and multi-objective collaborative inference.

[0107] Because the control light map provides structured node and edge information, graph neural networks can better handle this temporal and spatially coupled data, avoiding the data feature sparsity and spatial dimensionality curse problems caused by traditional centralized control methods when dealing with multi-source temporal video and discrete physical coordinates. This structural advantage makes graph neural networks more advantageous in handling global collaborative relationships. By comprehensively analyzing the video and installation features of surveyed and adjusted control light nodes, as well as their spatial topological location relationships, graph neural networks can find the optimal initial control scheme that achieves a balanced attraction and growth promotion effect across the entire cage while avoiding stress and tremors in the fish population caused by excessive local light intensity.

[0108] When processing the control light map, the graph neural network uses the node features of each survey control light node and each supplementary survey control light node as the initial embedding. It performs message passing along the edges connecting nodes, aggregating the fish behavior response data and installation information of adjacent nodes to the target node to update the target node's feature representation. The relative azimuth angle and spatial distance recorded by the edges determine the influence weight of adjacent nodes during aggregation. Two nodes that are spatially close and have overlapping illumination coverage have a stronger mutual influence. This allows the graph neural network to perceive the synergistic gain or mutual cancellation effect of the attraction performance of surrounding lights on the current light position when updating the features of a node. After multiple rounds of message passing, the embedded features of each node incorporate the attraction experience of all spatially associated nodes. The fully connected layer then maps the updated node embeddings to the optical power output space, calculating the initial output optical power for each fish-attracting and growth-promoting control light in the cage to achieve a balanced attraction effect globally when multiple lights work together. Different power configuration combinations have different emphases on the uniformity of fish aggregation and the intensity of phototactic response. The model can derive multiple different power configuration combinations, each of which covers a preliminary output light power corresponding to each of the fish gathering and growth promotion control lights in the net cage. Finally, it outputs multiple linkage preliminary control schemes for the fish gathering and growth promotion control lights in the net cage.

[0109] Step S63: Based on the fish behavior response video in the net cage corresponding to each survey and control light, the fish behavior response video in the net cage corresponding to each supplementary survey and control light, the light intensity tomographic scan image of the aquaculture water area, and the multiple linkage preliminary control schemes of the fish gathering and growth promoting control lights in the net cage, generate a simulation video of the fish behavior response in the net cage under each linkage preliminary control scheme.

[0110] In some embodiments, a behavioral response simulation model can be used to generate simulated videos of fish behavior responses within net cages under each preliminary coordinated control scheme. The behavioral response simulation model is a generative adversarial network (GAN). The inputs to the behavioral response simulation model are the fish behavior response videos within net cages corresponding to each survey and control light, the fish behavior response videos within net cages corresponding to each supplementary survey and control light, the tomographic image of the light intensity in the aquaculture area, and multiple preliminary coordinated control schemes for the fish-gathering and growth-promoting control lights within the net cages. The output of the behavioral response simulation model is the simulated video of fish behavior responses within net cages under each preliminary coordinated control scheme.

[0111] Generative Adversarial Networks (GANs) consist of two interplaying subnetworks: a generator and a discriminator. The generator captures the probability distribution of real data and generates highly realistic simulation data, while the discriminator distinguishes whether the input data originates from real samples or the generator. Through adversarial training, the GAN continuously optimizes the generator's parameters, ultimately enabling it to produce highly realistic simulation data.

[0112] The simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme is generated by the behavior response simulation model. It simulates the dynamic process of the fish behavior response in the net cage when all the fish gathering and growth promotion control lights are linked for illumination under each preliminary linkage control scheme.

[0113] The simulation video of fish behavior response in the net cage under each preliminary linkage control scheme can simulate the phototaxis aggregation process of the fish, the change of the swimming direction of the fish, and the temporal change of the fish density under the preliminary linkage control scheme.

