Method for training a remote semen collection of a local boar based on interaction with a simulation person

Abstract

The application discloses a local boar remote semen collection training method based on human simulation interaction, and particularly relates to the field of artificial reproductive technology, which comprises obtaining environmental parameters to calculate a situation interaction adaptation index, executing a human simulation behavior path and collecting boar response data to construct a behavior cognition connection index when the situation interaction adaptation index meets preset conditions, inputting the two indexes into a training model to output a cognition transition prediction score, reconstructing the behavior path and reevaluating if the score is lower than a threshold, triggering a real person intervention process if the score is higher than the threshold for m consecutive times and the connection index fluctuation is less than a set range, and the like. The application realizes quantitative evaluation of pig house environmental adaptability, dynamic perception of target boar behavior response, and training decision closed-loop control based on double indexes and a machine learning model, ensures efficient promotion of semen collection training under the condition of strong environmental adaptability, sufficient individual response and stable cognitive trust, and improves training success rate and intelligence level.

Description

Technical Field

[0001] This invention relates to the field of assisted reproductive technology, and more specifically, to a method for training local boars to remotely collect semen based on simulated human interaction. Background Technology

[0002] Artificial semen collection, a core technique in modern pig breeding, is widely used in various commercial pig farms. In standardized white pig breeds, due to their long-term domestication, stable temperament, and high tolerance for human intervention, their cooperation during artificial semen collection is relatively high. However, in non-scale herds, where they generally retain strong primitive behavioral characteristics and wariness instincts, semen collectors often encounter behavioral problems such as individual avoidance, resistance to approach, stressful barking, and even aggression. These stressful behavioral responses not only significantly increase the difficulty and time cost of artificial semen collection but may also interfere with semen quality and collection success rates, hindering the industrial development of local pig genetic resources.

[0003] Existing methods for adaptive modification of semen collection behavior mainly include:

[0004] Positive reinforcement therapy: This method uses food incentives and physical rewards to create positive associations in boars, reducing stress responses when strangers intervene. However, this method is highly dependent on human experience, requires long repetition times, cannot be implemented remotely, and has poor adaptability.

[0005] Gradual acclimatization method: Fixed personnel enter the shed daily to allow the boars to gradually adapt to their presence, but this requires long-term and repeated intervention and is difficult to maintain when manpower is insufficient.

[0006] Simulation induction devices (such as dummy sows): are used to stimulate boar sexual behavior and assist in positioning and semen collection, but they do not have the ability to interact and guide the boar and cannot play a role in establishing behavioral trust.

[0007] Furthermore, the universality of training can fluctuate depending on individual circumstances and the specific conditions of local pig farms. If training proves ineffective, adjustments such as replacing personnel are typically necessary, which is not only inefficient but also heavily reliant on the experience and judgment of the operators. This approach fails to achieve standardized and intelligent training interventions, and even situations where reasonable human behavior may be misinterpreted or insufficiently perceived as a threat by the target boar. Therefore, this invention proposes a remote semen collection training method for local boars based on simulated human interaction to address these issues. Summary of the Invention

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for training local boars to remotely collect semen based on simulated human interaction includes the following steps:

[0010] Before the first intervention of the simulated human, the ambient light intensity, visual occlusion degree and odor transmission trend of the pig house are obtained through illumination measurement, spatial structure modeling and air flow field analysis. The situational interaction adaptation index is calculated to determine the degree of environmental condition suitability for performing training behaviors. If the environment changes abruptly during the training process, the index is re-evaluated.

[0011] When the situational interaction adaptation index meets the prerequisite for triggering humanoid behavior, the initial behavior path of the simulated human is executed, including posture changes, voice output and odor release. At the same time, the response behavior data of the target boar is acquired. Based on the frequency of behavior approach, the magnitude of physiological stress reduction and the dwell time, a behavior-cognitive connection index is constructed to reflect the degree to which the humanoid behavior is cognitively accepted by the target individual.

[0012] The situational interaction adaptation index and the behavioral cognition connection index are input into a machine learning model built based on historical training data. The model outputs a cognitive transition prediction score, which is then compared with a set transition threshold to determine whether there is a behavioral trust basis for completing sperm collection preparation.

[0013] When the cognitive transition prediction score is lower than the transition threshold, the behavioral path is reconstructed by combining the relative offset direction and feedback characteristics of the situational interaction adaptation index and the behavioral cognitive connection index. This includes adjusting the amplitude of the movement, changing the phonological rhythm and optimizing the odor release rhythm. An interactive behavior sequence matching the current individual response characteristics is generated. After execution, feedback is collected again and the behavioral cognitive connection index is recalculated.

[0014] When the cognitive transition prediction score is not lower than the transition threshold for m consecutive times, and the behavioral cognitive connection index fluctuates within the evaluation window without exceeding the set stability threshold, the human intervention handover process is initiated. By controlling the human behavioral parameters to be consistent with the humanoid path, the semen collection behavior is seamlessly transferred based on the target boar's cognition.

[0015] In a preferred embodiment, during the process of obtaining the ambient light intensity, visual obstruction degree and odor transmission trend of the pig house through light measurement, spatial structure modeling and air flow field analysis, the ambient light intensity is calculated as follows: after collecting the illuminance value per unit area at multiple fixed measuring points, the average illuminance of all measuring points is calculated as the current ambient light intensity of the pig house.

[0016] The calculation method for visual occlusion is as follows: a three-dimensional model of the spatial structure is established, and the ratio of the connection interruption rate to the total number of visible connections is calculated based on whether the visible connection between each node of the simulated human's behavioral path and the head of the target boar passes through obstacles. This ratio is used as the degree of visual occlusion.

[0017] The odor transmission trend is calculated as follows: Based on the airflow field analysis, a pre-set numerical prediction model based on gas diffusion simulation is used to generate a diffusion range coverage map per unit time at the pre-set odor release point. The outer contour of the location where the odor concentration is equal to 20% of the initial value is defined as the high-concentration odor transmission boundary, and the position of this area changes over time. At the same time, through the continuous position trajectory of the target boar in the first five minutes of training, a time-weighted average method is used to establish a distribution map of the target boar's dwell probability. The average dwell area of ​​the boar is defined as the spatial cluster area where its dwell probability is greater than 60%. The area of ​​the overlap between the high-concentration odor transmission boundary and the average dwell area of ​​the boar in space is divided by the area of ​​the odor transmission boundary to obtain the quantitative expression value of the odor transmission trend.

[0018] In a preferred embodiment, the method for establishing the target boar residence probability distribution map using the time-weighted average method is as follows:

[0019] Before the initial intervention of the simulated human, the location data of the target boar within a preset time window is collected using a positioning and recognition device. The pigsty space is divided into several equidistant grid units, and at each moment, the grid number of the target boar and the relative time weight of that moment with respect to the entire time window are recorded. The time weight is defined as the ratio of the difference between the current moment and the end time of the window to the total time length, forming a weighted dwell value for each position. Then, all weighted dwell values ​​in each grid unit are summed and divided by the total number of times the grid is visited to obtain the weighted dwell score for each grid. Finally, the weighted dwell scores of all grids are normalized to obtain a spatial distribution map of the dwell probability within the time window.

[0020] In a preferred embodiment, calculating the contextual interaction adaptation index includes the following steps:

[0021] First, the ambient light intensity, visual occlusion degree and odor transmission trend are normalized to obtain variables L, V and G respectively. Variable L represents the normalized ambient light intensity, V represents the reciprocal function value of the normalized occlusion rate, and G represents the normalized odor coverage probability region.

