A method and system for intelligent grading and separate-pond culture of seahorses based on appearance characteristics

Abstract

This invention relates to the field of intelligent aquaculture and automated processing technology, specifically to a method and system for intelligent grading and pond separation of seahorses based on appearance characteristics. The method includes: S1, collecting sparse visual data and multi-source physiological data of seahorses; S2, generating a three-dimensional digital twin model based on the sparse visual data; calculating growth status grading indicators based on the three-dimensional digital twin model; S3, calculating a physiological stress index based on the multi-source physiological data; S4, comparing the real-time calculated physiological stress index with preset safe stress thresholds and critical stress thresholds to generate a grading adaptive correction strategy; S5, controlling a flexible robotic arm to perform harvesting and pond separation operations according to the grading adaptive correction strategy. This invention solves the technical problem of traditional physical intervention causing target stress reactions, making it impossible to assess the true quality of the seahorses and potentially causing damage, by simultaneously collecting the appearance and physiological data of seahorses in a non-contact manner, thus ensuring the authenticity and validity of the original data.

Description

Technical Field

[0001] This invention relates to the field of intelligent aquaculture and automated processing technology, specifically to a method and system for intelligent grading and pond-separation of seahorses based on their appearance characteristics. Background Technology

[0002] In the current high-value live aquatic product farming industry, such as seahorse farming, the grading of the product's growth status, such as separating them into different ponds based on size, is a core link that determines its farming efficiency and survival rate. To ensure the accuracy of grading and subsequent packaging operations, a detailed assessment of the appearance characteristics of live seahorses is required. Traditional harvesting and grading processes rely heavily on direct physical intervention, which can easily trigger stress responses in seahorses. These stress responses can not only cause injury to the seahorses but also alter their body color, posture, and other key quality characteristics, making it impossible to accurately assess their true grade in their natural state. This seriously affects the quality of automated pond separation operations and subsequent farming efficiency.

[0003] In existing technologies, fishing and pond separation operations are usually one-way physical operations, lacking real-time perception and feedback on the physiological state of the target organisms. On the one hand, in order to accurately assess appearance, it is necessary to obtain high-precision features at close range and from multiple angles; on the other hand, any close observation or physical contact will inevitably disturb the seahorse, causing it to enter a stress state, which in turn leads to the distortion of assessment data.

[0004] Existing technologies have failed to effectively resolve the conflict between high-precision feature recognition and maintaining the natural state of the target, resulting in insufficient objectivity and consistency of grading results, and posing a risk of damaging the target during the harvesting process. Therefore, how to provide a technical solution that enables non-destructive, precise, and intelligent grading and pooling of live seahorses, completes high-fidelity appearance assessment without physical intervention, and guides adaptive flexible harvesting and pooling operations by real-time monitoring of their physiological state is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention discloses a method and system for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics. Specifically, the technical solution of this invention is as follows:

[0006] A method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics includes the following steps:

[0007] S1. Collect sparse visual data and multi-source physiological data of the hippocampus;

[0008] S2. Generate a 3D digital twin model based on sparse visual data; calculate growth state grading index based on the 3D digital twin model;

[0009] S3. Calculate the physiological stress index based on multi-source physiological data;

[0010] S4. Compare the real-time calculated physiological stress index with the preset safe stress threshold and critical stress threshold to generate a graded adaptive correction strategy.

[0011] S5. Based on the hierarchical adaptive correction strategy, control the flexible robotic arm to perform fishing and pool separation operations.

[0012] Preferably, S1 specifically includes:

[0013] A sparse visual dataset is constructed by acquiring low-resolution two-dimensional images of the target from multiple preset azimuth angles using a long-range visual sensor.

[0014] Multi-source physiological data is collected using non-contact visual sensors; these data include gill movement frequency, limb movement frequency and amplitude, and instantaneous body color change rate.

[0015] Preferably, S2 specifically includes:

[0016] A sparse visual dataset is input into a pre-trained deep generative model to generate a 3D digital twin model.

[0017] Extract a set of growth characteristic parameters from a 3D digital twin model; among them, the growth characteristic parameters include the estimated values ​​of body length, height, weight, and body proportions.

[0018] The growth characteristic parameter set is input into a preset grading model, and weighted calculations are performed to generate a growth status grading index.

[0019] Preferably, S3 specifically includes:

[0020] Normalize the multi-source physiological data;

[0021] Based on the preset sensitivity weighting coefficients, the normalized multi-source physiological data are weighted and calculated to generate a physiological stress index.

[0022] Preferably, S4 specifically includes:

[0023] When the physiological stress index is not higher than the safe stress threshold, the stress state level is determined to be the safe zone, and an execution strategy to maintain the current optimal path is generated.