[0114] The video recordings of fish behavior responses in the net cages corresponding to each survey and control light, along with the video recordings of fish behavior responses in the net cages corresponding to each supplementary survey and control light, collectively record the actual phototaxis and aggregation process, swimming direction changes, and density temporal changes of fish when different light positions are individually turned on. This constitutes a sample library of the real behavioral patterns of fish under single-light stimulation at different light positions, providing an independent lighting behavior reference for the model to learn how the behavior of fish evolves when multiple fish-gathering and growth-promoting control lights are linked. The tomographic images of light intensity in the aquaculture water area are labeled with the light intensity distribution at different water depths, enabling the model to consider the natural attenuation effect of water on light when generating linked simulation images, ensuring that the light performance at each water depth in the simulation video conforms to the actual light intensity distribution characteristics of the water area.

[0115] The generator of the generative adversarial network (GAN) analyzes the behavioral response videos of fish in the cages corresponding to each survey and control light and each supplementary survey and control light, enabling it to grasp the temporal evolution and pixel gradation patterns of fish swimming direction changes. After constructing a 3D background image of the basic illumination attenuation of the aquaculture water body from a tomographic scan of the illumination intensity, the generator can analyze the initial output light power of each of the multiple linked initial control schemes for fish-attracting and growth-promoting lights in the cages. It then superimposes the initial output light power of multiple light sources onto the background image to render a dynamic global light field distribution image. Based on the learned pixel displacement patterns of the fish, the generator synthesizes temporal change frames of fish density within the rendered global light field, generating a continuous series of video frames. The discriminator receives a series of video frames output by the generator, and simultaneously takes in videos of fish behavior responses in the net cage corresponding to each survey and control light and each supplementary survey and control light as benchmarks. It then compares the pixel-level quantization differences between the two in terms of halo attenuation gradient edges and fish aggregation density. When the discriminator identifies a sudden change in light intensity or a fish swimming trajectory that does not conform to actual behavior in the series of video frames synthesized by the generator, it feeds back the error partial derivative gradient to the generator. The generator adjusts the multi-light source fusion attenuation weights and the fish pixel displacement inertia parameter scalar in the synthesis network according to the error partial derivative gradient. The generative adversarial network continuously narrows the digital fitting distance between the synthesized image and the real physical performance through repeated adversarial training. Finally, it synthesizes a continuous dynamic evolution process for each set of linkage initial selection control schemes and outputs a simulation video of fish behavior responses in the net cage under each linkage initial selection control scheme.

[0116] Step S64: Determine the optimal output light power of each fish-gathering and growth-promoting control lamp based on the simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme.

[0117] In some embodiments, an optical power determination model can be used to determine the optimal output optical power of each fish-attracting and growth-promoting control lamp. The optical power determination model is a gated loop unit. The input to the optical power determination model is a simulation video of the fish behavior response within the net cage under each preliminary linkage control scheme, and the output of the optical power determination model is the optimal output optical power of each fish-attracting and growth-promoting control lamp.

[0118] The optimal output optical power of each fish-attracting and growth-promoting control lamp is determined by an optical power determination model, which is used to ultimately control the optimal light power value of each fish-attracting and growth-promoting control lamp in the deep-sea cage for actual operation.

[0119] The simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme simulates the phototaxis aggregation process, swimming direction changes and density evolution of the fish under the corresponding linkage power configuration. The final aggregation density and uniformity of the fish in the video can reflect the quality of the attraction effect of the preliminary linkage control scheme, so that the model can use the fish behavior results in the simulation video as the data basis for judging the quality of different power configurations.

