Intelligent anesthesia method and device for fish based on biomimetic electro-optical combined neural inhibition

Abstract

This invention relates to the field of fish anesthesia equipment technology, particularly to a smart anesthesia method and device for fish based on biomimetic electro-optic combined neural inhibition. It includes an AI intelligent recognition module, a biomimetic electrical stimulation module, an optical neural inhibition module, and a safety protection module. The AI ​​intelligent recognition module identifies the species using an underwater camera; the biomimetic electrical stimulation module simulates the characteristics of eel electroorgans, generating intermittent high-frequency cluster pulses; the optical neural inhibition module uses a near-infrared light source to prolong the neural inhibition time; and the safety protection module includes adaptive conductive gel electrodes and a magnetic field cancellation device. This invention achieves precise and intelligent anesthesia for different species of marine fish through the combination of rapid electroanesthesia and long-term photo-inhibition, making it suitable for large-scale fishing vessel sorting lines, scientific sampling, and protective fishing scenarios, significantly reducing fish stress responses.

Description

Technical Field

[0001] This invention relates to the field of fish anesthesia equipment technology, and in particular to a smart anesthesia method and device for fish based on biomimetic electro-optic combined nerve inhibition. Background Technology

[0002] Fish often suffer injuries due to struggling during fishing, sorting, transportation, or scientific sampling. Anesthesia can humanely manage these injuries, reducing fish suffering and improving survival rates and meat quality. Traditional fish anesthesia methods mainly include chemical anesthesia and electroconvulsive anesthesia. While chemical anesthesia is simple to perform, it has drawbacks such as drug residues, environmental pollution, and long recovery times. Furthermore, it is difficult to control the depth of anesthesia, potentially causing irreversible damage to fish. Electroconvulsive anesthesia, as a physical method, has advantages such as rapid onset and no drug residue. However, existing electroconvulsive devices mostly use fixed-frequency square wave signals, resulting in a monotonous waveform that does not conform to the natural characteristics of the biological nervous system. This can easily induce strong muscle contractions and stress responses. Additionally, the lack of intelligent recognition capabilities means that anesthesia parameters cannot be automatically adjusted according to the species and size of different fish, leading to uneven anesthesia effects, short duration of anesthesia, and rapid recovery, making subsequent processing difficult. The current diffusion range is also difficult to control, potentially harming surrounding marine life. Furthermore, the lack of a safety monitoring mechanism makes it impossible to monitor the fish's physiological state in real time, posing safety hazards. Summary of the Invention

[0003] This invention overcomes the shortcomings of the prior art and provides a method and device for intelligent anesthesia of fish based on biomimetic electro-optic combined neural inhibition.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] The first aspect of this invention provides a smart anesthesia method for fish based on biomimetic electro-optical combined neural inhibition, comprising the following steps:

[0006] S102: Capture and photograph fish using an underwater camera to obtain real-time image information of the fish to be anesthetized, and simultaneously acquire underwater point cloud, water environment data and underwater positioning information of the target fish area through a multi-sensor fusion array;

[0007] S104: Combine water environment data and underwater positioning information to project the underwater point cloud onto the semantically segmented real-shot image information for mask rendering, generate an underwater enhanced motion simulation field, and identify the characteristics and movement of the fish to be anesthetized expressed by the underwater enhanced motion simulation field based on the image processing unit, and obtain the species information, body size and movement trajectory of the fish to be anesthetized.

[0008] S106: Based on empirical cases of fish anesthesia, a discrete value mapping library of preset fish-anesthesia parameters is constructed by discrete combination. Simultaneously, anesthesia constraints to be tested are created based on species information, body size, and movement trajectory and anesthesia parameters. Different anesthesia constraints to be tested are tested through the discrete value mapping library, and the optimal anesthesia parameters are output.

[0009] S108: Controls the intelligent electrode array to contact the fish body according to the optimal anesthesia parameters, releases biomimetic cluster pulses to quickly anesthetize, controls the near-infrared laser to irradiate the nerve-dense area of ​​the fish to be anesthetized, and obtains several actual physiological parameter values ​​of the fish to be anesthetized in different physiological activity indicators in real time through the safety protection module.

[0010] S110: Construct specific clause bottom boundaries that do not touch the physiological safety threshold based on the normal physiological operation mechanism and physiological safety thresholds of different physiological activity indicators of the fish to be anesthetized. Use the specific clause bottom boundaries to perform anomaly judgment analysis on the actual physiological parameter values ​​to respond to the control and early warning device and anesthesia procedure.

[0011] More specifically, step S104 includes the following steps:

[0012] The BiSe semantic segmentation network is invoked to perform different pixel-level classification on each pixel source point contained in the real-shot image information, generate the category confidence distribution probability of each pixel source point, and construct the semantic segmentation map of the real-shot image information based on the category confidence distribution probability.

[0013] The perception strategy and calibration logic of the multi-sensor fusion array are obtained. The multi-sensor calibration matrix is ​​established by combining the underwater positioning information, perception strategy and calibration criteria. Based on the water environment data, the multi-sensor calibration matrix is ​​used to transform and project the intrinsic and extrinsic parameters of each point cloud seed in the underwater point cloud onto the image plane of the real image information to generate a projection joint dependency map of point cloud seed-pixel source point.

[0014] Using the source pixel corresponding to each point cloud seed in the projection joint dependency graph as the source object, the semantic category confidence distribution of the source object is queried from the semantic segmentation graph to obtain the one-hot vector mask of each source object. Based on the projection joint dependency graph, the one-hot vector mask is colored and applied to the origin feature of the corresponding point cloud seed of each source object for highlighting rendering construction, so as to obtain the underwater enhanced motion simulation field carrying the anesthetized fish in the target fish area.

[0015] The deep learning inference engine, feature extraction operator library, and trajectory prediction architecture of the image processing unit are acquired. Based on the deep learning inference engine, dynamic voxel feature processing is performed on the underwater enhanced motion simulation field to obtain the BEV feature trajectory map of the fish to be anesthetized in a time-series dynamic migration. Based on the feature extraction operator library and trajectory prediction architecture, the BEV feature trajectory map is identified, and the species information, body size, and movement trajectory of the fish to be anesthetized are output.

[0016] More specifically, the deep learning inference engine, feature extraction operator library, and trajectory prediction architecture of the image processing unit perform dynamic voxel feature processing on the underwater enhanced motion simulation field based on the deep learning inference engine to obtain the BEV feature trajectory map of the fish to be anesthetized in a time-series dynamic migration. Based on the feature extraction operator library and trajectory prediction architecture, the BEV feature trajectory map is identified, and the species information, body size, and movement trajectory of the fish to be anesthetized are output. The specific steps include:

[0017] The deep learning inference engine, feature extraction operator library and trajectory prediction architecture of the image processing unit are obtained. The sparse voxel algorithm is introduced to weave the network system of feature recognition and perception of the deep learning inference engine, and a multi-dimensional voxel perception network is constructed.

[0018] The multidimensional voxel sensing network is used to voxelize the point cloud region where the fish to be anesthetized is located in the target fish region in an underwater enhanced motion simulation field, and to obtain the global voxel feature map of the fish to be anesthetized affected by the environmental changes of the target fish region.

[0019] By sampling the starting point cloud region of the fish to be anesthetized in an underwater enhanced motion simulation field, multiple key temporal transition sites of the starting point cloud were obtained. The key temporal transition sites were interpolated from the global voxel feature map to obtain the voxel features of each key temporal transition site, which are defined as temporal transition voxel features.