[0024] When the physiological stress index is higher than the safe stress threshold but not higher than the critical stress threshold, the stress state level is determined as the warning zone, and a correction strategy to dynamically reduce the movement speed of the robotic arm is generated.

[0025] When the physiological stress index is higher than the critical stress threshold, the stress state level is determined to be a danger zone, and an avoidance strategy is generated to terminate the task and retreat to a safe distance.

[0026] The preferred correction strategy for the warning zone is as follows:

[0027] The physiological stress index, safe stress threshold, and critical stress threshold are input into a preset speed correction function to calculate the corrected robotic arm movement speed. The corrected robotic arm movement speed decreases smoothly and non-linearly as the physiological stress index increases.

[0028] Preferably, after the danger zone avoidance strategy is implemented, it also includes:

[0029] Continuously monitor the physiological stress index, and resume fishing and pond separation operations when the physiological stress index falls below the safe stress threshold.

[0030] A smart grading and pond-separating seahorse culture system based on appearance characteristics includes:

[0031] The data acquisition module is used to collect sparse visual data and multi-source physiological data from the hippocampus.

[0032] The modeling and evaluation module is used to generate a three-dimensional digital twin model based on sparse visual data and to calculate the growth status grading index based on the three-dimensional digital twin model.

[0033] The status monitoring module is used to calculate the physiological stress index based on multi-source physiological data;

[0034] The decision control module is used to compare the physiological stress index with the preset safe stress threshold and critical stress threshold, and generate a graded adaptive correction strategy.

[0035] The job execution module receives the hierarchical adaptive correction strategy and controls the flexible robotic arm to perform fishing and pool separation operations.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. This invention synchronously collects the appearance and physiological data of the seahorse in a non-contact manner, which solves the technical problem that traditional physical intervention causes stress response in the target, making it impossible to assess its true quality and potentially causing damage, thus ensuring the authenticity and validity of the original data.

[0038] 2. This invention generates a high-fidelity three-dimensional digital twin model based on sparse visual data, and deconstructs the abstract concept of growth into calculable objective indicators. It transforms the evaluation of hippocampal growth status from relying on subjective experience to an automated and standardized quantitative process, which significantly improves the objectivity, consistency and precision of the grading results.

[0039] 3. This invention constructs a closed-loop adaptive control system dominated by biological signals, which dynamically modifies the fishing strategy by monitoring the physiological stress index in real time, transforming the traditional one-way fishing behavior into a two-way adaptive interaction, minimizing the intensity of physical intervention, and achieving safe and non-destructive acquisition of live targets;

[0040] 4. This invention deeply integrates the biological characteristics and physiological state of seahorses. Through non-interventional modeling and evaluation and biosignal-driven adaptive harvesting, it effectively solves the technical contradiction between high-precision characteristic recognition and maintaining the natural state of the target, and greatly improves the automated pooling efficiency and aquaculture benefits of precious live aquatic products. Attached Figure Description

[0041] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0042] Figure 1 This is a flowchart of the method of the present invention;

[0043] Figure 2 This is a flowchart of the system of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0045] Example 1:

[0046] Please see Figure 1 A method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics includes the following steps:

[0047] S1. Collect sparse visual data and multi-source physiological data of the hippocampus;

[0048] S2. Generate a 3D digital twin model based on sparse visual data; calculate growth state grading index based on the 3D digital twin model;

[0049] S3. Calculate the physiological stress index based on multi-source physiological data;

[0050] S4. Compare the real-time calculated physiological stress index with the preset safe stress threshold and critical stress threshold to generate a graded adaptive correction strategy.

[0051] S5. Based on the hierarchical adaptive correction strategy, control the flexible robotic arm to perform fishing and pool separation operations.

[0052] This embodiment provides a method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics. This method aims to solve the technical problem that stress response caused by physical intervention during the grading and harvesting process of live seahorses makes it impossible to accurately assess their true quality and may cause damage. This method achieves non-destructive, precise, and separate-pond rearing of seahorses by constructing a complete technical process from non-interventional data acquisition, high-fidelity modeling and evaluation to biological signal-driven closed-loop adaptive control.

[0053] The specific process of this method includes the following steps:

[0054] S1. Acquisition of sparse visual data and multi-source physiological data of seahorses: To obtain the appearance and physiological information of seahorses without disturbing their natural state, this step involves non-contact data acquisition. Sparse visual data refers to a dataset consisting of low-resolution two-dimensional images of the target captured rapidly from multiple preset azimuth angles under natural light conditions in the breeding environment using a long-distance visual sensor. Its purpose is to provide basic input for subsequent three-dimensional modeling. Multi-source physiological data refers to dynamic data that can reflect the physiological stress state of seahorses in multiple dimensions. Its purpose is to quantify the stress level of seahorses and provide a basis for decision-making in subsequent adaptive harvesting.