[0022] Then, the information entropy εi of each variable i is calculated as an objective measure of the contribution of variable i to the sample space, where i is the index corresponding to variables L, V, and G. The calculation method is as follows:

[0023] k is a preset regularization coefficient, Pij represents the proportion of the normalized value of the i-th index (i.e., illumination, visual occlusion, and odor transmission) in the j-th time slice relative to the sum of all time slices for that index, and n represents the total number of time slices before the first intervention of the simulated human.

[0024] Next, calculate the variable weights wi: And assign variable weights w1, w2, and w3 to variables L, V, and G respectively;

[0025] Define the ideal environment point as [d1, d2, d3], and perform the following calculations with the variable combinations under each time slice:

[0026] D represents the deviation distance, and its maximum value is set to Dmax;

[0027] Finally, substituting into the formula for calculating the contextual interaction adaptation index: S = e -Dmax S represents the contextual interaction adaptation index.

[0028] In a preferred embodiment, during the process of executing the initial behavioral path of the simulated human, including posture changes, voice output and odor release, while acquiring the response behavior data of the target boar, the behavioral approach frequency is calculated as follows: within the observation window during the execution of the behavioral path by the simulated human, the number of times the target boar moves from the initial position toward the simulated human and approaches within a preset distance threshold is recorded, and the number of times is divided by the observation window duration to obtain the behavioral approach frequency per unit time.

[0029] The method for calculating the decrease in physiological stress is as follows: the average value of the preset stress parameters of the target boar at two time points before and after the execution of the behavior path is collected, and the relative decrease ratio is obtained by dividing the difference between the two time points by the previous value, which is used as the decrease in physiological stress.

[0030] The calculation method for dwell time is as follows: During the execution of the behavior path, the cumulative time that the target boar stays in the area affected by the simulated human behavior is identified. This area is defined as the space within a specified radius with the center of the simulated human behavior as the center. The dwell time of the target boar in this area is counted each time and summed to obtain the dwell time within the complete behavior path execution cycle.

[0031] In a preferred embodiment, constructing the behavioral cognitive cohesion index includes the following steps:

[0032] Obtain the standard behavioral response intervals corresponding to the frequency of behavioral approaches, the magnitude of physiological stress reduction, and the dwell time, and extract the median of each standard behavioral response interval to form a standard triplet;

[0033] The behavioral approach frequency, physiological stress decline magnitude, and dwell time were processed using the S-shaped growth function, bimodal stabilization function, and logarithmic enhancement function, respectively, so that each data value was mapped to a membership value between zero and one. The response triples were obtained by arranging the same data type in the standard triplet corresponding to the three membership values ​​in order.

[0034] Calculate the Euclidean distance between the response triplet and the standard triplet, which is equal to the value of the behavioral-cognitive connectivity index.

[0035] In a preferred embodiment, the machine learning model is a support vector regression model, which is obtained by fitting and training based on historical training data. The contextual interaction adaptation index and the behavioral cognition connection index are used as two-dimensional input vectors, and the output is a cognitive transition prediction score between zero and one.

[0036] In a preferred embodiment, a behavior path reconstruction strategy is adopted when performing behavior path reconstruction. The behavior path reconstruction strategy is established based on a preset response classification rule base and is used to dynamically generate an interactive behavior sequence that matches the current individual response characteristics and re-execute it.

[0037] In a preferred embodiment, while the behavioral cognitive connection index fluctuates within an evaluation window without exceeding a set threshold range, the evaluation window is defined as m consecutive complete behavioral path execution cycles. After each cycle, the output value of the behavioral cognitive connection index is recorded to obtain m sets of continuous index values. Then, the difference between the maximum and minimum values ​​in these m sets of values ​​is calculated as the fluctuation range of the behavioral cognitive connection index. If the fluctuation range is less than or equal to a preset stability threshold, and the cognitive transition prediction score is not lower than the transition threshold for m consecutive times, then the current behavioral cognitive state is determined to have a continuous acceptance characteristic.

[0038] The technical effects and advantages of this invention are as follows:

[0039] This invention achieves a quantitative judgment on whether the internal environmental conditions of the pigsty are suitable for interactive training behavior by performing environmental lighting measurement, spatial structure modeling, and airflow field analysis before the initial intervention of the simulated human, and calculating the situational interaction adaptation index accordingly. The calculation of this index ensures that the simulated human behavior path is initiated only under the premise of moderate lighting intensity, acceptable visual obstruction, and effective coverage of odor transmission trends, avoiding the failure of training effect due to sudden changes in the external environment, and providing a highly reliable initiation judgment mechanism for the training process.

[0040] This invention, during the execution of an initial behavioral path by a simulated human, synchronously collects the target boar's behavioral approach frequency, physiological stress reduction magnitude, and dwell time, and constructs a behavioral cognitive connectivity index based on this data. This establishes a quantitative expression of the target individual's response to the simulated human interactive behavior. This index can accurately reflect the target boar's cognitive acceptance of the current behavioral path, thus providing a basis for the formation of behavioral trust. This allows the training system to evaluate behavioral effects in real time at the individual response level and support adaptive adjustments.

[0041] This invention further introduces a collaborative decision-making mechanism combining a contextual interaction adaptation index and a behavioral cognitive connection index. By constructing a machine learning model trained on historical training data, a cognitive transition prediction score is calculated and compared with a set threshold to dynamically assess the completion status of semen collection preparation. When the score is low, behavioral path reconstruction is automatically triggered; if the score consistently and stably exceeds the threshold, the human handover process is initiated. This mechanism achieves a closed-loop process control from humanoid behavioral paths to human behavioral transfer, ensuring that semen collection is only performed when the target individual has stable cognitive acceptance, thus guaranteeing the intelligence, consistency, and practicality of the training system. Attached Figure Description

[0042] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0043] Figure 1 This is a schematic diagram of the principle of the remote semen collection training method for local boars based on simulated human interaction in this invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] Reference Figure 1 The following examples were obtained:

[0046] Example 1: A method for training local boars to remotely collect semen based on simulated human interaction, comprising the following steps:

[0047] Step 1: Environmental Adaptation Assessment Before Initial Human Intervention. Before the initial human intervention, the ambient light intensity, visual obstruction level, and odor transmission trend of the pigsty are obtained through illumination measurement, spatial structure modeling, and airflow field analysis. A situational interaction adaptation index is calculated to determine the degree of environmental condition suitability for the training behavior. If the environment undergoes abrupt changes during training, the index is reassessed. The purpose of this step is to perform multi-dimensional perceptual modeling and adaptability assessment of the pigsty's current environment before the human performs any behavior, ensuring the accessibility of behavioral stimuli in terms of spatial structure, light visibility, and odor propagation paths. Illumination measurement confirms whether the target boar's visually active area possesses the basic lighting conditions for observing human-like behavior. Spatial structure modeling identifies whether there are obstructions, blind spots, or blocking areas between the human-like path and the boar. Airflow field analysis provides a basis for assessing the coverage effect of odor transmission. By normalizing the above three parameters and inputting them into the index model, the situational interaction adaptation index is calculated, enabling a quantitative judgment on whether the current environment meets the triggering prerequisites for human-like behavior. For example, the humanoid behavior trigger condition is met only when the situational interaction adaptation index reaches the preset conditions. If changes such as a sudden drop in light intensity, a sudden change in wind speed, or an adjustment of the pigsty space are detected, the system will re-evaluate the index to dynamically adjust the timing and method of humanoid behavior intervention. The changes such as a sudden drop in light intensity, a sudden change in wind speed, or an adjustment of the pigsty space mentioned here refer to changes that exceed the preset allowable range. If the change is only minor, no re-evaluation will be performed. For example, the preset allowable range for light intensity fluctuation is within ±10%. If the initial average illuminance of the environment is 200 lx (lux), the system allows light intensity fluctuations within the range of 180–220 lx during training. When the measured illuminance is lower than 180 lx (such as a sudden drop to 150 lx), it is judged as a sudden drop in light intensity, and the index will be recalculated. However, if the fluctuation is 195 lx or 210 lx, it is considered a minor change and no re-evaluation will be performed.