[0120] The gated loop unit uses an update gate to unfold the simulated video of fish behavior response in the net cage under each initial linkage control scheme frame by frame along the time series. This allows for the calculation and extraction of the stability of the fish's phototactic aggregation process as a numerical pixel feature, while simultaneously tracking the discrete variance distribution of the temporal fluctuations in fish density. After the reset gate removes the footage of fish swimming direction changes causing stress oscillations in the initial stage of the simulated video, the gated loop unit accumulates and records the relative duration of fish distribution within the aggregation area within each time step loop. This allows for the analysis of the equilibrium and long-term steady-state of fish spatial distribution under different initial linkage control schemes. When processing simulation videos of fish behavior response in a net cage under a certain preliminary linkage control scheme, and finding that the fish frequently dodge in high-light-intensity areas and exhibit huge variance in density temporal fluctuations, the parameter evaluation network within the gated loop unit performs a strong downgrade calculation on the effectiveness of the preliminary linkage control scheme. Conversely, when analyzing simulation videos of fish behavior response under another preliminary linkage control scheme, and finding that the fish steadily and uniformly converge with regular cruising patterns, the parameter evaluation network confirms that the preliminary linkage control scheme effectively promotes growth in terms of light distribution without overstimulation. The gated loop unit performs horizontal numerical comparison and sequential arrangement of the temporal evaluations of each preliminary linkage control scheme on a one-dimensional coordinate system, selecting the scheme with the highest overall stability score. It then deeply analyzes and obtains the specific quantitative parameter values ​​configured within this scheme, ultimately outputting the obtained parameter values ​​precisely as the optimal output light power for each fish-gathering and growth-promoting control lamp.

[0121] Step S7: Based on the optimal output light power of each fish-attracting and growth-promoting control lamp, control each fish-attracting and growth-promoting control lamp to emit light to the fish in the deep-sea cage.

[0122] Once the optimal output light power of each fish-attracting and growth-promoting control lamp is determined, each lamp is controlled to emit light to the fish in the deep-sea cage based on its optimal output light power, so as to achieve stable phototaxis and aggregation of the fish in the deep-sea cage, as well as balance and stability of the fish spatial distribution.

[0123] This step is existing technology and will not be described in detail here.

[0124] Based on the same inventive concept Figure 6 A schematic diagram of a control system for an artificial intelligence-based underwater lighting device is provided as an embodiment of the present invention. The control system for the artificial intelligence-based underwater lighting device includes:

[0125] The acquisition module 81 is used to acquire initial monitoring video of the spatial distribution of fish in the net cage, tomographic scan image of light intensity in the aquaculture water area, and installation information of each fish gathering and growth promoting control lamp;

[0126] The initial parameter determination module 82 is used to determine multiple survey and control lights and the initial output light power of each survey and control light based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic scan image of the light intensity of the aquaculture water area, and the installation information of each fish gathering and growth promoting control light.

[0127] The survey lighting control module 83 is used to control each survey control light to emit light according to the corresponding initial output light power, and to acquire the fish behavior response video in the net cage corresponding to each survey control light.

[0128] The supplementary parameter determination module 84 is used to determine the supplementary output optical power of multiple supplementary survey and control lights and each supplementary survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light.

[0129] The supplementary lighting control module 85 is used to control each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and to acquire the fish behavior response video in the net cage corresponding to each supplementary survey and control light.

[0130] The optimal parameter determination module 86 is used to determine the optimal output light power of each fish gathering and growth promoting control lamp based on the tomographic scan image of the light intensity of the aquaculture water area, the installation information of each fish gathering and growth promoting control lamp, the fish behavior response video in the net cage corresponding to each survey control lamp, and the fish behavior response video in the net cage corresponding to each supplementary survey control lamp.

[0131] The optimal illumination control module 87 is used to control each fish gathering and growth promoting control lamp to emit light to the fish in the deep-sea cage based on the optimal output light power of each fish gathering and growth promoting control lamp.