[0055] S2. Generate a 3D digital twin model based on sparse visual data; calculate growth status grading index based on the 3D digital twin model: The core of this step is to accurately quantify and evaluate the growth status of the hippocampus. Specifically, a pre-trained deep generative model is used to process the sparse visual dataset collected in step S1, rendering a high-precision, full-view 3D digital twin model of the hippocampus in a zero-stress state in real time. This model is a digital reproduction of the hippocampus's true 3D morphology, providing high-fidelity data objects for subsequent morphological analysis. Based on this model, the system automatically extracts and quantifies a series of preset growth characteristic parameters, and calculates the final growth status grading index through a non-linear evaluation model. This score is a single, quantifiable, comprehensive index used to objectively assess the hippocampus's growth stage and specifications, such as size.

[0056] S3. Calculate the physiological stress index based on multi-source physiological data: In order to perceive the stress state of the hippocampus in real time, this step inputs the multi-source physiological parameters collected in S1 into a comprehensive evaluation model and dynamically calculates a physiological stress index that reflects the current stress level of the target. This physiological stress index is a quantitative indicator that transforms the invisible internal stress level of the hippocampus into a measurable dynamic signal, which serves as the core feedback variable of the subsequent closed-loop control system.

[0057] S4. Compare the real-time calculated physiological stress index with the preset safe stress threshold and critical stress threshold to generate a graded adaptive correction strategy: This step is the decision-making center connecting perception and action; the system continuously compares the physiological stress index calculated in real-time in step S3 with two preset thresholds to determine the stress level of the hippocampus; the safe stress threshold refers to the index value corresponding to the inflection point where the slope of the hippocampus stress response curve first shows a significant increase, representing the critical point at which the hippocampus begins to show initial tension; the critical stress threshold refers to the index value corresponding to when the hippocampus begins to exhibit violent escape or rigidity behaviors, representing that the hippocampus is already in a state of high tension; based on the comparison results, the system will generate a corresponding graded adaptive correction strategy, which aims to dynamically adjust the execution mode of harvesting and pool separation operations based on the real-time feedback of the hippocampus;

[0058] S5. Based on the hierarchical adaptive correction strategy, control the flexible robotic arm to perform fishing and pool separation operations: As the final execution link of the technical process, this step transforms the strategy generated in step S4 into specific physical actions; the flexible robotic arm is an actuator capable of gentle, biomimetic operations, used to complete fishing without damaging the seahorse; it will precisely control its own trajectory and speed according to the received strategy, such as normal execution, slowing down, or stopping and retreating, thereby achieving adaptive interaction with the seahorse;

[0059] Through the above steps, this invention constructs an intelligent decision-making and control closed loop that deeply integrates the biological characteristics and physiological state of seahorses. Through non-interventional modeling and evaluation, it achieves objective and accurate grading of their growth status. During the harvesting process, by monitoring their physiological stress signals in real time, it transforms the traditional one-way harvesting behavior into a two-way adaptive interaction dominated by biological signals, solving the technical contradiction between high-precision feature recognition and maintaining the natural state of the target. This enables non-destructive grading and efficient acquisition of live targets, improving the quality of automated processing and aquaculture efficiency of precious aquatic products.

[0060] Example 2:

[0061] S1 specifically includes:

[0062] A sparse visual dataset is constructed by acquiring low-resolution two-dimensional images of the target from multiple preset azimuth angles using a long-range visual sensor.

[0063] Multi-source physiological data is collected using non-contact visual sensors; these data include gill movement frequency, limb movement frequency and amplitude, and instantaneous body color change rate.

[0064] This embodiment is based on Embodiment 1, focusing on the data acquisition method in step S1;

[0065] In a preferred embodiment, sparse visual data is acquired by deploying a long-range visual sensor in the aquaculture environment. The sensor is configured to rapidly capture images of the target from multiple preset azimuth angles to obtain low-resolution two-dimensional images. The long-range design aims to minimize visual interference from the sensor to the seahorse. The multiple azimuth angles ensure that the acquired data, although sparse, still covers key morphological information, providing sufficient geometric constraints for 3D reconstruction. The low resolution ensures the acquisition speed, enabling instantaneous snapshots and avoiding image blurring caused by seahorse movement. These images are then integrated to construct a sparse visual dataset.