[0048] Step Two: Execute the initial behavioral path of the simulated human and collect the response data of the target boar. The initial behavioral path of the simulated human is executed, including posture changes, speech output, and odor release. Simultaneously, the response behavior data of the target boar is acquired. Based on the frequency of behavioral approach, the magnitude of physiological stress reduction, and the duration of stay, a behavioral cognitive connection index is constructed to reflect the degree to which the simulated human behavior is cognitively accepted by the target individual. This step is executed after environmental adaptation conditions are met and represents the initial establishment of a behavioral interaction relationship between the simulated human and the target boar. The simulated human behavioral path is preset as a set of structured actions and multimodal stimulus signals, including amplitude control of body movements, the rhythm of simulated speech playback, and the frequency and timing of odor release. The behavioral response of the target boar is simultaneously collected, including the frequency of actively approaching the simulated human, the proportion of changes in physiological parameters before and after stress, and the cumulative time spent within the influence range of the simulated human behavior. These three indicators reflect whether the target individual exhibits cognitive acceptance behaviors such as trust, attachment, or curiosity towards the current simulated human behavior. The three data points are processed through normalization and function mapping to calculate the behavioral cognitive connection index, forming a quantifiable behavioral acceptance score.

[0049] Step 3: The cognitive transition prediction scoring model determines the degree of behavioral preparation readiness. The situational interaction adaptation index and the behavioral cognitive connection index are input into a machine learning model built based on historical training data. The model outputs a cognitive transition prediction score, which is compared with a set transition threshold to determine whether the behavioral trust basis for semen collection preparation is met. After obtaining these two indices, the system uses them as bivariate inputs into pre-trained prediction models such as a support vector regression model. This model uses supervised learning modeling based on a large number of historical behavioral samples and has the ability to identify whether a trust loop has been formed between the current environmental conditions and the boar's behavior. The model output is a continuous score value, representing the degree of cognitive transition readiness from humanoid behavior to real-human behavior. The system compares this score with a preset transition threshold in real time. If the score is higher than the threshold, it indicates that the conditions for semen collection behavior intervention have been met; if the score is lower than the threshold, subsequent behavioral optimization processes need to be executed to continue training.

[0050] Step Four: Behavioral Path Reconstruction and Re-collection of Response Data. When the cognitive transition prediction score falls below the transition threshold, the system reconstructs the behavioral path by combining the relative offset direction and feedback characteristics of the contextual interaction adaptation index and the behavioral cognitive connection index. This includes adjusting the amplitude of movements, changing the speech rhythm, and optimizing the odor release rhythm. An interactive behavioral sequence matching the current individual's response characteristics is generated. After execution, feedback is collected again, and the behavioral cognitive connection index is recalculated. When the score fails to meet the target, the system adjusts the structure of the existing humanoid behavioral path to continuously improve the behavioral guidance effect, creating an adaptive behavioral sequence for the current target individual. This adjustment process triggers different behavioral parameter combination strategies based on the offset direction between the two indices. For example, if the environment adapts well but the behavioral response is weak, or the behavioral response is strong but the environment adapts poorly. These strategies may include amplifying the amplitude of movements, accelerating the speech rhythm, or shortening the odor release interval. The behavioral path reconstruction strategy selects matching rules from a preset response rule base to recombine humanoid behavioral steps and implement training. After executing the new behavioral path, the system collects behavioral responses again and recalculates the behavioral cognitive connection index, forming a self-optimizing closed loop.

[0051] Step 5: After assessing cognitive acceptance stability, initiate the human intervention handover. When the cognitive transition prediction score remains above the transition threshold for m consecutive times, and the behavioral cognitive connection index fluctuates within the assessment window without exceeding the set stability threshold, initiate the human intervention handover process. By controlling the human's behavioral parameters to maintain consistency with the anthropomorphic path, seamless transfer of semen collection behavior based on the target boar's cognitive foundation is achieved. This step, as the endpoint of the anthropomorphic interaction phase, employs a dual confirmation mechanism: first, the cognitive transition prediction score reaches or exceeds the predetermined threshold three consecutive times; second, the fluctuation range of the behavioral cognitive connection index within three consecutive behavioral cycles does not exceed the initially set stability threshold, typically one-tenth. This combined strategy avoids short-term fluctuations from misleading the judgment. After successful judgment, the system triggers the human interaction process. The human operator will mimic the actions based on the anthropomorphic execution history path, with their voice output, behavioral rhythm, and spatial displacement all referencing the anthropomorphic path to ensure continuity and consistency of behavioral style and rhythm at the perceptual level, guaranteeing that semen collection is completed without cognitive disruption.

[0052] Before the initial intervention of the simulated human, to achieve multi-dimensional quantitative perception of the current pigsty environment, the system divides the training preparation phase into n consecutive time slices. Environmental parameters of the pigsty are independently collected within each time slice, and the following three analytical operations are performed: illumination measurement, spatial structure modeling, and airflow field analysis. This yields three fundamental variables for subsequent calculation of the situational interaction adaptation index: ambient light intensity, visual occlusion level, and odor transmission trend. Ambient light intensity refers to the visible light illuminance level per unit area at different locations within the pigsty, used to assess the visual observability of the simulated human behavior. Its calculation method is as follows: within the current time slice, illuminance sensors collect the illuminance value per unit area at multiple pre-set fixed points in the pigsty space, recording it in lux. The illuminance values ​​of all points within the time slice are then arithmetically averaged, and the resulting mean is taken as the ambient light intensity for that time slice.

[0053] Visual occlusion refers to the degree to which the line of sight between the simulated human and the target boar is interrupted by structures, facilities, or other obstructions within the pigsty during the execution of a behavioral path. This parameter is obtained through spatial structure modeling. Specifically, a 3D spatial model of the current pigsty is constructed, the position of each behavioral node in the simulated human's behavioral path is set, and it is calculated whether the visible line from that node to the target boar's head crosses an obstacle. If the line crosses an obstructing volume, it is defined as an interruption. The ratio of the number of interrupted lines to the total number of visible lines generated by all behavioral nodes within the time slice is counted as the visual occlusion degree of the current time slice.

[0054] Odor transmission trend refers to the ability of an odor released by a simulated human to diffuse in the airflow field and cover the active behavioral area of ​​the target boar, reflecting the accessibility of olfactory interaction signals. Within the current time slice, a flow field simulation is first performed using aerodynamic modeling tools. Then, using a pre-defined gas diffusion model, the diffusion range per unit time at the odor release points along the simulated human's behavioral path is calculated, forming an odor diffusion concentration distribution map. A boundary where the odor concentration equals 20% of the initial value is defined as the high-concentration odor propagation boundary, indicating that within this boundary, the target individual can clearly perceive the odor signal.