[0132] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0133] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A control method for an underwater lighting device based on artificial intelligence, characterized in that, include: Acquire initial monitoring video of the spatial distribution of fish in the net cage, tomographic scan image of light intensity in the aquaculture area, and installation information of each fish-attracting and growth-promoting control lamp; Based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic image of the light intensity in the aquaculture area, and the installation information of each fish gathering and growth promoting control lamp, the initial output light power of multiple survey control lamps and each survey control lamp is determined. Each survey and control light is controlled to emit light according to its corresponding initial output light power, and video of the fish behavior response in the net cage corresponding to each survey and control light is acquired. Based on the fish behavior response video in the cage corresponding to each survey and control light, determine multiple supplementary survey and control lights and the supplementary output light power of each supplementary survey and control light; Control each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and acquire video of the fish behavior response in the net cage corresponding to each supplementary survey and control light; Based on the tomographic scan image of the light intensity of the aquaculture water area, the installation information of each fish gathering and growth promoting control lamp, the video of the fish behavior response in the net cage corresponding to each survey control lamp, and the video of the fish behavior response in the net cage corresponding to each supplementary survey control lamp, the optimal output light power of each fish gathering and growth promoting control lamp is determined. Based on the optimal output light power of each fish-attracting and growth-promoting control lamp, each fish-attracting and growth-promoting control lamp emits light to illuminate the fish in the deep-sea cage.

2. The control method for an artificial intelligence-based underwater lighting device as described in claim 1, characterized in that, The determination of the optimal output light power for each fish-attracting and growth-promoting control lamp, based on the tomographic scan image of the light intensity in the aquaculture area, the installation information of each fish-attracting and growth-promoting control lamp, the video of the fish behavior response in the net cage corresponding to each survey control lamp, and the video of the fish behavior response in the net cage corresponding to each supplementary survey control lamp, includes: A control light map is constructed, which includes multiple nodes and multiple edges between the nodes. The multiple nodes include multiple survey control light nodes and multiple supplementary survey control light nodes. The node characteristics of each survey control light node are the fish behavior response video in the net cage corresponding to each survey control light and the installation information of each fish gathering and growth promoting control light. The node characteristics of each supplementary survey control light node are the fish behavior response video in the net cage corresponding to each supplementary survey control light and the installation information of each fish gathering and growth promoting control light. The edges represent the positional relationship information between the nodes. Based on the control scheme determination model, the control lamp spectrum is processed to determine multiple linkage preliminary control schemes for the fish gathering and growth promoting control lamps in the net cage. Each linkage preliminary control scheme includes a preliminary output light power for each fish gathering and growth promoting control lamp. Based on the fish behavior response video in the net cage corresponding to each survey and control light, the fish behavior response video in the net cage corresponding to each supplementary survey and control light, the light intensity tomographic scan image of the aquaculture water area, and multiple linkage preliminary control schemes of the fish gathering and growth promoting control lights in the net cage, a simulation video of the fish behavior response in the net cage under each linkage preliminary control scheme is generated. Based on the simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme, the optimal output light power of each fish gathering and growth promotion control lamp is determined.

3. The control method for an artificial intelligence-based underwater lighting device as described in claim 1, characterized in that, The determination of multiple supplementary survey and control lights and the supplementary output optical power of each supplementary survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light includes: Based on the fish behavior response video in the net cage corresponding to each survey and control light, multiple response fish movement status information corresponding to each survey and control light are determined. Clustering is performed based on the multiple response fish movement state information corresponding to each survey and control light to obtain multiple behavioral feature clusters corresponding to each survey and control light; Based on the multiple behavioral feature clusters corresponding to each survey and control light, multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light are determined. Based on the installation information of each fish-attracting and growth-promoting control light, and the multiple overactive and sluggish behavior zones corresponding to each survey control light, multiple supplementary survey control lights and the supplementary output light power of each supplementary survey control light are determined.

4. The control method for an artificial intelligence-based underwater lighting device as described in claim 2, characterized in that, The control scheme is determined by a graph neural network model.