[0066] Meanwhile, the acquisition of multi-source physiological data is accomplished through a non-contact visual sensor. This sensor continuously monitors the seahorse via video throughout the entire process of the robotic arm preparing to approach and capture it. By analyzing subtle changes in video frames, the system can extract core parameters reflecting the seahorse's physiological stress state. In this embodiment, the multi-source physiological data specifically includes: gill movement frequency, limb movement frequency and amplitude, and instantaneous body color change rate due to stress. Among them, gill movement frequency is obtained by analyzing the periodic changes in minute colors on the body surface; limb movement frequency and amplitude are calculated using optical flow or target tracking algorithms; and instantaneous body color change rate is obtained by analyzing the rate of change in chromaticity values ​​in specific areas of the body surface.

[0067] Through the specific acquisition methods described above, this invention can simultaneously and in parallel acquire two heterogeneous data sets for appearance modeling and physiological state monitoring without generating any physical contact or stress interference. This non-invasive dual-data-stream acquisition scheme provides high-quality, distortion-free raw data input for subsequent high-fidelity growth state assessment and high-sensitivity stress closed-loop control, ensuring the effectiveness and reliability of the entire technical solution.

[0068] To address challenges such as water turbidity, light variations, and occlusion in real-world aquaculture environments, the data acquisition module of this system can further integrate image enhancement and denoising algorithms. For example, a dark channel prior algorithm can be used for underwater image restoration. Simultaneously, by fusing data collected from multiple sensors at different locations and utilizing a target tracking algorithm to take consecutive snapshots of the same target within a short period, low-quality data due to occlusion or poor posture can be effectively eliminated, ensuring the quality of data input into subsequent models and thus enhancing the robustness of the entire system in complex real-world scenarios.

[0069] Example 3:

[0070] S2 specifically includes:

[0071] A sparse visual dataset is input into a pre-trained deep generative model to generate a 3D digital twin model.

[0072] Extract a set of growth characteristic parameters from a 3D digital twin model; among them, the growth characteristic parameters include the estimated values ​​of body length, height, weight, and body proportions.

[0073] The growth characteristic parameter set is input into a preset grading model, and weighted calculations are performed to generate a growth status grading index.

[0074] This embodiment is based on Embodiment 1, and focuses on the growth status evaluation process in step S2;

[0075] To further clarify, the generation process of the 3D digital twin model is as follows: The sparse visual dataset collected in the aforementioned steps is input into a pre-trained deep generative model to generate a high-fidelity 3D digital twin model; this process can be described by the following formula:

[0076]

[0077] in, : Represents the final generated high-fidelity 3D digital twin model, which serves as the data object for subsequent growth status evaluation. The data type is 3D point cloud or mesh model.

[0078] : Represents a pre-trained deep generative model function, whose function is to predict and render a complete 3D model based on sparse 2D input; in this embodiment, the model can adopt a neural radiation field or generative adversarial network architecture known in the art.

[0079] : Represents the sparse two-dimensional image dataset that serves as the input to the model, which was acquired in the preceding steps;

[0080] : Represents the internal network weight parameters that the generative model has converged and solidified; these parameters were determined through offline training on a professional dataset containing thousands of sets of omnidirectional high-precision 3D scan data and corresponding sparse angular 2D images, and are the training results of machine learning.

[0081] This method utilizes the powerful prior knowledge of deep learning models to solve the technical contradiction of traditional 3D reconstruction technology, which requires multi-angle and long-term scanning and causes strong stress interference to living organisms, thus achieving observation without intervention.

[0082] The system automatically extracts a set of growth characteristic parameters from the generated 3D digital twin model. In this embodiment, the set of growth characteristic parameters specifically includes: body length, body height, estimated weight, and body proportion. Body length is obtained by calculating the maximum distance along the main axis between the head apex and the tail tip on the 3D digital twin model. Body height is obtained by calculating the maximum width of the torso portion of the model, such as from the back to the abdomen, perpendicular to the main axis. Estimated weight is obtained by calculating the mesh volume of the 3D model and multiplying it by a preset species density coefficient. Body proportion is obtained by calculating the ratio between the body length and body height.

[0083] Based on the above parameter set, to ensure the stability and repeatability of the evaluation results, the extracted growth feature parameter set is rigorously standardized before being input into the nonlinear evaluation model; this standardization aims to eliminate the original values ​​of different growth features. The impact of differences in units, dimensions, or numerical ranges; specifically, the system will, based on historical statistical data for each feature, use Z-score standardization or Min-Max normalization methods to transform the original quantized values... Convert to a unified, dimensionless standardized value ;

[0084] Standardized values Input a pre-defined nonlinear evaluation model, perform weighted calculations, and generate the final growth state classification index; this calculation process can be defined by the following growth state classification index function:

[0085]

[0086] in, : Represents the final growth status classification index, which is a dimensionless scalar and its function is to provide a direct basis for intelligent classification.