[0055] The gas diffusion model used in this invention belongs to a standard simulation modeling method based on numerical calculation. Its theoretical basis comes from the principles of fluid dynamics and mass transport, and can be implemented in the following way: Modeling principle source: The model is based on the establishment of air flow state and turbulent behavior in the target pig house environment, with Newtonian fluid as the basic assumption, and simulates the diffusion path of gas in three-dimensional space as it changes over time. The diffusion path is not only affected by the release intensity, but also coupled and controlled by factors such as air velocity, obstacle position, vent orientation, and the airtightness of the pig house space.

[0056] The basic simulation process is as follows: First, an aerodynamic modeling tool (such as a commercial general-purpose fluid simulation software) is used to reconstruct the three-dimensional spatial structure of the target pigsty, and basic meteorological parameters such as wind speed, temperature, and humidity are imported. Then, based on multiple odor release points preset in the simulated human behavior path, conditions for continuous odor release are set in the simulation environment, and the diffusion path, velocity, and concentration changes of the gas from the release point are tracked per unit time.

[0057] Numerical simulation details: The simulation process involves refining the pigsty space and setting up multiple computational grid cells. At each time point, the concentration changes of gas components in each grid cell are calculated. The calculation results of each grid cell are continuously updated based on the diffusion behavior of the gas in the space over time, thus forming a complete odor concentration distribution map.

[0058] Concentration boundary definition and determination method: In the 3D concentration map output by the simulation, the initial concentration at the odor release point is defined as 100%. Subsequently, the boundary range where the odor concentration drops to 20% of the initial value is identified in all computational grids, and the outer contour of this boundary is defined as the "high concentration propagation boundary", representing the spatial range in which the target individual can clearly perceive the odor. This boundary usually has an ellipsoidal or fan-shaped diffusion structure and exhibits temporal evolution.

[0059] Registration with behavioral data: The simulation results are spatially registered with the target boar's dwell probability distribution map, and then the overlap area and ratio between the odor propagation boundary and the average dwell area of ​​the boar are calculated to construct a quantitative indicator of odor transmission trend. The larger the overlap ratio, the more likely the odor release under the current anthropomorphic behavioral path is to be perceived by the target individual. The mentioned "odor diffusion model" is a well-known aerodynamic modeling technique, but in this invention, through fine integration with the semen collection behavior path and quantitative extraction of perception-related areas, the model has specific technical application value.

[0060] Meanwhile, in the n time slices prior to the initial intervention of the simulated human, the system uses positioning and recognition equipment to collect the continuous spatial trajectory of the target boar, and calculates the spatial distribution map of its dwell probability in each time slice using a time-weighted average method. Based on the probability distribution map, the area with a dwell probability greater than 60% is defined as the average dwell area of ​​the boar in the current time slice. Then, a geometric overlap analysis is performed in three-dimensional space between the high-concentration odor propagation boundary area and the average dwell area of ​​the boar in the current time slice, calculating the overlap area. The overlap area is divided by the area of ​​the high-concentration odor propagation boundary as a quantitative expression of the odor transmission trend in the current time slice. The values ​​of light intensity, visual obstruction, and odor transmission trend are all environmental variables collected and formed one by one in the n time slices divided before the initial intervention of the simulated human, used for subsequent calculation of the situational interaction adaptation index.

[0061] Before the initial intervention of the simulated human, a spatial distribution model of the target boar's natural behavioral activity range in the current environment was established. A time-weighted average method was used to construct a spatial distribution map of the target boar's dwell probability, which was then used to assist in calculating the matching of odor transmission trends with the target area for simulated human behavior. This method was constructed based on the collection and processing of continuous location data of the target boar within a preset time window.

[0062] First, within the current pigsty environment, the activity trajectory of the target boar is continuously sampled within a preset time window using a positioning and recognition device, forming two-dimensional or three-dimensional spatial location data corresponding to each time point. The duration of this time window is fixed, covering the complete preparation phase before the simulated human first executes the behavioral path. Subsequently, the system divides the entire pigsty space into several equidistant grid units, forming a regular spatial index structure, so that any location can be mapped to a unique grid number.

[0063] At each sampling moment within the time window, the grid number of the target boar is recorded, and the time weight value for that moment is calculated. The time weight value is defined as the ratio of the time difference between that moment and the end of the entire time window to the total duration of the time window. It represents the proximity of that moment to the end of the window, with a value between 0 and 1. This setting reflects the principle that data closer to the end of the window is assigned a higher weight in the model, thus highlighting the target boar's recent activity preferences.

[0064] Subsequently, all sampling points falling into each grid cell are accumulated to form a weighted dwell value sequence. For each grid cell, the total weighted dwell value of all its contained time weights is summed. To eliminate the influence of grid access frequency differences on distribution offset, the total weighted dwell value is divided by the total number of times the grid is accessed to obtain the weighted dwell score of the grid cell, which represents the spatial preference of the target boar for that location per unit of visit behavior.

[0065] After the above processing, the weighted dwell scores of all grids are normalized to ensure that the sum of the dwell probability values ​​in the entire space is 1. Finally, a spatial distribution map of dwell probability covering the entire pigsty is obtained, with each grid corresponding to a probability value between 0 and 1, representing the target boar's tendency to dwell at that location. This map can serve as the basis for static spatial behavioral feature modeling, used for subsequent identification of the average dwell area of ​​boars and determination of the effective coverage range of odor signals.

[0066] For example, in a 20m x 10m pigsty, the system divides the space into 2m x 2m grids, totaling 50 units. If a target boar stays at grid A 5 times within a time window corresponding to a time slice, at time points with weight values ​​of 0.4, 0.5, 0.6, 0.7, and 0.9 respectively, then the weighted dwell value of grid A is the sum of these five values, i.e., 3.1, with 5 visits and a weighted dwell score of 0.62. After normalizing the scores corresponding to all 50 grids, the probability corresponding to grid A reflects the spatial behavioral preference intensity of the target boar at that location.

[0067] To quantify the feasibility and effectiveness of simulated human behavior paths in a specific pigsty environment, the situational interaction adaptation index is calculated using a composite model based on information entropy weights and weighted Euclidean distance. The core logic of this method is as follows: by continuously monitoring and normalizing three dimensions—ambient light intensity, visual occlusion, and odor transmission trends—a distribution of environmental state evolution within the current time period is established. Then, by using the information entropy of each indicator within this distribution, its fluctuation intensity and weight difference are determined. Finally, the deviation between the environmental state in any time period and the ideal environmental point is nonlinearly decayed to obtain a numerical feasibility adaptation score. The introduction of information entropy aims to dynamically measure the dispersion and uncertainty of each indicator throughout the entire sample time period. A higher information entropy indicates stronger fluctuations and lower stability in different time periods, resulting in greater unpredictability of its impact on behavior; therefore, it is assigned a smaller weight in subsequent calculations. Conversely, variables with lower information entropy indicate more concentrated changes, a more thorough understanding by the system, and higher credibility in the behavioral interaction path; therefore, they are assigned a larger weight. This approach makes behavioral execution decisions more robust. To avoid simple weighted linear bias, the model uses the weighted Euclidean distance between the three weighted input indicators and the ideal point formed by the ideal environmental state as a measure of deviation. A larger deviation indicates that the current environmental state is less suitable for executing the preset humanoid path. Considering that behavioral triggering is a continuous process, the final situational interaction adaptation index is represented by an exponential decay model. This ensures that when the environment deviates significantly from the ideal state, the adaptation score rapidly decays to a low value, reflecting that the system will prevent the execution of the behavioral path and avoid inappropriate interactions causing fear or resistance in the target boar. The acquisition of the situational interaction adaptation index specifically includes the following steps:

[0068] First, the ambient light intensity, visual occlusion degree and odor transmission trend are normalized to obtain variables L, V and G respectively. Variable L represents the normalized ambient light intensity, V represents the reciprocal function value of the normalized occlusion rate, and G represents the normalized odor coverage probability region.