[0020] A BEV mesh for underwater enhanced motion simulation field is constructed. The temporal transition voxel features are fused and stitched with the point cloud features of each key temporal transition point according to the positional rules of the BEV mesh to form a BEV feature trajectory map of the dynamic migration of the fish to be anesthetized in the target fish area in time.

[0021] A lightweight recognition network is established based on the computational logic of the feature extraction operator library and trajectory prediction architecture. The lightweight recognition network performs comprehensive recognition of the fish to be anesthetized on the BEV feature trajectory map by distinguishing features, regressing positions, and predicting directions. It outputs feature classification scores and trajectory bounding box regression chains. Based on the feature classification scores and trajectory bounding box regression chains, the species information, body size, and movement trajectory of the fish to be anesthetized are determined.

[0022] More specifically, step S106 includes the following steps:

[0023] To obtain the predetermined anesthesia requirements of fish, several optimal anesthesia cases simulating the natural discharge characteristics of different fish were obtained through big data network retrieval.

[0024] By extracting the range of anesthesia parameters required for different preset fish species based on species information, body size, and movement trajectory from several optimal anesthesia cases, and discretizing the parameter values ​​for anesthesia of each preset fish species based on species information, body size, and movement trajectory according to the given anesthesia requirements, a discrete value mapping library of preset fish species and anesthesia parameters is constructed; among which, anesthesia parameters include voltage, frequency, and light intensity.

[0025] Anesthesia constraint prefixes are created based on species information, body size, and movement trajectory. Anesthesia parameters such as voltage, frequency, and light intensity are defined as anesthesia variable matrices and added to the anesthesia constraint prefixes to generate an anesthesia constraint test queue for fish to be anesthetized.

[0026] Each anesthesia constraint in the anesthesia constraint test queue is loaded in descending order and marked as an anesthesia constraint to be tested. A set of discrete values ​​is randomly extracted from the discrete value mapping library of fish-anesthesia parameters and injected into the anesthesia variables of the anesthesia constraint to be tested. The test processing is performed on the anesthesia constraint to be tested, and the constraint individuals corresponding to the set of discrete values ​​are obtained and marked as empirical anesthesia constraints.

[0027] Construct a candidate empty stack, introduce an expert hash algorithm to calculate the hash misalignment function of the anesthesia constraint to be tested relative to the empirical anesthesia constraint. If the hash misalignment function is greater than the preset hash misalignment function, then remove the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested; if the hash misalignment function is greater than the preset hash misalignment function, then transfer the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested and record them to the candidate empty stack.

[0028] Repeat the above steps of random injection and verification of discrete values ​​to verify and analyze the remaining anesthesia constraints to be tested in the anesthesia constraint verification queue. Finally, only the candidate discrete values ​​corresponding to the lowest hash misalignment function are extracted as the optimal anesthesia parameters output.

[0029] More specifically, step S110 includes the following steps:

[0030] Multiple physiological activity indicators of fish to be anesthetized were extracted and monitored to assess the anesthesia requirements based on the established anesthesia needs, and the physiological safety threshold of each physiological activity indicator was obtained; wherein, the physiological activity indicators include heart rate, respiratory rate, body temperature change and cortisol level;

[0031] Based on big data networks, the normal physiological operation mechanism of the fish to be anesthetized in the target fish area is obtained, and several past physiological parameter values ​​of various physiological activity indicators are obtained under the premise that they are in line with the normal physiological operation mechanism and do not exceed the physiological safety threshold.

[0032] A pattern matching algorithm is introduced, and based on the normal physiological operating mechanism and physiological safety threshold, several past physiological parameter values ​​are deduced in reverse in the pattern matching algorithm to construct a specific clause bottom boundary that does not touch the physiological safety threshold.

[0033] Based on several actual physiological parameter values ​​of each physiological activity indicator, an actual physiological clause exploration domain is established. The bottom boundary of the specific clause is used as the search strategy criterion for the normal physiological operation mechanism to perform a global traversal search on each inductive clause in the actual physiological clause exploration domain, and the coverage violation degree of each inductive clause is obtained.

[0034] If the fish to be anesthetized does not have actual physiological parameter values ​​corresponding to inductive clauses with coverage violation greater than the preset coverage violation, then the biomimetic electro-optical combined anesthesia procedure will continue to be executed.

[0035] If at least one or more inductive clauses with a coverage violation degree greater than the preset coverage violation degree have corresponding actual physiological parameter values, the early warning device will immediately trigger an audible and visual alarm and automatically interrupt the anesthesia procedure.

[0036] A second aspect of the present invention provides a smart anesthesia device for fish based on biomimetic electro-optical combined neural inhibition, the smart anesthesia device for fish comprising:

[0037] The AI ​​intelligent recognition module includes an underwater camera, an image processing unit, and a multi-sensor fusion array, used to identify comprehensive recognition data including fish species, body size parameters, spatial location, and movement trajectory.

[0038] A biomimetic electrical stimulation module, comprising a waveform generator, a power amplifier, and a smart electrode array, is used to simulate the natural discharge characteristics of electrical organs to generate electrical stimulation signals that more closely resemble physiological processes for biomimetic electrical stimulation anesthesia of fish.

[0039] An optical neural inhibition module, comprising a near-infrared laser, an optical fiber transmission unit, and an illumination control unit, is responsible for releasing near-infrared light on the basis of electrical stimulation anesthesia to act on the ion channels and membrane potential of fish neurons, thereby prolonging neural inhibition;

[0040] The safety protection module includes a magnetic field cancellation device, a biological monitoring unit, and an adaptive conductive gel device, which are responsible for protecting the operational safety of fish biomimetic electro-optic anesthesia.

[0041] The central control module includes a main controller, a parameter adjustment panel, and a display screen, and is responsible for receiving data, automatically switching suppression modes, and manually adjusting operating parameters.

[0042] Furthermore, in a preferred embodiment of the present invention, the multi-sensor fusion array includes an ultrasonic ranging sensor, an underwater lidar, a water flow velocity sensor, a water quality detection sensor, and an inertial measurement unit.

[0043] Furthermore, in a preferred embodiment of the present invention, the intelligent electrode array includes a multi-electrode array substrate, an electrode spacing adjustment mechanism, and a contact detection sensor.

[0044] Furthermore, in a preferred embodiment of the present invention, the optical fiber transmission unit includes a multimode fiber bundle, an optical fiber coupler, an optical fiber splitter, and an adjustable optical fiber probe array.

[0045] Furthermore, in a preferred embodiment of the present invention, the adaptive conductive gel device includes a gel dispensing device, a contact pressure sensor, and a gel composition adjustment unit.

[0046] This invention addresses the technical deficiencies in the prior art, and its beneficial technical effects are as follows:

[0047] Compared with existing technologies, this invention combines artificial intelligence recognition technology, biomimetic electrical stimulation technology, and optical nerve inhibition technology. Through artificial intelligence, it can identify species and body size, and connect to a cloud database to guide the intensity of the electric motor and light exposure, maximizing the protection of the anesthetized subject without causing damage. The combination of biomimetic electrical stimulation and optical nerve inhibition technologies allows the artificial intelligence module to provide the electric motor intensity and duration, as well as the light intensity and duration, which are then implemented by the electrical stimulation and light inhibition modules. Electrical stimulation is the main step in anesthesia, while the light inhibition module prolongs the anesthesia time and reduces muscle tension-related damage to the fish during the anesthesia process. This design considers more aspects of fish physiology, making the method more rational and applicable, achieving intelligent anesthesia for fish. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0049] Figure 1 A schematic diagram of the overall structure of a smart anesthesia device for fish based on biomimetic electro-optical combined neural inhibition is shown.