[0087] : Represents the total number of growth characteristic parameters;

[0088] : The dimensionless weight coefficients representing the i-th growth characteristic, with a total of 1. These weight coefficients are not subjectively set, but are determined by inviting several senior seahorse breeding experts and traders to conduct double-blind scoring on the standard sample set, and then using objective weighting methods such as the analytic hierarchy process or the entropy weight method to conduct statistical analysis on the expert scoring data, so as to ensure the objectivity of the weights and the industry's recognition.

[0089] : Represents the standardized value of the i-th growth feature extracted from model M, which is a uniform dimensionless value;

[0090] : Represents a standardized value The function that performs nonlinear mappings has the specific form of the standard Sigmoid function: This function can smoothly map standardized feature values ​​to... Intervals are used to achieve optimal differentiation.

[0091] By employing this method of evaluation based on deep learning modeling and multi-attribute decision theory, the present invention transforms the evaluation of hippocampal growth status from a traditional approach that relies on subjective experience into a fully automated, standardized, and highly accurate quantitative process. The deep generative model ensures that the model on which the evaluation is based is a true and natural reflection of the hippocampus, while the growth status grading index function deconstructs the abstract concept of growth into a computable objective index, thereby improving the objectivity, consistency, and precision of the grading results.

[0092] Example 4:

[0093] S3 specifically includes:

[0094] Normalize the multi-source physiological data;

[0095] Based on the preset sensitivity weighting coefficients, the normalized multi-source physiological data are weighted and calculated to generate a physiological stress index.

[0096] This embodiment is based on Embodiment 1, and focuses on the calculation method of the physiological stress index in step S3;

[0097] In a preferred embodiment, to effectively fuse physiological data from different sources and with different dimensions, it is necessary to normalize the multi-source physiological data. Specifically, for the collected raw data such as gill movement frequency, escape behavior frequency and amplitude, and body color change rate, the system linearly normalizes them according to their dynamic range in historical statistical data, mapping them to the dimensionless interval [0,1]. For example, the normalized value of the gill movement frequency change rate... Normalized values ​​of the frequency and amplitude of escape behavior and the normalized value of the rate of change in body color. All of these were obtained in this way; among them, the escape behavior indicator It is a comprehensive value obtained by weighting the normalized escape frequency and amplitude to fully reflect the intensity of its stress response behavior;

[0098] After normalization, the system will perform weighted calculations on the normalized multi-source physiological data according to preset sensitivity weight coefficients to generate the final physiological stress index. This calculation process is implemented through the following physiological stress index function:

[0099]

[0100] in, : Represents the final dimensionless physiological stress index; the higher the value, the stronger the stress level in the hippocampus.

[0101] : These represent the gill movement frequency change rate, escape behavior index, and body color change rate after the aforementioned normalization treatment, respectively. Their values ​​are all calculated from real-time monitoring data.

[0102] These are the dimensionless sensitivity weighting coefficients for the three physiological parameters mentioned above, and the sum of the three is 1. These weighting coefficients are not set equally, but are obtained through controlled experiments. In the experiment, by gradually applying external stimuli of different intensities to the sample and simultaneously monitoring the degree of change of each physiological indicator, the earliest and most sensitive stress response indicator will be assigned the highest weighting coefficient to ensure that the system has the strongest early warning capability.

[0103] By normalization and weighted calculation, this invention creates a quantitative indicator that can reflect the overall stress level within the hippocampus in real time and with high sensitivity. This method not only integrates information from multiple physiological dimensions, but more importantly, by introducing experimentally calibrated sensitivity weights, it enables the index to focus on the most critical early warning signals, thereby providing a high-quality, high-signal-to-noise-ratio core feedback signal for the subsequent closed-loop adaptive control system, ensuring that the fishing behavior can be precisely dominated by the physiological state of the seahorse.

[0104] It should be noted that the linear weighted model used in this embodiment is a simplified calculation of complex physiological stress responses. This model has good characterization ability and computational efficiency within the monotonically increasing stress level range. For specific breeds that may exhibit non-monotonic responses such as rigidity, quadratic terms can be introduced into the model or combined with other state judgment logic to improve the model's adaptability.

[0105] Example 5:

[0106] S4 specifically includes:

[0107] When the physiological stress index is not higher than the safe stress threshold, the stress state level is determined to be the safe zone, and an execution strategy to maintain the current optimal path is generated.

[0108] When the physiological stress index is higher than the safe stress threshold but not higher than the critical stress threshold, the stress state level is determined as the warning zone, and a correction strategy to dynamically reduce the movement speed of the robotic arm is generated.