[0069] Then, the information entropy εi of each variable i is calculated as an objective measure of the contribution of variable i to the sample space, where i is the index corresponding to variables L, V, and G. The calculation method is as follows:

[0070] k is a preset regularization coefficient, Pij represents the proportion of the normalized value of the i-th index (i.e., illumination, visual occlusion, and odor transmission) in the j-th time slice relative to the sum of all time slices for that index, and n represents the total number of time slices before the first intervention of the simulated human.

[0071] Next, calculate the variable weights wi: And assign variable weights w1, w2, and w3 to variables L, V, and G respectively;

[0072] Define the ideal environment point as [d1, d2, d3], and perform the following calculations with the variable combinations under each time slice:

[0073] D represents the deviation distance, and its maximum value is set as Dmax. When calculating the deviation between environmental parameters and the ideal point, to more accurately reflect the influence weight of each parameter on the overall behavioral adaptability, this invention introduces a weighted square operation on the basis of weighted Euclidean distance, forming an improved weighted distance model. The weight parameters in this model originate from the objective variable weights calculated based on information entropy, representing the relative importance of the degree of change of each variable in historical samples. To enhance the sensitivity of response to high-weight variables in distance calculation, the variable weights are squared when substituting them into the distance calculation for each dimension. The technical basis of this processing method can refer to the differentiation enhancement principle in the multivariate weighted aggregation evaluation model: that is, for variables that show strong stability, low entropy, and obvious information concentration characteristics in the sample, their corresponding weight values ​​are larger. After the weight squares are involved in the calculation, the influence on the overall distance value is amplified, thereby enabling key influencing factors in environmental conditions to dominate the overall judgment of adaptability, effectively avoiding the distortion of deviation evaluation caused by the accumulation of multiple low-weight variables. The squaring of weights essentially enhances the imbalance among variables in the weighted calculation, improving the ability to identify abnormal behavioral environmental conditions while maintaining a simple model structure. Therefore, using the square of the variable weight multiplied by the squared deviation of the corresponding parameter from the ideal point not only conforms to the mathematical modeling logic of highlighting the effects of key variables in multi-indicator comprehensive evaluation, but also improves the accuracy of judging the matching of behavioral path execution conditions, making the contextual interaction adaptation index more dynamic and accurate in its adaptation.

[0074] Finally, substituting into the formula for calculating the contextual interaction adaptation index: S = e -Dmax; S represents the contextual interaction adaptation index. The evaluation cycle of the contextual interaction adaptation index is set according to the rate of change of environmental variables. If the change of airflow disturbance exceeds the preset change threshold, or if the visual occlusion path is deviated due to the change of the target boar or obstacle position, the environmental parameter collection and the update calculation of the contextual interaction adaptation index will be retried.

[0075] In essence, the ideal environmental points [d1, d2, d3] are not arbitrarily specified values, but are extracted from the actual training data based on the statistically optimal response conditions. Those skilled in the art can statistically summarize the highly adaptable environmental parameter combinations in the successfully trained samples in the target boar semen collection pre-experiment, extract the environmental state with the highest cognitive acceptance rate, and use it as the ideal point for fitting calculation.

[0076] The specific method is as follows: For multiple training samples before successful semen collection, record the corresponding three parameter values: light intensity L (average illuminance in successful semen collection samples, such as 180 lx); occlusion degree V (average spatial visibility of paths with high line-of-sight connectivity); odor coverage G (i.e., the odor coverage map with the highest overlap between the target area and the residence area); use the sample parameters with the highest median or response score as the ideal point triplet [d1, d2, d3]. This method has statistical stability and correlation with behavioral goals, and belongs to the data-driven modeling method known in the field, which can be collected and reproduced through typical experiments. Therefore, the method of obtaining the ideal environmental point is based on data analysis of environmental factors that have been verified to be effective in semen collection from publicly available historical sample sets, and those skilled in the art can complete the parameter extraction without creative labor.

[0077] In this invention, to obtain ideal reference points for environmental parameters to calculate the fit between behavioral paths and desired states, three environmental parameters corresponding to multiple successful semen collection samples—ambient light intensity, visual occlusion level, and odor transmission trend—are centrally characterized. Considering that parameters may fluctuate during successful semen collection in different training samples, the median is used for central tendency extraction to avoid distortion of the ideal point due to extreme values. As a statistically robust statistic, the median is less affected by maxima or minima than the mean. When the sample size is moderate and the data distribution is not severely skewed, it can more stably reflect the core concentration range of the parameters. Although the median does not necessarily strictly correspond to the mathematical "center of gravity" (i.e., the position of the weighted average of all values), it can more effectively avoid evaluation bias caused by abnormal data skewing the mean in practical applications, especially suitable for situations where there is noise in the training samples or some semen collection behaviors deviate from environmental standards. Therefore, the median is introduced as the basis for constructing the ideal environment point triplet in this invention to enhance robustness and stability, ensuring that the calculated situational adaptation index more realistically reflects the core environmental state adapted to most successful semen collection processes. If the actual data distribution exhibits a multi-peaked distribution or strong local concentration, the sample parameter with the highest response score can be used as a substitute, thereby improving the approximation of the optimal adaptation situation.

[0078] In order to determine which training sample has the optimal environmental state among multiple successful semen collection samples and thus extract the ideal environmental point [d1,d2,d3], it is necessary to quantitatively score the behavioral response of each successful sample. This scoring index is defined as the "response score," which measures the overall degree of adaptation of the simulated human behavior to the target boar's cognitive acceptance in the current sample. The specific calculation method is as follows: Input dimension selection: For each successful semen collection sample, the following three behavioral response parameters are collected: behavioral approach frequency (representing the frequency of the boar actively approaching the simulated human per unit time); physiological stress reduction magnitude (reflecting the degree of improvement in the boar's stress level before and after the simulated human intervention); and dwell time (representing the duration of the boar's stay in the area affected by the simulated human behavior). Normalization processing: The three parameters are respectively subjected to max-min standardization processing to normalize all parameters to the [0,1] interval. The three normalized parameters are weighted and summed using empirical weighting factors to obtain the response score R. The weight coefficients corresponding to the empirical weighting factors adopt the initial values ​​of expert experience, which can be set to 3:2:2 by default or dynamically adjusted through Bayesian optimization or other methods. All successful semen collection samples are sorted from high to low according to their response score R. The sample with the highest score is the most cognitively acceptable training case, and its corresponding environmental parameter combination is used as the ideal environment point [d1, d2, d3].

[0079] Assuming there are 5 successful samples, the corresponding behavioral response parameters and normalized scores are shown in Table 1 below: Successful Sample Scoring Table:

[0080]

[0081] The environmental parameter combination of sample C is selected as the ideal environmental point [d1,d2,d3].