5. A control system for an underwater lighting device based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire initial monitoring videos of the spatial distribution of fish in the net cage, tomographic scan images of light intensity in the aquaculture area, and installation information of each fish-gathering and growth-promoting control lamp; The initial parameter determination module is used to determine multiple survey and control lights and the initial output light power of each survey and control light based on the initial monitoring video of the spatial distribution of fish in the net cage, the tomographic scan image of the light intensity of the aquaculture water area, and the installation information of each fish gathering and growth promoting control light. The survey lighting control module is used to control each survey control light to emit light according to the corresponding initial output light power, and to acquire the fish behavior response video in the net cage corresponding to each survey control light. The supplementary parameter determination module is used to determine the supplementary output optical power of multiple supplementary survey and control lights and each supplementary survey and control light based on the fish behavior response video in the net cage corresponding to each survey and control light. The supplementary lighting control module is used to control each supplementary survey and control light to emit light according to the corresponding supplementary output light power, and to acquire the fish behavior response video in the net cage corresponding to each supplementary survey and control light. The optimal parameter determination module is used to determine the optimal output light power of each fish gathering and growth promoting control lamp based on the tomographic scan image of the light intensity of the aquaculture water area, the installation information of each fish gathering and growth promoting control lamp, the fish behavior response video in the net cage corresponding to each survey control lamp, and the fish behavior response video in the net cage corresponding to each supplementary survey control lamp. The optimal illumination control module is used to control the illumination emission of each fish-attracting and growth-promoting control lamp to the fish in the deep-sea cage based on the optimal output light power of each fish-attracting and growth-promoting control lamp.

6. The control system for the artificial intelligence-based underwater lighting device as described in claim 5, characterized in that, The optimal parameter determination module is also used for: A control light map is constructed, which includes multiple nodes and multiple edges between the nodes. The multiple nodes include multiple survey control light nodes and multiple supplementary survey control light nodes. The node characteristics of each survey control light node are the fish behavior response video in the net cage corresponding to each survey control light and the installation information of each fish gathering and growth promoting control light. The node characteristics of each supplementary survey control light node are the fish behavior response video in the net cage corresponding to each supplementary survey control light and the installation information of each fish gathering and growth promoting control light. The edges represent the positional relationship information between the nodes. Based on the control scheme determination model, the control lamp spectrum is processed to determine multiple linkage preliminary control schemes for the fish gathering and growth promoting control lamps in the net cage. Each linkage preliminary control scheme includes a preliminary output light power for each fish gathering and growth promoting control lamp. Based on the fish behavior response video in the net cage corresponding to each survey and control light, the fish behavior response video in the net cage corresponding to each supplementary survey and control light, the light intensity tomographic scan image of the aquaculture water area, and multiple linkage preliminary control schemes of the fish gathering and growth promoting control lights in the net cage, a simulation video of the fish behavior response in the net cage under each linkage preliminary control scheme is generated. Based on the simulation video of the fish behavior response in the net cage under each preliminary linkage control scheme, the optimal output light power of each fish gathering and growth promotion control lamp is determined.

7. The control system for the artificial intelligence-based underwater lighting device as described in claim 5, characterized in that, The supplementary parameter determination module is also used for: Based on the fish behavior response video in the net cage corresponding to each survey and control light, multiple response fish movement status information corresponding to each survey and control light are determined. Clustering is performed based on the multiple response fish movement state information corresponding to each survey and control light to obtain multiple behavioral feature clusters corresponding to each survey and control light; Based on the multiple behavioral feature clusters corresponding to each survey and control light, multiple overactive behavior zones and multiple sluggish behavior zones corresponding to each survey and control light are determined. Based on the installation information of each fish-attracting and growth-promoting control light, and the multiple overactive and sluggish behavior zones corresponding to each survey control light, multiple supplementary survey control lights and the supplementary output light power of each supplementary survey control light are determined.

8. The control system for the artificial intelligence-based underwater lighting device as described in claim 6, characterized in that, The control scheme is determined by a graph neural network model.

9. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the control method for an artificial intelligence-based underwater illuminator as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the control method for an artificial intelligence-based underwater lighting device as described in any one of claims 1 to 4.

Images