[0109] When the physiological stress index is higher than the critical stress threshold, the stress state level is determined to be a danger zone, and an avoidance strategy is generated to terminate the task and retreat to a safe distance.

[0110] The specific correction strategy for the warning zone is as follows:

[0111] The physiological stress index, safe stress threshold, and critical stress threshold are input into a preset speed correction function to calculate the corrected robotic arm movement speed. The corrected robotic arm movement speed decreases smoothly and non-linearly as the physiological stress index increases.

[0112] After the danger zone avoidance strategy is implemented, it also includes:

[0113] Continuously monitor the physiological stress index, and resume fishing and pond separation operations when the physiological stress index falls below the safe stress threshold.

[0114] This embodiment, based on embodiment 1, focuses on the graded adaptive correction strategy in step S4. This is a complete data-driven decision-making and control process that incorporates the physiological stress index. This translates into precise control commands for the robotic arm;

[0115] This strategy is based on the physiological stress index. With safety stress threshold Critical stress threshold The comparison results show that the specific values ​​of these two thresholds are not fixed, but need to be experimentally calibrated for specific hippocampal species before use. The calibration process includes: in a controlled environment, a robotic arm gradually approaches the sample hippocampus at different speeds while continuously monitoring its physiological stress index. The changes were observed. Through analysis of multiple sets of experimental data, it was statistically determined at what index level the hippocampus generally began to exhibit initial signs of stress, which was then identified as... At what index level do violent escape or freeze behavior begin to appear? This allows for the preset threshold; the stress state of the hippocampus is divided into three levels, and the corresponding control logic is executed:

[0116] Level 1 state: When the physiological stress index is not higher than the safe stress threshold, i.e. The system determines the stress state level as a safe zone. In this state, the system believes that the activities of the robotic arm have not caused significant interference to the seahorse. Therefore, the decision control module will generate an execution strategy to maintain the current optimal path, and instruct the flexible robotic arm of the operation execution module to perform the fishing and pool separation operations normally according to the optimal path and speed pre-planned based on the three-dimensional digital twin model.

[0117] Secondary state: When the physiological stress index is higher than the safe stress threshold but not higher than the critical stress threshold, i.e. The system determines the stress state level to be in the warning zone; this indicates that the hippocampus has begun to perceive the threat and generate initial tension; at this time, the system will generate a correction strategy to dynamically reduce the movement speed of the robotic arm in order to actively reduce interference with the target.

[0118] The specific correction strategy for this warning zone is as follows: real-time physiological stress index... Safety stress threshold With critical stress threshold We input a preset speed correction function to calculate the corrected robotic arm movement speed. The function is designed as follows:

[0119]

[0120] in, : Represents the corrected speed output to the robotic arm, its dimensions are... Consistent;

[0121] : Represents the original speed of the current plan, given by the optimal path planning algorithm;

[0122] These are the real-time stress index, the safety threshold, and the critical threshold, all of which are dimensionless parameters.

[0123] : is a dimensionless modulating factor greater than 1, used to adjust the nonlinearity of the rate decay curve; its value is determined based on variety sensitivity experimental data, where sensitivity can be quantified as the stress index under standard incremental stimulation. The average slope of the response curve; the steeper the slope, the larger the value should be. value;

[0124] The corrected robotic arm movement speed will vary with the physiological stress index. The increase is smooth and non-linear from The intensity of the intervention decreases to zero; this intelligent approach behavior can reduce the intensity of the intervention in a gradual and safe manner, effectively calming the target and preventing further escalation of the stress state.

[0125] Level 3 state: When the physiological stress index is higher than the critical stress threshold, i.e. The system determines the stress level to be a danger zone; this indicates that the seahorse is in a state of high stress, and any continued approach may cause it to struggle violently or cause physiological damage; at this time, the system will generate an avoidance strategy to stop the task and retreat to a safe distance; the operation execution module will immediately instruct the robotic arm to stop the current fishing task and actively move backward to a safe distance that will not cause stress to the seahorse, and then suspend the operation.

[0126] After implementing the danger zone avoidance strategy, the system does not permanently stop but enters a waiting mode. In this mode, the status monitoring module continuously monitors the physiological stress index, and when the index reaches a certain level... Falling back to the safe stress threshold Once the system determines that the seahorse's stress state has returned to normal, it will resume the harvesting and separation operations.

[0127] Through this hierarchical adaptive correction and recovery logic, this invention constructs a truly biosignal-driven closed-loop control system. This system can not only respond proportionally and smoothly to the stress level of the seahorse, but also execute decisive abortion and avoidance operations in extreme situations, and intelligently resume the task after the danger has passed. This refined, multi-level control strategy ensures that the harvesting and pool separation operations are always carried out within the physiological tolerance range of the seahorse, minimizing physical intervention and achieving safe and non-destructive acquisition of live targets.