[0050] Figure 2 The module architecture diagram of the AI ​​intelligent recognition module is shown;

[0051] Figure 3 The array structure diagram of the multi-sensor fusion array is shown;

[0052] Figure 4 The array structure diagram of the smart electrode array is shown;

[0053] Figure 5 The unit architecture diagram of the optical fiber transmission unit is shown;

[0054] Figure 6 A structural diagram of the adaptive conductive gel device is shown.

[0055] Figure 7 A flowchart of a smart anesthesia method for fish based on biomimetic electro-optical combined neural inhibition is shown. Detailed Implementation

[0056] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0057] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0058] The first aspect of this invention provides a smart anesthesia method for fish based on biomimetic electro-optical combined neural inhibition, such as... Figure 7 As shown, it includes the following steps:

[0059] S102: Capture and photograph fish using an underwater camera to obtain real-time image information of the fish to be anesthetized, and simultaneously acquire underwater point cloud, water environment data and underwater positioning information of the target fish area through a multi-sensor fusion array;

[0060] S104: Combine water environment data and underwater positioning information to project the underwater point cloud onto the semantically segmented real-shot image information for mask rendering, generate an underwater enhanced motion simulation field, and identify the characteristics and movement of the fish to be anesthetized expressed by the underwater enhanced motion simulation field based on the image processing unit, and obtain the species information, body size and movement trajectory of the fish to be anesthetized.

[0061] S106: Based on empirical cases of fish anesthesia, a discrete value mapping library of preset fish-anesthesia parameters is constructed by discrete combination. Simultaneously, anesthesia constraints to be tested are created based on species information, body size, and movement trajectory and anesthesia parameters. Different anesthesia constraints to be tested are tested through the discrete value mapping library, and the optimal anesthesia parameters are output.

[0062] S108: Controls the intelligent electrode array to contact the fish body according to the optimal anesthesia parameters, releases biomimetic cluster pulses to quickly anesthetize, controls the near-infrared laser to irradiate the nerve-dense area of ​​the fish to be anesthetized, and obtains several actual physiological parameter values ​​of the fish to be anesthetized in different physiological activity indicators in real time through the safety protection module.

[0063] S110: Construct specific clause bottom boundaries that do not touch the physiological safety threshold based on the normal physiological operation mechanism and physiological safety thresholds of different physiological activity indicators of the fish to be anesthetized. Use the specific clause bottom boundaries to perform anomaly judgment analysis on the actual physiological parameter values ​​to respond to the control and early warning device and anesthesia procedure.

[0064] It should be noted that the water environment data includes water temperature, salinity, pH value, and dissolved oxygen. Underwater positioning information includes the direction and velocity of water flow, as well as the attitude and location of the equipment.

[0065] More specifically, step S104 includes the following steps:

[0066] The BiSe semantic segmentation network is invoked to perform different pixel-level classification on each pixel source point contained in the real-shot image information, generate the category confidence distribution probability of each pixel source point, and construct the semantic segmentation map of the real-shot image information based on the category confidence distribution probability.

[0067] The perception strategy and calibration logic of the multi-sensor fusion array are obtained. The multi-sensor calibration matrix is ​​established by combining the underwater positioning information, perception strategy and calibration criteria. Based on the water environment data, the multi-sensor calibration matrix is ​​used to transform and project the intrinsic and extrinsic parameters of each point cloud seed in the underwater point cloud onto the image plane of the real image information to generate a projection joint dependency map of point cloud seed-pixel source point.

[0068] Using the source pixel corresponding to the projection of each point cloud seed in the projection dependency graph as the source object, query the semantic category confidence distribution of the source object from the semantic segmentation graph, and obtain the one-hot vector mask of each source object.

[0069] Based on the projection joint dependency graph, the unique hot vector mask is added to the origin feature of the corresponding point cloud seed of each source object for highlighting rendering construction, so as to obtain the underwater enhanced motion simulation field carrying the anesthetized fish in the target fish area.

[0070] The deep learning inference engine, feature extraction operator library, and trajectory prediction architecture of the image processing unit are acquired. Based on the deep learning inference engine, dynamic voxel feature processing is performed on the underwater enhanced motion simulation field to obtain the BEV feature trajectory map of the fish to be anesthetized in a time-series dynamic migration. Based on the feature extraction operator library and trajectory prediction architecture, the BEV feature trajectory map is identified, and the species information, body size, and movement trajectory of the fish to be anesthetized are output.

[0071] It should be noted that, in order to simulate the natural discharge characteristics of fish and achieve precise automated anesthesia, this invention employs pre-trained convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to identify fish species and estimate body size, and Kalman filtering and particle filtering to predict movement trajectories. However, since the multi-sensor fusion array visualizes the underwater environment through point cloud reconstruction, while the identification of the fish to be anesthetized relies on image information from a high-definition camera, pixel incompatibility may occur in the representation of fish images within the aquatic point cloud. This leads to significant errors in the identification of fish species, body size, and movement trajectory prediction by CNNs, RNNs, Kalman filters, and particle filters, greatly affecting the subsequent decision-making of anesthesia parameters and the accuracy of anesthesia. To address this, this method utilizes a BiSe semantic segmentation network to perform pixel-level semantic classification on the real-time images of the fish to be anesthetized, thereby clarifying the semantic category probability distribution of each pixel in the image. This effectively compensates for the lack of information in distant or sparse point cloud regions in real-time images, and enhances the point cloud's understanding of real-time images by utilizing the rich texture and color information of the image, providing an interpretable basis for the projection of the subsequent perfect fusion of the point cloud and the image. The multi-sensor calibration matrix established by underwater positioning information, perception strategy and calibration logic is the point cloud coordinate tracking mark of the fish to be anesthetized in the dynamic aquatic environment by the multi-sensor fusion array, which can further make the fusion of image and point cloud more accurate.

[0072] It should be noted that, subsequently, based on the aquatic environment data, a multi-sensor calibration matrix (point cloud coordinate cues) is used to project and map each point cloud seed in the underwater point cloud to the corresponding pixel position in the captured image. This effectively establishes the spatial correspondence between the image and the point cloud, namely, the projection dependency graph of point cloud seed-pixel source point. This aligns the spatial data of two different modalities, achieving cross-modal multi-source data fusion and laying a solid foundation for obtaining the corresponding image semantic information from the point cloud. Then, the semantic segmentation graph is used to query the corresponding pixel source point after projection of each point cloud seed in the projection dependency graph. The semantic label (one-hot vector mask) of the corresponding pixel source point of the point cloud seed in the captured image is attached to its feature to highlight the rendering structure. This can significantly improve the semantic expressive ability of the point cloud for the image, thus enabling the existence of perceptible fish image semantics in the sparse areas of the underwater point cloud environment. Finally, an underwater enhanced motion simulation field is formed, depicting the anesthetized fish in a dynamically changing underwater environment, enhancing the model's ability to distinguish fish species and body sizes. This method can fuse the point cloud modal of the underwater environment with the modal projection of fish images, thereby improving the recognition accuracy of subsequent convolutional neural networks, recurrent neural networks, Kalman filters, and particle filters for the dynamic aquatic environment affecting the fish to be anesthetized under different data modal conditions, and ensuring the anesthesia performance.