[0082] While executing the initial behavioral path of the simulated human, including posture changes, voice output, and odor release, the system acquires the target boar's response behavior data through three behavioral indicators: behavioral approach frequency, physiological stress reduction magnitude, and dwell time. These three data constitute the basic input variables of the behavioral cognitive connection index, which is used to reflect the degree to which the simulated human behavior is cognitively accepted by the target individual.

[0083] The behavioral approach frequency describes whether a target boar actively approaches a humanoid area under the influence of humanoid behavioral stimuli. It is calculated as follows: within the observation window during the humanoid's behavioral path, the target boar's position is continuously tracked. Each instance within this time window where the target boar starts from its initial position, moves towards the humanoid, and eventually approaches the center of the humanoid behavior within a preset distance threshold (e.g., 1.5 meters), is recorded. The cumulative number of these events is used as the approach count. This count is then divided by the window duration to obtain the behavioral approach frequency per unit time, which can be set to "times / minute". A higher frequency value indicates a stronger willingness of the target boar to approach the area under that behavioral path.

[0084] The reduction in physiological stress is used to assess changes in the physiological stress level of a target boar before and after behavioral stimulation. This indicator is calculated as follows: Pre-defined stress parameters of the target boar are collected at two time points, before and after the implementation of the humanoid behavior pathway. These parameters may include, but are not limited to, heart rate, skin temperature, or respiratory rate, all acquired using non-contact sensing devices. The specific devices selected are determined by the user based on actual usage. Multiple samples are collected at each time point, and their average values ​​are calculated, recorded as the pre-value and post-value. The relative decrease rate of the stress parameter is obtained by subtracting the post-value from the pre-value and then dividing by the pre-value, as the reduction in physiological stress. This value is in percentage form; a positive value indicates a decrease in stress, and a negative value indicates an increase in stress. A larger value indicates a calming or maturing effect of the humanoid behavior on the target boar.

[0085] Dwell time represents the cumulative time a target boar voluntarily stays within the area influenced by humanoid behavior, reflecting the degree of interaction pauses and attachment intentions. The area of ​​influence is defined as a spatial range with a fixed radius (e.g., 2 meters) centered on the center point of the humanoid behavior. During the execution of the humanoid behavior path, the system continuously collects the real-time location coordinates of the target boar and determines whether it falls within this spatial area. Timing begins each time the boar enters the area and pauses upon exiting. The dwell time for all periods within this area is accumulated to obtain the dwell time within the complete behavior path execution cycle, in seconds. A higher value indicates a stronger stability and affinity in the interaction process between the target boar and the humanoid behavior.

[0086] For example, during a 60-second behavioral path execution, the target boar actively approached the humanoid model within 1 meter three times, at a frequency of 0.05 times / second (i.e., 3 times / minute). The average skin temperature collected before and after the behavior was 36.9℃ and 36.2℃, respectively. Therefore, the decrease in physiological stress was (36.9-36.2) / 36.9≈0.019, or 1.9%. During this period, the target boar stayed within a 2-meter radius of the humanoid model twice, for 12 seconds and 9 seconds respectively, for a total stay time of 21 seconds. These three values, as the quantitative results of the behavioral response, will be further input into the behavioral cognitive cohesion index calculation model for scoring.

[0087] Constructing a behavioral-cognitive cohesion index includes the following steps:

[0088] Behavioral approach frequency represents the frequency with which the target boar actively approaches the human-like path during the execution of the human-like behavior; physiological stress reduction magnitude represents the relative decrease in the target boar's physiological stress parameters before and after the start of the behavior; and dwell time represents the cumulative duration for which the target boar voluntarily dwells within the influence range of the human-like behavior. To unify these three behavioral data into comparable acceptability metrics, they need to be mapped using nonlinear functions. A standard behavioral response interval is established for each indicator. The standard behavioral response interval is a reasonable and effective range statistically obtained from historical interaction experiment samples. For example, the standard response interval for behavioral approach frequency can be [2 times / minute, 6 times / minute], the standard response interval for physiological stress reduction magnitude can be [5%, 25%], and the standard response interval for dwell time can be [10 seconds, 50 seconds]. For each interval, the median is extracted as a representative response baseline value, and the standard behavioral response intervals corresponding to behavioral approach frequency, physiological stress reduction magnitude, and dwell time are obtained. Furthermore, the medians of each standard behavioral response interval are extracted to form a standard triplet.

[0089] The behavioral approach frequency, physiological stress decline magnitude, and dwell time were processed using the S-shaped growth function, bimodal stabilization function, and logarithmic enhancement function, respectively, so that each data value was mapped to a membership value between zero and one. The response triples were obtained by arranging the same data type in the standard triplet corresponding to the three membership values ​​in order.

[0090] The S-shaped growth function (used for behavioral approach frequency) reflects the process of "poor response near zero, rapid acceptance after exceeding a certain threshold," and takes the following form: Here, X represents the frequency normalization value of the behavior, a represents the growth rate (e.g., 15), and b represents the inflection point (e.g., 0.3). The S-shaped growth function is close to 0 at low values ​​and rapidly approaches 1 after exceeding the critical value.

[0091] The bimodal stabilization function (used to measure the decrease in physiological stress) reflects the process where "too low or too high is not good, and the middle is most suitable": Here, Y represents the normalized value of the decrease in physiological stress, c represents the optimal decrease center (e.g., 0.2), σ represents the behavioral response tolerance bandwidth (e.g., 0.05), and the bimodal stability function has the center value as the peak, and the decrease is faster the farther away from the center.

[0092] The logarithmic enhancement function (used for dwell time) reflects the process of "the longer the dwell time, the better, but the growth rate gradually slows down": Here, Z represents the normalized value of the dwell time, and t represents the amplification factor (e.g., 9). The logarithmic enhancement function grows rapidly in the early stage, then stabilizes in the later stage, and is compressed into the range of 0-1.

[0093] Calculate the Euclidean distance between the response triplet and the standard triplet. This Euclidean distance is equal to the behavioral-cognitive cohesion index. Since the standard triplet is considered to be all 1s after function mapping (i.e., the response strength should be close to the full value in the ideal state), the smaller the distance value, the closer the current response is to the ideal level, the more complete the cognitive cohesion, and the smaller the cohesion index, the stronger the behavioral acceptability.

[0094] The machine learning model is a support vector regression model, which is obtained by fitting and training based on historical training data. The situational interaction adaptation index and the behavioral cognition connection index are used as two-dimensional input vectors, and the output is a cognitive transition prediction score between zero and one. During the training phase, the system uses historical humanoid interaction behavior samples as the dataset. Each sample record includes three types of structured content: (1) the actual situational interaction adaptation index value collected in this training; (2) the behavioral cognition connection index value calculated in this training; (3) after the completion of the humanoid behavior, the system judges whether the training has successfully established a behavioral trust foundation through expert annotation or semen collection intervention results feedback, and obtains the corresponding score label. All score labels are normalized to between zero and one, indicating that the target boar has the confidence level of semen collection preparation after this round of training. For example, a score of 0.85 means that the system believes that the behavioral transformation conditions in this training round are relatively mature, while a score of 0.25 means that behavioral trust has not yet been established.