[0128] This scheme optimizes the physiological stress index function and the velocity correction function; for the physiological stress index... To avoid normalization failures caused by fluctuations in historical data ranges, the system introduces a dynamic safety margin in its calculations. When any raw physiological data exceeds its historical maximum value, the system adaptively expands the normalized dynamic range of that feature to ensure its normalized value. It always remains within the valid range [0,1].

[0129] In the velocity correction function, when the critical stress threshold is... With safety stress threshold The difference is less than a preset minimum value. To prevent the denominator from approaching zero and causing computational instability, the system will directly implement a danger zone avoidance strategy; this strategy can be defined as: if Then execute directly. ;

[0130] This correction ensures that even in rare cases where the threshold settings are extremely close, the system can make a safe and decisive response, preventing the robotic arm from making uncontrollable movements due to calculation errors.

[0131] In addition, to address situations where sensor data is temporarily lost or of low quality, the decision control module also incorporates data validity verification logic. If valid physiological data is not received within a preset time window, the system will default to entering the warning zone and implement a slow approach strategy instead of maintaining high-speed movement to ensure safety. If data loss continues for more than a longer threshold, a danger zone avoidance strategy will be implemented until the data returns to normal, thereby ensuring the system's fault-tolerant capability in the event of sensor malfunctions.

[0132] Example 6:

[0133] Please see Figure 2 A smart grading and pond-separating seahorse culture system based on appearance characteristics, comprising:

[0134] The data acquisition module is used to collect sparse visual data and multi-source physiological data from the hippocampus.

[0135] The modeling and evaluation module is used to generate a three-dimensional digital twin model based on sparse visual data and to calculate the growth status grading index based on the three-dimensional digital twin model.

[0136] The status monitoring module is used to calculate the physiological stress index based on multi-source physiological data;

[0137] The decision control module is used to compare the physiological stress index with the preset safe stress threshold and critical stress threshold, and generate a graded adaptive correction strategy.

[0138] The job execution module receives the hierarchical adaptive correction strategy and controls the flexible robotic arm to perform fishing and pool separation operations.

[0139] This embodiment provides a seahorse intelligent grading and pond-separation breeding system based on appearance characteristics. This system is a combination of hardware and software designed to implement the aforementioned method, and includes:

[0140] A data acquisition module is designed to acquire raw data from seahorses in a non-contact manner. In this embodiment, the module consists of a long-range visual sensor and a non-contact visual sensor array deployed in the breeding environment. It is responsible for performing the aforementioned S1 step, which is to acquire sparse visual data and multi-source physiological data of seahorses and transmit these data to subsequent modules in real time.

[0141] A modeling and evaluation module is designed to quantitatively evaluate the growth status of the hippocampus. This module is typically a server or edge computing device that integrates high-performance computing units, such as GPUs, and internally embeds pre-trained deep generative models and nonlinear evaluation models. It is responsible for performing the S2 step, which generates a three-dimensional digital twin model based on sparse visual data and calculates the growth status grading index based on the model.

[0142] A state monitoring module is designed to quantify the physiological stress level of the hippocampus in real time. This module can be a standalone signal processing unit or a software algorithm integrated into the main controller. It has an embedded physiological stress index function. It is responsible for executing step S3, which is used to calculate the physiological stress index in real time based on multi-source physiological data.

[0143] A decision control module is designed to generate intelligent control strategies based on real-time status. This module is the central control unit of the system, usually implemented by a central processing unit (CPU) or microcontroller (MCU). Its core is hierarchical adaptive correction logic. It is responsible for executing step S4, which compares the physiological stress index calculated by the status monitoring module with the preset safety and critical stress thresholds and generates the corresponding hierarchical adaptive correction strategy.

[0144] A job execution module is designed to perform the final physical operations. This module mainly consists of a high-precision flexible robotic arm and its motion controller. It is responsible for executing the S5 step, which is to receive the hierarchical adaptive correction strategy generated by the decision control module and precisely control the flexible robotic arm to perform a series of actions such as normal movement, dynamic deceleration, stopping and reversing.