[0073] More specifically, the deep learning inference engine, feature extraction operator library, and trajectory prediction architecture of the image processing unit perform dynamic voxel feature processing on the underwater enhanced motion simulation field based on the deep learning inference engine to obtain the BEV feature trajectory map of the fish to be anesthetized in a time-series dynamic migration. Based on the feature extraction operator library and trajectory prediction architecture, the BEV feature trajectory map is identified, and the species information, body size, and movement trajectory of the fish to be anesthetized are output. The specific steps include:

[0074] The deep learning inference engine, feature extraction operator library and trajectory prediction architecture of the image processing unit are obtained. The sparse voxel algorithm is introduced to weave the network system of feature recognition and perception of the deep learning inference engine, and a multi-dimensional voxel perception network is constructed.

[0075] The multidimensional voxel sensing network is used to voxelize the point cloud region where the fish to be anesthetized is located in the target fish region in an underwater enhanced motion simulation field, and to obtain the global voxel feature map of the fish to be anesthetized affected by the environmental changes of the target fish region.

[0076] By sampling the starting point cloud region of the fish to be anesthetized in an underwater enhanced motion simulation field, multiple key temporal transition sites of the starting point cloud were obtained. The key temporal transition sites were interpolated from the global voxel feature map to obtain the voxel features of each key temporal transition site, which are defined as temporal transition voxel features.

[0077] A BEV mesh for underwater enhanced motion simulation field is constructed. The temporal transition voxel features are fused and stitched with the point cloud features of each key temporal transition point according to the positional rules of the BEV mesh to form a BEV feature trajectory map of the dynamic migration of the fish to be anesthetized in the target fish area in time.

[0078] A lightweight recognition network is established based on the computational logic of the feature extraction operator library and trajectory prediction architecture. The lightweight recognition network performs comprehensive recognition of the fish to be anesthetized on the BEV feature trajectory map by distinguishing features, regressing positions, and predicting directions. It outputs feature classification scores and trajectory bounding box regression chains. Based on the feature classification scores and trajectory bounding box regression chains, the species information, body size, and movement trajectory of the fish to be anesthetized are determined.

[0079] It should be noted that the deep learning inference engine is based on a hybrid architecture of convolutional neural networks and recurrent neural networks; the feature extraction operator library integrates morphological feature extraction operators, texture feature analysis operators, and motion pattern recognition operators; the trajectory prediction architecture adopts a fusion architecture based on Kalman filtering and particle filtering algorithms. Traditional fish anesthesia methods often struggle to achieve the coordinated recognition and trajectory prediction of fish affected by dynamic aquatic environments using convolutional neural networks (CNN), recurrent neural networks (RNN), Kalman filtering, and particle filtering, thus reducing the efficiency of fish parameter acquisition. To address this, our method employs a sparse voxel algorithm to weave a deep learning inference engine into a network system capable of voxel feature recognition and perception, constructing a multi-dimensional voxel-aware network. This network is used to voxelize the dynamically changing point cloud regions of the anesthetized fish within the target fish region in an underwater enhanced motion simulation field. This process divides the dense original point cloud into a regular grid and simultaneously calculates the features within each voxel. This transforms sparse and irregular point cloud data into a structure suitable for convolutional neural networks and recurrent neural networks to perform convolution operations, ultimately forming a global voxel feature map that reflects the influence of environmental changes on the features of the anesthetized fish. This global voxel feature map is spatially semantically aware, providing global feature semantic information. This enables convolutional neural networks and recurrent neural networks to understand large-scale spatial relationships, resulting in a more complete and clearer characterization of both local and global features of the anesthetized fish, significantly improving the extraction efficiency and stability of the anesthetized fish features. Next, the starting point cloud region of the fish to be anesthetized is sampled to obtain the key points of each starting point cloud, namely the key temporal transition sites. These key temporal transition sites are the migration positions of the fish to be anesthetized under the influence of aquatic environmental factors in the temporal progression. The voxel features of the global voxel feature map are interpolated and fused to each key point to generate key point features that retain details and have global motion process context. Combining the fine structure of the point cloud and the global semantics of voxels, the dynamic key points obtain richer and more robust local features, enhance the dynamic expression ability of the target features (fish to be anesthetized) key points, and ensure that the prediction of subsequent motion trajectories is more accurate.

[0080] It should be noted that, following the BEV mesh format of the underwater enhanced motion simulation field, the temporal transition voxel features are fused and stitched with the point cloud features of each key temporal transition point. This introduces multi-scale contextual information from the local neighborhood to each key point while filtering out potential target paths for the target's motion features, forming a BEV feature trajectory map that displays both the individual characteristics of the fish to be anesthetized and its motion migration vector. The BEV perspective provides clear spatial relationships and interpretable obstacle locations, greatly improving the processing efficiency of feature recognition and motion prediction. Finally, a lightweight recognition network is established by combining a feature extraction operator library with a trajectory prediction architecture. Target detection is performed on the extracted BEV features to identify the species of the fish to be anesthetized, estimate its size, and predict the position and direction of its motion trajectory. The Chinese name for the BEV feature trajectory map is "bird's-eye view feature trajectory map."

[0081] More specifically, step S106 includes the following steps:

[0082] To obtain the predetermined anesthesia requirements of fish, several optimal anesthesia cases simulating the natural discharge characteristics of different fish were obtained through big data network retrieval.

[0083] By extracting the range of anesthesia parameters required for different preset fish species based on species information, body size, and movement trajectory from several optimal anesthesia cases, and discretizing the parameter values ​​for anesthesia of each preset fish species based on species information, body size, and movement trajectory according to the given anesthesia requirements, a discrete value mapping library of preset fish species and anesthesia parameters is constructed; among which, anesthesia parameters include voltage, frequency, and light intensity.

[0084] Anesthesia constraint prefixes are created based on species information, body size, and movement trajectory. Anesthesia parameters such as voltage, frequency, and light intensity are defined as anesthesia variable matrices and added to the anesthesia constraint prefixes to generate an anesthesia constraint test queue for fish to be anesthetized.

[0085] Each anesthesia constraint in the anesthesia constraint test queue is loaded in descending order and marked as an anesthesia constraint to be tested. A set of discrete values ​​is randomly extracted from the discrete value mapping library of fish-anesthesia parameters and injected into the anesthesia variables of the anesthesia constraint to be tested. The test processing is performed on the anesthesia constraint to be tested, and the constraint individuals corresponding to the set of discrete values ​​are obtained and marked as empirical anesthesia constraints.

[0086] Construct a candidate empty stack, introduce an expert hash algorithm to calculate the hash misalignment function of the anesthesia constraint to be tested relative to the empirical anesthesia constraint. If the hash misalignment function is greater than the preset hash misalignment function, then remove the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested; if the hash misalignment function is greater than the preset hash misalignment function, then transfer the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested and record them to the candidate empty stack.

[0087] Repeat the above steps of random injection and verification of discrete values ​​to verify and analyze the remaining anesthesia constraints to be tested in the anesthesia constraint verification queue. Finally, only the candidate discrete values ​​corresponding to the lowest hash misalignment function are extracted as the optimal anesthesia parameters output.