[0095] During model training, the system uses the contextual interaction adaptation index and the behavioral cognition connection index as two-dimensional input features, forming an input vector [x1, x2], where x1 represents the environmental condition fit and x2 represents the target individual's behavioral response level. The support vector regression model uses a Gaussian radial basis kernel function to establish a feature mapping relationship and is trained and optimized based on the ∈-insensitive loss function. By continuously adjusting the kernel parameters and penalty coefficients, the model fits a nonlinear function f(x1, x2) that maps input features to output scores, and the maximum absolute error of this function on all training samples does not exceed a preset threshold. In the model application phase, when the system collects new contextual interaction adaptation indices and behavioral cognition connection indices, they can be fed as a set of two-dimensional inputs into the trained support vector regression model to obtain a cognitive transition prediction score for the current behavioral state. This score is a real value between 0 and 1; the higher the value, the closer the current humanoid behavior is to the ideal behavioral trust state in terms of environmental conditions and behavioral response.

[0096] For example, if the contextual interaction adaptation index is 0.78 and the behavioral cognition connection index is 0.19 in the current training round, the system combines them into an input vector [0.78, 0.19]. After inputting this vector into the model, the output prediction score is 0.43, indicating that behavioral trust has not yet been formed. The system will then perform path reconstruction and re-collection. If the prediction score is higher than the set transition threshold (e.g., 0.7) for m consecutive rounds, and the fluctuation range of the behavioral cognition connection index is stable, then the subsequent human intervention process is triggered to achieve behavioral transfer.

[0097] When the cognitive transition prediction score is below the transition threshold, the system, combining the relative offset direction and feedback characteristics of the situational interaction adaptation index and the behavioral cognitive connection index, performs behavioral path reconstruction. During this process, the system matches preset response classification rules based on the combined performance type of the two indices, thereby resetting the three behavioral parameter dimensions of the current simulated human's execution sequence: motion amplitude, speech rhythm, and odor release rhythm, and generating a new interactive behavior sequence for re-execution. Motion amplitude refers to the spatial displacement and limb extension angle of the simulated human's posture change in a single behavioral node; speech rhythm refers to the frequency distribution and duration variation of the simulated human's output audio; and odor release rhythm refers to the frequency and time interval of odor release events triggered per unit time. In the specific behavioral path reconstruction, the system first determines the relative offset direction based on the numerical positional relationship between the situational interaction adaptation index and the behavioral cognitive connection index in the current round of evaluation, and matches the following three typical offset states and their corresponding processing logic:

[0098] When the situational interaction adaptation index is high but the behavioral cognition cohesion index is low, it is identified as insufficient behavioral perception sensitivity of the target boar, meaning that the external environment adapts well, but the boar itself does not fully accept or perceive the humanoid behavior. In this case, the system uses the current humanoid behavior sequence as a basis and enhances the parameters without changing the execution nodes: the amplitude value of the action in the current node is increased by 30% to improve dynamic attractiveness; the frequency of the speech rhythm is increased by 20% to enhance the recognizability of the sound rhythm and the intensity of the stimulus; and the time interval between two odor releases is shortened by 40% to accelerate the odor coverage update speed and increase the density of olfactory stimulation.

[0099] When the situational interaction adaptation index is low while the behavioral cognition cohesion index is high, it is identified as insufficient environmental adaptation leading to behavioral occlusion. This means the target boar possesses high behavioral response potential, but due to environmental limitations such as occlusion, lighting, and airflow, the behavioral effect is not fully expressed. In this case, the system does not adjust the original action amplitude value, maintaining stable execution of the humanoid behavior within the existing action scale. Simultaneously, it spatially offsets the positional parameters of the behavioral nodes, repositioning their execution location within the pigsty structural model to optimize the visual angle of the simulated human relative to the target boar's head. Furthermore, it reduces the speech frequency by 10% to minimize reflection interference from high-frequency sound waves in narrow structural areas and increases the odor release delay by one second to avoid peak ventilation fluctuations and improve odor retention time.

[0100] When both indices are low, it is identified as a mismatch between the overall environment and behavior, indicating that the humanoid path execution conditions are not adapted to the target individual's current cognitive state. In this case, the system does not adjust the original behavioral nodes but triggers a full parameter reset strategy. It re-matches the behavioral scenario template most similar to the current feedback pattern in the response classification rule base and selects a new sequence of interaction nodes. Node reset includes the order of behavioral types, the execution timestamp and location information of each node, thereby generating a complete sequence of interactive behaviors for the next round of execution and feedback collection. This behavioral path reconstruction strategy is based on a preset response classification rule base. This rule base analyzes the relationship between index offset patterns and response types in a large number of historical training samples to construct a mapping table with dual-index states as input and behavioral adjustment strategies as output. Each path reconstruction automatically calls the corresponding strategy parameters in the rule base according to the current state to dynamically generate an interactive behavior sequence matching the current individual's response characteristics and re-execute it, achieving self-updating and feedback loop of the behavioral adaptation strategy.

[0101] While the behavioral cognitive connectivity index fluctuates within the evaluation window but does not exceed a set threshold, the system first defines the "evaluation window" as m consecutive complete behavioral path execution cycles. Each complete cycle refers to the entire process from the initial humanoid behavioral path, completing a complete sequence of interactions including posture, voice, and odor, collecting response data, calculating the behavioral cognitive connectivity index, and completing the cognitive transition prediction score, ensuring that each round of interaction has the same structure and functional closed loop. Within this evaluation window, after each cycle is completed, the system immediately records the output value of the behavioral cognitive connectivity index, accumulating a total of m sets of continuous index values, forming a time series of behavioral cognitive responses. Subsequently, the system identifies the maximum and minimum values ​​in this series and calculates the difference between them as the "behavioral cognitive connectivity index fluctuation amplitude" within the evaluation window. If this fluctuation amplitude is less than or equal to the system's preset stability threshold, and the cognitive transition prediction score continuously exceeds the set transition threshold within these m cycles, it can be determined that the target boar's current behavioral cognitive state has "persistent acceptance characteristics."

[0102] The stability threshold is uniformly set during the training initialization phase, with a default value of 0.1, indicating that the behavioral cognitive connection index is allowed to fluctuate within a range of no more than one-tenth over m consecutive cycles. For example, in a certain round of evaluation, the system records the behavioral cognitive connection index as 0.72, 0.78, and 0.74, with a fluctuation range of 0.78-0.72=0.06, which is less than the threshold of 0.1; at the same time, the cognitive transition prediction scores are 0.81, 0.83, and 0.86, all higher than the transition threshold (e.g., set to 0.7). This group of states then satisfies the dual conditions of "stable behavioral continuity" and "achievement of cognitive trust." The two conditions of "fluctuation range judgment" and "continuously high cognitive transition prediction scores" together constitute the necessary prerequisites for triggering the human intervention handover process. Only after the system meets this judgment logic can it initiate the mechanism for mimicking human behavior parameters to migrate to human behavior, controlling the human to perform consistent behavior according to the mimicking human execution path, thus achieving seamless cognitive transfer of semen collection behavior. This mechanism ensures that the target boars not only respond positively in a single round of behavior, but also exhibit a stable and predictable acceptance tendency in multiple cycles, thereby ensuring cognitive continuity and behavioral smoothness in subsequent semen collection operations and avoiding the risk of semen collection failure or interruption.