[0145] This system integrates data acquisition, modeling and evaluation, status monitoring, intelligent decision-making, and physical execution through the organic combination and collaborative work of the above modules, forming a complete and automated intelligent grading and pond-based aquaculture system. Each module performs its own function, and the data flow and control flow are clear and efficient, thereby transforming complex methods and processes into a stable, reliable, and deployable industrial solution, providing solid technical support for achieving high-quality and standardized processing of precious live aquatic products.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics, characterized in that, Includes the following steps: S1. Collect sparse visual data and multi-source physiological data of the hippocampus; S2. Generate a 3D digital twin model based on sparse visual data; calculate growth state grading index based on the 3D digital twin model; S3. Calculate the physiological stress index based on multi-source physiological data; S4. Compare the real-time calculated physiological stress index with the preset safe stress threshold and critical stress threshold to generate a graded adaptive correction strategy. S5. Based on the hierarchical adaptive correction strategy, control the flexible robotic arm to perform fishing and pool separation operations; S1 specifically includes: A sparse visual dataset is constructed by acquiring low-resolution two-dimensional images of the target from multiple preset azimuth angles using a long-range visual sensor. Multi-source physiological data is collected using non-contact visual sensors; these multi-source physiological data include gill movement frequency, limb movement frequency and amplitude, and instantaneous body color change rate. S3 specifically includes: Normalize the multi-source physiological data; Based on the preset sensitivity weighting coefficients, the normalized multi-source physiological data are weighted and calculated to generate a physiological stress index. The calculation process is achieved through the following physiological stress exponential function: ； in, : Represents the final dimensionless physiological stress index; the higher the value, the stronger the stress level in the hippocampus. : These represent the gill movement frequency change rate, escape behavior index, and body color change rate after the aforementioned normalization treatment, respectively. Their values ​​are all calculated from real-time monitoring data. : These are the dimensionless sensitivity weighting coefficients for the gill movement frequency change rate, escape behavior index, and body color change rate mentioned above, and the sum of the three is 1; When the physiological stress index is higher than the critical stress threshold, the stress state level is determined to be a danger zone, and an avoidance strategy of terminating the task and retreating to a safe distance is generated. When the physiological stress index is not higher than the safe stress threshold, i.e. The system determines the stress state level as a safe zone; When the physiological stress index is higher than the safe stress threshold but not higher than the critical stress threshold, that is... The system determines the stress state level as the warning zone; when the physiological stress index exceeds the critical stress threshold, i.e. The system determined the stress state level to be a danger zone; S4 specifically includes: When the physiological stress index is not higher than the safe stress threshold, the stress state level is determined to be the safe zone, and an execution strategy to maintain the current optimal path is generated. When the physiological stress index is higher than the safe stress threshold but not higher than the critical stress threshold, the stress state level is determined as the warning zone, and a correction strategy to dynamically reduce the movement speed of the robotic arm is generated. The specific correction strategy for the warning zone is as follows: The physiological stress index, safe stress threshold, and critical stress threshold are input into a preset speed correction function to calculate the corrected robotic arm movement speed; wherein, the corrected robotic arm movement speed decreases smoothly and non-linearly with the increase of the physiological stress index. The function is designed as follows: ； in, : Represents the corrected speed output to the robotic arm, its dimensions are... Consistent; : Represents the original speed of the current plan, given by the optimal path planning algorithm; These are the real-time stress index, the safety threshold, and the critical threshold, all of which are dimensionless parameters. : is a dimensionless regulatory factor greater than 1.

2. The method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics according to claim 1, characterized in that, S2 specifically includes: A sparse visual dataset is input into a pre-trained deep generative model to generate a 3D digital twin model. Extract a set of growth characteristic parameters from a 3D digital twin model; among them, the growth characteristic parameters include the estimated values ​​of body length, height, weight, and body proportions. The growth characteristic parameter set is input into a preset grading model, and weighted calculations are performed to generate a growth status grading index.

3. The method for intelligent grading and separate-pond rearing of seahorses based on appearance characteristics according to claim 2, characterized in that, After the danger zone avoidance strategy is implemented, it also includes: Continuously monitor the physiological stress index, and resume fishing and pond separation operations when the physiological stress index falls below the safe stress threshold.

4. A seahorse intelligent grading and pond-separation culture system based on appearance characteristics, based on the seahorse intelligent grading and pond-separation culture method based on appearance characteristics according to any one of claims 1-3, characterized in that, include: The data acquisition module is used to collect sparse visual data and multi-source physiological data from the hippocampus. The modeling and evaluation module is used to generate a three-dimensional digital twin model based on sparse visual data and to calculate the growth status grading index based on the three-dimensional digital twin model. The status monitoring module is used to calculate the physiological stress index based on multi-source physiological data; The decision control module is used to compare the physiological stress index with the preset safe stress threshold and critical stress threshold, and generate a graded adaptive correction strategy. The job execution module receives the hierarchical adaptive correction strategy and controls the flexible robotic arm to perform fishing and pool separation operations.

Images