[0088] It should be noted that after obtaining the species, size, and movement trajectory of the fish to be anesthetized, expert algorithms can be used to retrieve the optimal anesthesia parameters that match the discharge characteristics of simulated deep-sea or surface-sea fish. However, traditional fish databases, even with the support of expert analysis algorithms, typically rely on similarity comparisons of species, size, and movement characteristics to determine parameter decisions for different anesthesia effects. These parameters have a wide range of values; for example, the voltage of the fish to be anesthetized may be between 50-100V and the frequency between 20-40Hz for a highly similar surface-sea fish, making it difficult to determine the exact anesthesia parameters. This can lead to incomplete or excessive anesthesia, resulting in relatively low decision precision and matching reliability. To address this, this method extracts the required range of anesthesia parameters for different preset fish species based on species information, size, and movement trajectory through optimal anesthesia cases. Based on predetermined anesthesia requirements, these parameter values ​​are discretely combined to provide a more refined value scheme for anesthesia parameters required for different species and size differences—essentially a discrete value mapping library of preset fish and anesthesia parameters. It is worth mentioning that the range of anesthesia parameters was extracted from the fish database based on empirical data from optimal anesthesia cases, and the discrete value mapping library was synchronously merged and stored in the fish database. Then, species information, body size, and movement trajectory were used as constraints for the best anesthesia effect. In other words, the decision on the optimal anesthesia parameters was to better adapt to and serve the species information, body size, and movement trajectory of the fish to be anesthetized, and therefore they were set as the anesthesia constraints to be tested.

[0089] It should be noted that each anesthesia constraint to be tested depends on different anesthesia variables. For example, species information requires precise adjustment of anesthesia parameters for voltage and frequency. Therefore, the form of the anesthesia constraint to be tested is species [voltage, frequency]. Then, a set of discrete values ​​is randomly extracted from the constructed discrete value mapping library and added to the anesthesia variables, for example: species [voltage 80V, frequency 35Hz], thereby performing the optimal anesthesia test based on empirical data. If the hash misalignment function of the anesthesia constraint to be tested relative to the empirical anesthesia constraint is greater than the preset hash misalignment function, it means that the set of discrete values ​​used in the previous empirical anesthesia constraint is difficult to simulate and adapt to the natural discharge characteristics of the fish to be anesthetized. Therefore, it cannot achieve the best effect when used on the fish to be anesthetized, and the parameter scheme of the set of discrete values ​​is withdrawn. Conversely, it means that the set of discrete values ​​used in the empirical anesthesia constraint can simulate and conform to the natural discharge characteristics of the fish to be anesthetized, thereby achieving the ideal anesthesia effect. Therefore, the set of discrete values ​​is adopted to achieve precise value positioning. This method allows for the determination of specific and reasonable parameters that can achieve the best anesthetic effect for fish from a broad range of anesthetic values, based on the species, size, and movement trajectory of the fish to be anesthetized. This maximizes the simulation of the natural discharge characteristics of the fish's electrical organs, achieving precise, rapid, and low-damage anesthetic effects, and significantly reducing muscle rigidity and stress damage in fish.

[0090] More specifically, step S108 includes the following steps:

[0091] Multiple physiological activity indicators of fish to be anesthetized were extracted and monitored to assess the anesthesia requirements based on the established anesthesia needs, and the physiological safety threshold of each physiological activity indicator was obtained; wherein, the physiological activity indicators include heart rate, respiratory rate, body temperature change and cortisol level;

[0092] Based on big data networks, the normal physiological operation mechanism of the fish to be anesthetized in the target fish area is obtained, and several past physiological parameter values ​​of various physiological activity indicators are obtained under the premise that they are in line with the normal physiological operation mechanism and do not exceed the physiological safety threshold.

[0093] A pattern matching algorithm is introduced, and based on the normal physiological operating mechanism and physiological safety threshold, several past physiological parameter values ​​are deduced in reverse in the pattern matching algorithm to construct a specific clause bottom boundary that does not touch the physiological safety threshold.

[0094] Based on several actual physiological parameter values ​​of each physiological activity indicator, an actual physiological clause exploration domain is established. The bottom boundary of the specific clause is used as the search strategy criterion for the normal physiological operation mechanism to perform a global traversal search on each inductive clause in the actual physiological clause exploration domain, and the coverage violation degree of each inductive clause is obtained.

[0095] If the fish to be anesthetized does not have actual physiological parameter values ​​corresponding to inductive clauses with coverage violation greater than the preset coverage violation, then the biomimetic electro-optical combined anesthesia procedure will continue to be executed.

[0096] If at least one or more inductive clauses with a coverage violation degree greater than the preset coverage violation degree have corresponding actual physiological parameter values, the early warning device will immediately trigger an audible and visual alarm and automatically interrupt the anesthesia procedure.

[0097] It should be noted that while the safety protection module can monitor the physiological status of the fish to be anesthetized in real time and execute corresponding early warning actions when abnormalities occur, existing early warning methods, when applied to fish anesthesia, struggle to perform abnormal threshold analysis of physiological parameters according to the natural norms of their normal physiological mechanisms. This leads to subtle monitoring and judgment biases, potentially resulting in serious false alarms or missed alarms in the abnormal monitoring response to physiological safety issues. To address this, this method uses a pattern matching algorithm to reverse-engineer past physiological parameter values ​​based on the underlying logic that the normal physiological operation mechanism of the fish to be anesthetized must be within a physiological safety threshold. This applies a premise of index coupling that follows the normal physiological mechanisms of fish to determine the safety anomalies of physiological parameters—a specific clause bottom boundary. This further enhances the constraint and traceability of the physiological safety status monitoring and judgment response among different physiological parameters of the fish to be anesthetized, improving the correlation strength of safety early warnings based on physiological indicators. Next, the coupling premise of this indicator (the bottom boundary of the specific clause) is used as the search strategy criterion for the normal physiological operation mechanism to perform a global traversal search on the monitored actual physiological parameter values. In this case, the exploration domain of the actual physiological clause is established based on the actual physiological parameter values ​​of each physiological activity indicator to ensure that the bottom boundary of the specific clause has a good summarizing ability for the constraint judgment of the actual physiological parameter values. This can effectively improve the understanding of the mechanism coupling between the actual physiological parameter values ​​under the premise of judging the normal physiological operation mechanism, and ensure the accuracy of the judgment of physiological safety anomalies.

[0098] It should be noted that if the physiological morphology of the fish to be anesthetized does not have any actual physiological parameter values ​​corresponding to inductive clauses with coverage violations exceeding the preset coverage violation, it indicates that the fish to be anesthetized is within the physiological safety range under the premise of following normal physiological mechanisms, meaning that the physiological characteristics of the fish to be anesthetized are maintained at a comprehensive normal level, and therefore there is no need to terminate the bionic electro-optic combined anesthesia procedure. If the fish to be anesthetized has at least one or more actual physiological parameter values ​​corresponding to inductive clauses with coverage violations exceeding the preset coverage violation, it indicates that the fish to be anesthetized, under the premise of following normal physiological mechanisms, has exceeded the physiological safety specifications, representing that there is an abnormal coupling interaction of certain physiological indicators in the fish to be anesthetized, causing the physiological characteristics of the fish to tend to an abnormal level. Therefore, it is necessary to control the triggering warning device to issue an audible and visual alarm to prompt the staff to interrupt the anesthesia procedure, thereby protecting the safety of the fish to be anesthetized. This method can monitor and determine abnormalities in the physiological indicators of fish by focusing on the constraints of normal physiological operating mechanisms, improve the safety and controllability coefficient of bionic electro-optic anesthesia for fish, and achieve intelligent protection of neuroanesthesia effects.

[0099] A second aspect of this invention provides a smart anesthesia device for fish based on biomimetic electro-optical combined neural inhibition, such as... Figure 1-6 As shown, the intelligent fish anesthesia device includes:

[0100] AI intelligent recognition module 11, which includes an underwater camera 12, an image processing unit 13 and a multi-sensor fusion array 14, is used to identify comprehensive recognition data including fish species, body size parameters, spatial location and movement trajectory;

[0101] The bionic electrical stimulation module 15 includes a waveform generator 16, a power amplifier 17, and a smart electrode array 18, which is used to simulate the natural discharge characteristics of electrical organs to generate electrical stimulation signals that are closer to physiological processes to perform bionic electrical stimulation anesthesia on fish.

[0102] The optical nerve inhibition module 19 includes a near-infrared laser 20, an optical fiber transmission unit 21, and an illumination control unit 22, which is responsible for releasing near-infrared light to act on the ion channels and membrane potential of fish neurons on the basis of electrical stimulation anesthesia, thereby prolonging nerve inhibition.

[0103] Safety protection module 23, which includes magnetic field cancellation device 24, biological monitoring unit 25 and adaptive conductive gel device 26, is responsible for protecting the safety of fish biomimetic electro-optic anesthesia operation;

[0104] The central control module 27 includes a main controller 28, a parameter adjustment panel 29, and a display screen 30, and is responsible for receiving data, automatically switching suppression modes, and manually adjusting working parameters.

[0105] It should be noted that the underwater camera is used to acquire real-time images of fish; the image processing unit is used to identify fish species, measure body size, and predict movement trajectories; and the multi-sensor fusion array is used to assist in positioning and environmental perception. The waveform generator is used to simulate the discharge characteristics of eel electroorgans to generate intermittent high-frequency cluster pulses; the power amplifier is used to amplify the pulse signal to an adjustable voltage of 50–800V; and the surface of the intelligent electrode array is coated with an adaptive conductive gel layer for contact with the fish and conduction of electrical pulses. The near-infrared laser is used to generate near-infrared light with wavelengths of 810–980nm; the fiber optic transmission unit is used to precisely guide the laser to the target fish; and the illumination control unit is used to control the illumination timing, intensity adjustment, and delay suppression, achieving coordinated operation with the electrical stimulation system. The magnetic field cancellation device is used to limit the diffusion range of the current in the water to protect surrounding marine life; the biological monitoring unit is used to monitor fish physiological indicators in real time to ensure safe operation; and the adaptive conductive gel device is used to prevent localized burns caused by poor electrode contact. The main controller receives data from the AI ​​recognition system and automatically switches the inhibition mode according to the fish species. It uses a short-time high-pressure pulse mode for surface fish and a low-frequency long-time stimulation mode for deep-sea fish. The parameter adjustment panel is used to manually adjust the operating parameters of each system. The display screen is used to display the device status and parameter information.

[0106] It should be noted that the image processing unit includes an image preprocessing module, a deep learning recognition engine, a feature extraction algorithm library, and a trajectory prediction calculation module. The image preprocessing module is used to denoise, enhance, and color correct underwater images. The deep learning recognition engine is based on a pre-trained convolutional neural network model and can recognize more than 500 common marine fish species with an accuracy rate of no less than 95%. The feature extraction algorithm library is used to extract the length, width, body shape features, and swimming posture parameters of fish. The trajectory prediction calculation module predicts the movement trajectory of fish within the next 0.5 to 2 seconds based on the Kalman filter algorithm. The waveform generator includes an eel bioelectric signal analysis unit, a pulse sequence generator, a frequency modulator, and a waveform shaper. The eel bioelectric signal analysis unit stores and analyzes the discharge characteristics data of eel electroorgans, including parameters such as discharge voltage, frequency, pulse width, and interval time. The pulse sequence generator generates biomimetic pulse sequences based on the analyzed data, simulating the intermittent high-frequency cluster discharge pattern of fish electroorgans. The frequency modulator adjusts the pulse frequency range from 0.1 to 100 Hz according to the target fish species and body size. The waveform shaper optimizes the pulse waveform to maximize the electrical stimulation effect and reduce tissue damage. The power amplifier includes a multi-stage amplification circuit, a voltage regulation module, a current limiter, and a power monitoring unit. The multi-stage amplification circuit amplifies the low-power signal output from the waveform generator stage by stage. The voltage regulation module automatically adjusts the output voltage according to the fish species, outputting 500-800V high-voltage short pulses for surface fish and 50-200V low-voltage long pulses for deep-sea fish. The current limiter prevents overcurrent and protects the fish and equipment. The power monitoring unit monitors the output power in real time and feeds it back to the control system.

[0107] It should be noted that the near-infrared laser includes a laser diode chip, a temperature control module, a beam shaping lens group, and a power adjustment circuit. The laser diode chip is made of InGaAs material and is used to generate near-infrared laser with a tunable wavelength of 810-980nm. The temperature control module is used to maintain the laser's operating temperature and ensure constant output power. The beam shaping lens group is used to shape the laser beam into a uniform circular spot. The power adjustment circuit is used to achieve continuous power adjustment of 1-50mW according to the fish species and body size.

[0108] The specific workflow of this device is as follows:

[0109] In the application embodiment of the fishing vessel sorting production line, this device works in conjunction with the automatic sorting system. Fish enter the device's detection area (0.5-10 meters) via a conveyor belt at a speed of 0.5 m / s. An underwater camera automatically captures images, sonar acquires location information, and the AI ​​intelligent recognition module identifies the species and measures the size of the fish to be anesthetized within 0.3-0.5 seconds, while simultaneously predicting its movement trajectory. Subsequently, the central control module automatically selects anesthesia parameters based on the fish database and expert algorithms. These parameters can be manually fine-tuned via the parameter adjustment panel: Surface fish: 500-800V high-voltage short pulse (1-3ms), 20-50Hz frequency, 5-15mW near-infrared light. Deep-sea fish: 50-200V low-voltage long pulse (10-50ms), 1-10Hz frequency, 20-50mW near-infrared light. Next comes the electro-optical combined anesthesia stage. The intelligent electrode array of the bionic electrical stimulation module contacts the fish and releases a bionic cluster pulse (1-5 seconds) of electrical stimulation to quickly anesthetize the fish. Within 0.1-2 seconds after the electrical stimulation, the near-infrared laser (810-980nm) of the optical nerve inhibition module irradiates the head or spinal nerve-dense areas of the fish through a fiber optic probe to inhibit the activity of neuronal ion channels. The illumination lasts for 1-10 minutes (automatically adjusted according to the fish's recovery), prolonging the anesthesia state, which is sufficient to complete subsequent sorting, weighing, and packaging operations. Compared with traditional manual operation, this anesthesia method significantly improves sorting efficiency, reduces fish stress response, and significantly improves the quality of fish products. The electrical stimulation and optical nerve inhibition work together to reduce stress damage.

[0110] Furthermore, in a preferred embodiment of the present invention, the multi-sensor fusion array 14 includes an ultrasonic ranging sensor 31, an underwater lidar 32, a water flow velocity sensor 33, a water quality detection sensor 34, and an inertial measurement unit 35.

[0111] It should be noted that the sonar positioning sensor is used to accurately determine the relative distance and azimuth angle between the fish and the device, with a ranging accuracy of ±2cm; the water flow velocity sensor is used to monitor the water flow direction and velocity to correct the trajectory prediction algorithm; the water temperature sensor, water pressure sensor, and conductivity sensor are used to monitor water environment parameters in real time, providing environmental reference data for the adaptive adjustment of electrical stimulation parameters. The AI ​​intelligent recognition module uses a multi-sensor data fusion algorithm to comprehensively process visual recognition, sonar positioning, and environmental perception information to achieve three-dimensional spatial positioning of the target fish, with a response time of no more than 0.5 seconds. It can maintain effective recognition in turbid water with visibility of less than 1 meter, and finally outputs comprehensive recognition data including fish species, body size parameters, spatial position, and movement trajectory, providing precise control basis for subsequent electrical stimulation and optical inhibition systems.

[0112] Furthermore, in a preferred embodiment of the present invention, the intelligent electrode array 18 includes a multi-electrode array substrate 36, an electrode spacing adjustment mechanism 37, and a contact detection sensor 38.

[0113] It should be noted that the multi-electrode array substrate includes 6-12 independently controllable electrode units. Each electrode unit is made of biocompatible stainless steel, and the electrode surface is coated with an adaptive conductive gel layer. This gel has good conductivity, biocompatibility, and adhesion, and can adapt to different fish surface shapes, preventing localized current concentration and burns caused by poor contact. The electrode spacing adjustment mechanism is used to automatically adjust the distance between electrodes according to the size of the fish, ensuring a uniform electric field distribution. The contact detection sensor is used to detect the contact state between the electrode and the fish and feed it back to the control system. The biomimetic electrical stimulation module generates electrical stimulation signals that more closely resemble physiological processes by simulating the natural discharge characteristics of fish electroorgans. This enables precise, rapid, and low-damage anesthesia for different species of marine fish. Compared with traditional fixed-frequency square wave electroconvulsive therapy devices, it significantly reduces muscle rigidity and stress damage, improves fish meat quality, and ensures the safety and effectiveness of current conduction through the adaptive characteristics of the intelligent electrode array.

[0114] Furthermore, in a preferred embodiment of the present invention, the optical fiber transmission unit 21 includes a multimode fiber bundle 39, an optical fiber coupler 40, an optical fiber splitter 41, and an adjustable optical fiber probe array 42.

[0115] It should be noted that the multimode fiber bundle uses quartz fiber material with a core diameter of 50-200μm for low-loss transmission of near-infrared laser light; the fiber coupler is used to efficiently couple the laser output light into the fiber, with a coupling efficiency of ≥85%; the fiber beam splitter is used to distribute a single laser beam to multiple output ports to achieve simultaneous multi-point irradiation; the adjustable fiber probe array includes 6-12 independently adjustable probes, each with 360° rotation and ±45° pitch adjustment capabilities, used for precise directional irradiation of the nerve-dense areas of the target fish. The optical neural inhibition module acts on the ion channels and membrane potential of fish neurons through the photobiological modulation effect of near-infrared light. Based on rapid anesthesia by electrical stimulation, it prolongs the neural inhibition time by 3-10 minutes, significantly reducing muscle rigidity and stress response. At the same time, through multi-point precise irradiation and real-time monitoring feedback, it ensures the uniformity and safety of the inhibition effect, ultimately achieving a mild, long-lasting, and controllable neural inhibition effect on marine fish.

[0116] Furthermore, in a preferred embodiment of the present invention, the adaptive conductive gel device 26 includes a gel dispensing device 43, a contact pressure sensor 44, and a gel composition adjustment unit 45.

[0117] It should be noted that the gel dispensing device includes a gel storage tank, a micro-pump, and a multi-way dispensing valve for precisely dispensing the appropriate amount of gel to each electrode. The device stores a biocompatible conductive gel composed of sodium alginate, sodium chloride, glycerin, and an antibacterial agent, which has good conductivity and biocompatibility. A contact pressure sensor is embedded in the surface of each electrode to detect the contact pressure between the electrode and the fish body to determine the contact status. The gel composition adjustment unit automatically adjusts the viscosity, conductivity, and buffering performance of the gel according to the skin characteristics of different fish and the contact pressure data.

[0118] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart anesthesia method for fish based on biomimetic electro-optical combined neural inhibition, characterized in that, Includes the following steps: S102: Capture and photograph fish using an underwater camera to obtain real-time image information of the fish to be anesthetized, and simultaneously acquire underwater point cloud, water environment data and underwater positioning information of the target fish area through a multi-sensor fusion array; S104: Combine water environment data and underwater positioning information to project the underwater point cloud onto the semantically segmented real-shot image information for mask rendering, generate an underwater enhanced motion simulation field, and identify the characteristics and movement of the fish to be anesthetized expressed by the underwater enhanced motion simulation field based on the image processing unit, and obtain the species information, body size and movement trajectory of the fish to be anesthetized. S106: Based on empirical cases of fish anesthesia, a discrete value mapping library of preset fish-anesthesia parameters is constructed by discrete combination. Simultaneously, anesthesia constraints to be tested are created based on species information, body size, and movement trajectory and anesthesia parameters. Different anesthesia constraints to be tested are tested through the discrete value mapping library, and the optimal anesthesia parameters are output. S108: Controls the intelligent electrode array to contact the fish body according to the optimal anesthesia parameters, releases biomimetic cluster pulses to quickly anesthetize, controls the near-infrared laser to irradiate the nerve-dense area of ​​the fish to be anesthetized, and obtains several actual physiological parameter values ​​of the fish to be anesthetized in different physiological activity indicators in real time through the safety protection module. S110: Construct specific clause bottom boundaries that do not touch the physiological safety threshold based on the normal physiological operation mechanism and physiological safety thresholds of different physiological activity indicators of the fish to be anesthetized. Use the specific clause bottom boundaries to perform anomaly judgment analysis on the actual physiological parameter values ​​to respond to the control and early warning device and anesthesia procedure.

2. The intelligent anesthesia method for fish based on biomimetic electro-optical combined neural inhibition according to claim 1, characterized in that, Step S104 specifically includes the following steps: The BiSe semantic segmentation network is invoked to perform different pixel-level classification on each pixel source point contained in the real-shot image information, generate the category confidence distribution probability of each pixel source point, and construct the semantic segmentation map of the real-shot image information based on the category confidence distribution probability. The perception strategy and calibration logic of the multi-sensor fusion array are obtained. The multi-sensor calibration matrix is ​​established by combining the underwater positioning information, perception strategy and calibration criteria. Based on the water environment data, the multi-sensor calibration matrix is ​​used to transform and project the intrinsic and extrinsic parameters of each point cloud seed in the underwater point cloud onto the image plane of the real image information to generate a projection joint dependency map of point cloud seed-pixel source point. Using the source pixel corresponding to each point cloud seed in the projection joint dependency graph as the source object, the semantic category confidence distribution of the source object is queried from the semantic segmentation graph to obtain the one-hot vector mask of each source object. Based on the projection joint dependency graph, the one-hot vector mask is colored and applied to the origin feature of the corresponding point cloud seed of each source object for highlighting rendering construction, so as to obtain the underwater enhanced motion simulation field carrying the anesthetized fish in the target fish area. The deep learning inference engine, feature extraction operator library, and trajectory prediction architecture of the image processing unit are acquired. Based on the deep learning inference engine, dynamic voxel feature processing is performed on the underwater enhanced motion simulation field to obtain the BEV feature trajectory map of the fish to be anesthetized in a time-series dynamic migration. Based on the feature extraction operator library and trajectory prediction architecture, the BEV feature trajectory map is identified, and the species information, body size, and movement trajectory of the fish to be anesthetized are output.

3. The intelligent anesthesia method for fish based on biomimetic electro-optical combined neural inhibition according to claim 2, characterized in that, The image processing unit acquires a deep learning inference engine, a feature extraction operator library, and a trajectory prediction architecture. Based on the deep learning inference engine, it performs dynamic voxel feature processing on the underwater enhanced motion simulation field to obtain a BEV feature trajectory map of the fish to be anesthetized undergoing dynamic migration in time. Based on the feature extraction operator library and trajectory prediction architecture, it identifies the BEV feature trajectory map and outputs the species information, body size, and movement trajectory of the fish to be anesthetized. Specifically, this includes the following steps: The deep learning inference engine, feature extraction operator library and trajectory prediction architecture of the image processing unit are obtained. The sparse voxel algorithm is introduced to weave the network system of feature recognition and perception of the deep learning inference engine, and a multi-dimensional voxel perception network is constructed. The multidimensional voxel sensing network is used to voxelize the point cloud region where the fish to be anesthetized is located in the target fish region in an underwater enhanced motion simulation field, and to obtain the global voxel feature map of the fish to be anesthetized affected by the environmental changes of the target fish region. By sampling the starting point cloud region of the fish to be anesthetized in an underwater enhanced motion simulation field, multiple key temporal transition sites of the starting point cloud were obtained. The key temporal transition sites were interpolated from the global voxel feature map to obtain the voxel features of each key temporal transition site, which are defined as temporal transition voxel features. A BEV mesh for underwater enhanced motion simulation field is constructed. The temporal transition voxel features are fused and stitched with the point cloud features of each key temporal transition point according to the positional rules of the BEV mesh to form a BEV feature trajectory map of the dynamic migration of the fish to be anesthetized in the target fish area in time. A lightweight recognition network is established based on the computational logic of the feature extraction operator library and trajectory prediction architecture. The lightweight recognition network performs comprehensive recognition of the fish to be anesthetized on the BEV feature trajectory map by distinguishing features, regressing positions, and predicting directions. It outputs feature classification scores and trajectory bounding box regression chains. Based on the feature classification scores and trajectory bounding box regression chains, the species information, body size, and movement trajectory of the fish to be anesthetized are determined.

4. The intelligent anesthesia method for fish based on biomimetic electro-optical combined neural inhibition according to claim 1, characterized in that, Step S106 specifically includes the following steps: To obtain the predetermined anesthesia requirements of fish, several optimal anesthesia cases simulating the natural discharge characteristics of different fish were obtained through big data network retrieval. By extracting the range of anesthesia parameters required for different preset fish species based on species information, body size, and movement trajectory from several optimal anesthesia cases, and discretizing the parameter values ​​for anesthesia of each preset fish species based on species information, body size, and movement trajectory according to the given anesthesia requirements, a discrete value mapping library of preset fish species and anesthesia parameters is constructed; among which, anesthesia parameters include voltage, frequency, and light intensity. Anesthesia constraint prefixes are created based on species information, body size, and movement trajectory. Anesthesia parameters such as voltage, frequency, and light intensity are defined as anesthesia variable matrices and added to the anesthesia constraint prefixes to generate an anesthesia constraint test queue for fish to be anesthetized. Each anesthesia constraint in the anesthesia constraint test queue is loaded in descending order and marked as an anesthesia constraint to be tested. A set of discrete values ​​is randomly extracted from the discrete value mapping library of fish-anesthesia parameters and injected into the anesthesia variables of the anesthesia constraint to be tested. The test processing is performed on the anesthesia constraint to be tested, and the constraint individuals corresponding to the set of discrete values ​​are obtained and marked as empirical anesthesia constraints. Construct a candidate empty stack, introduce an expert hash algorithm to calculate the hash misalignment function of the anesthesia constraint to be tested relative to the empirical anesthesia constraint. If the hash misalignment function is greater than the preset hash misalignment function, then remove the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested; if the hash misalignment function is greater than the preset hash misalignment function, then transfer the set of discrete values ​​from the anesthesia variables of the anesthesia constraint to be tested and record them to the candidate empty stack. Repeat the above steps of random injection and verification of discrete values ​​to verify and analyze the remaining anesthesia constraints to be tested in the anesthesia constraint verification queue. Finally, only the candidate discrete values ​​corresponding to the lowest hash misalignment function are extracted as the optimal anesthesia parameters output.

5. The intelligent anesthesia method for fish based on biomimetic electro-optical combined neural inhibition according to claim 1, characterized in that, Step S110 specifically includes the following steps: Multiple physiological activity indicators of fish to be anesthetized were extracted and monitored to assess the anesthesia requirements based on the established anesthesia needs, and the physiological safety threshold of each physiological activity indicator was obtained; wherein, the physiological activity indicators include heart rate, respiratory rate, body temperature change and cortisol level; Based on big data networks, the normal physiological operation mechanism of the fish to be anesthetized in the target fish area is obtained, and several past physiological parameter values ​​of various physiological activity indicators are obtained under the premise that they are in line with the normal physiological operation mechanism and do not exceed the physiological safety threshold. A pattern matching algorithm is introduced, and based on the normal physiological operating mechanism and physiological safety threshold, several past physiological parameter values ​​are deduced in reverse in the pattern matching algorithm to construct a specific clause bottom boundary that does not touch the physiological safety threshold. Based on several actual physiological parameter values ​​of each physiological activity indicator, an actual physiological clause exploration domain is established. The bottom boundary of the specific clause is used as the search strategy criterion for the normal physiological operation mechanism to perform a global traversal search on each inductive clause in the actual physiological clause exploration domain, and the coverage violation degree of each inductive clause is obtained. If the fish to be anesthetized does not have actual physiological parameter values ​​corresponding to inductive clauses with coverage violation greater than the preset coverage violation, then the biomimetic electro-optical combined anesthesia procedure will continue. If at least one or more inductive clauses with a coverage violation degree greater than the preset coverage violation degree have corresponding actual physiological parameter values, the early warning device will immediately trigger an audible and visual alarm and automatically interrupt the anesthesia procedure.

6. A smart anesthesia device for fish based on biomimetic electro-optical combined neural inhibition, characterized in that, The intelligent anesthesia device for fish includes: The AI ​​intelligent recognition module includes an underwater camera, an image processing unit, and a multi-sensor fusion array, used to identify comprehensive recognition data including fish species, body size parameters, spatial location, and movement trajectory. A biomimetic electrical stimulation module, comprising a waveform generator, a power amplifier, and a smart electrode array, is used to simulate the natural discharge characteristics of electrical organs to generate electrical stimulation signals that more closely resemble physiological processes for biomimetic electrical stimulation anesthesia of fish. An optical neural inhibition module, comprising a near-infrared laser, an optical fiber transmission unit, and an illumination control unit, is responsible for releasing near-infrared light on the basis of electrical stimulation anesthesia to act on the ion channels and membrane potential of fish neurons, thereby prolonging neural inhibition; The safety protection module includes a magnetic field cancellation device, a biological monitoring unit, and an adaptive conductive gel device, which are responsible for protecting the operational safety of fish biomimetic electro-optic anesthesia. The central control module includes a main controller, a parameter adjustment panel, and a display screen, and is responsible for receiving data, automatically switching suppression modes, and manually adjusting operating parameters.

7. The intelligent anesthesia device for fish based on biomimetic electro-optic combined neural inhibition according to claim 6, characterized in that: The multi-sensor fusion array includes an ultrasonic ranging sensor, an underwater lidar, a water flow velocity sensor, a water quality detection sensor, and an inertial measurement unit.

8. The intelligent anesthesia device for fish based on biomimetic electro-optic combined neural inhibition according to claim 6, characterized in that: The intelligent electrode array includes a multi-electrode array substrate, an electrode spacing adjustment mechanism, and a contact detection sensor.

9. The intelligent anesthesia device for fish based on biomimetic electro-optic combined neural inhibition according to claim 6, characterized in that: The optical fiber transmission unit includes a multimode fiber bundle, an optical fiber coupler, an optical fiber splitter, and an adjustable optical fiber probe array.

10. The intelligent fish anesthesia device based on biomimetic electro-optic combined neural inhibition according to claim 6, characterized in that: The adaptive conductive gel device includes a gel dispensing device, a contact pressure sensor, and a gel composition adjustment unit.

Images