[0103] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0104] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0105] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0106] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for training local boars to remotely collect semen based on simulated human interaction, characterized in that: Includes the following steps: Before the first intervention of the simulated human, the ambient light intensity, visual occlusion degree and odor transmission trend of the pig house are obtained through illumination measurement, spatial structure modeling and air flow field analysis. The situational interaction adaptation index is calculated to determine the degree of environmental condition suitability for performing training behaviors. If the environment changes abruptly during the training process, the index is re-evaluated. When the situational interaction adaptation index meets the prerequisite for triggering humanoid behavior, the initial behavior path of the simulated human is executed, including posture changes, voice output and odor release. At the same time, the response behavior data of the target boar is acquired. Based on the frequency of behavior approach, the magnitude of physiological stress reduction and the dwell time, a behavior-cognitive connection index is constructed to reflect the degree to which the humanoid behavior is cognitively accepted by the target individual. The situational interaction adaptation index and the behavioral cognition connection index are input into a machine learning model built based on historical training data. The model outputs a cognitive transition prediction score, which is then compared with a set transition threshold to determine whether there is a behavioral trust basis for completing sperm collection preparation. When the cognitive transition prediction score is lower than the transition threshold, the behavioral path is reconstructed by combining the relative offset direction and feedback characteristics of the situational interaction adaptation index and the behavioral cognitive connection index. This includes adjusting the amplitude of the movement, changing the phonological rhythm and optimizing the odor release rhythm. An interactive behavior sequence matching the current individual response characteristics is generated. After execution, feedback is collected again and the behavioral cognitive connection index is recalculated. When the cognitive transition prediction score is not lower than the transition threshold for m consecutive times, and the behavioral cognitive connection index fluctuates within the assessment window without exceeding the set stability threshold, the human intervention handover process is initiated.

2. The method for training local boars to remotely collect semen based on simulated human interaction as described in claim 1, characterized in that, In the process of obtaining the ambient light intensity, visual obstruction degree and odor transmission trend of the pig house through light measurement, spatial structure modeling and air flow field analysis, the ambient light intensity is calculated as follows: after collecting the illuminance value per unit area at multiple fixed measuring points, the average illuminance of all measuring points is calculated as the current ambient light intensity of the pig house. The calculation method for visual occlusion is as follows: a three-dimensional model of the spatial structure is established, and the ratio of the connection interruption rate to the total number of visible connections is calculated based on whether the visible connection between each node of the simulated human's behavioral path and the head of the target boar passes through obstacles. This ratio is used as the degree of visual occlusion. The odor transmission trend is calculated as follows: Based on the airflow field analysis, a pre-set numerical prediction model based on gas diffusion simulation is used to generate a diffusion range coverage map per unit time at the pre-set odor release point. The outer contour of the location where the odor concentration is equal to 20% of the initial value is defined as the high-concentration odor transmission boundary, and the position of this area changes over time. At the same time, through the continuous position trajectory of the target boar in the first five minutes of training, a time-weighted average method is used to establish a distribution map of the target boar's dwell probability. The average dwell area of ​​the boar is defined as the spatial cluster area where its dwell probability is greater than 60%. The area of ​​the overlap between the high-concentration odor transmission boundary and the average dwell area of ​​the boar in space is divided by the area of ​​the odor transmission boundary to obtain the quantitative expression value of the odor transmission trend.

3. The method for training local boars to remotely collect semen based on simulated human interaction according to claim 2, characterized in that, The method for constructing the probability distribution map of the target boar's stay using the time-weighted average method is as follows: Before the initial intervention of the simulated human, the location data of the target boar within a preset time window is collected using a positioning and recognition device. The pigsty space is divided into several equidistant grid units, and at each moment, the grid number of the target boar and the relative time weight of that moment with respect to the entire time window are recorded. The time weight is defined as the ratio of the difference between the current moment and the end time of the window to the total time length, forming a weighted dwell value for each position. Then, all weighted dwell values ​​in each grid unit are summed and divided by the total number of times the grid is visited to obtain the weighted dwell score for each grid. Finally, the weighted dwell scores of all grids are normalized to obtain a spatial distribution map of the dwell probability within the time window.

4. The method for training local boars to remotely collect semen based on simulated human interaction according to claim 3, characterized in that, Calculating the contextual interaction adaptation index includes the following steps: First, the ambient light intensity, visual occlusion degree and odor transmission trend are normalized to obtain variables L, V and G respectively. Variable L represents the normalized ambient light intensity, V represents the reciprocal function value of the normalized occlusion rate, and G represents the normalized odor coverage probability region. Then calculate the information entropy for each variable i. As an objective measure of the contribution of variable i in the sample space, i is the index corresponding to variables L, V, and G, and is calculated as follows: ; The preset regularization coefficients, This represents the percentage of the normalized value of the i-th metric (i.e., illumination, visual occlusion, and odor transmission) in the j-th time slice relative to the sum of that metric across all time slices. This represents the total number of time slices allocated before the initial intervention of the simulated human; Next, calculate the variable weights. : ; and assign weights to variables L, V, and G respectively. , , ; Define the ideal environment point as The following calculations are performed in combination with the variables in each time slice: ; This represents the deviation distance, and its maximum value is then set to... ; Finally, substituting into the formula for calculating the contextual interaction adaptation index: ; This represents the contextual interaction adaptation index.

5. The method for training local boars to remotely collect semen based on simulated human interaction according to claim 4, characterized in that, During the execution of the simulated human's initial behavioral path, including posture changes, voice output, and odor release, while acquiring the target boar's response behavior data, the behavioral approach frequency is calculated as follows: within the observation window during the simulated human's execution of the behavioral path, the number of times the target boar moves from the initial position toward the simulated human and approaches within a preset distance threshold is recorded. This number is then divided by the observation window duration to obtain the behavioral approach frequency per unit time. The method for calculating the decrease in physiological stress is as follows: the average value of the preset stress parameters of the target boar at two time points before and after the execution of the behavior path is collected, and the relative decrease ratio is obtained by dividing the difference between the two time points by the previous value, which is used as the decrease in physiological stress. The calculation method for dwell time is as follows: During the execution of the behavior path, the cumulative time that the target boar stays in the area affected by the simulated human behavior is identified. This area is defined as the space within a specified radius with the center of the simulated human behavior as the center. The dwell time of the target boar in this area is counted each time and summed to obtain the dwell time within the complete behavior path execution cycle.

6. The method for training local boars to remotely collect semen based on simulated human interaction according to claim 5, characterized in that, Constructing a behavioral-cognitive cohesion index includes the following steps: Obtain the standard behavioral response intervals corresponding to the frequency of behavioral approaches, the magnitude of physiological stress reduction, and the dwell time, and extract the median of each standard behavioral response interval to form a standard triplet; The behavioral approach frequency, physiological stress decline magnitude, and dwell time were processed using the S-shaped growth function, bimodal stabilization function, and logarithmic enhancement function, respectively, so that each data value was mapped to a membership value between zero and one. The response triples were obtained by arranging the same data type in the standard triplet corresponding to the three membership values ​​in order. Calculate the Euclidean distance between the response triplet and the standard triplet, which is equal to the value of the behavioral-cognitive connectivity index.

7. The method for training local boars to remotely collect semen based on simulated human interaction as described in claim 6, characterized in that, The machine learning model is a support vector regression model, which is obtained by fitting and training based on historical training data. It uses the contextual interaction adaptation index and the behavioral cognition connection index as two-dimensional input vectors and outputs a cognitive transition prediction score between zero and one.

8. The method for training local boars to remotely collect semen based on simulated human interaction according to claim 7, characterized in that, When performing behavior path reconstruction, a behavior path reconstruction strategy is adopted, and the behavior path reconstruction strategy is established based on a preset response classification rule library. It is used to dynamically generate an interactive behavior sequence that matches the current individual's response characteristics and re-execute it.

9. The method for training local boars to remotely collect semen based on simulated human interaction as described in claim 8, characterized in that, During the process where the behavioral cognition connection index fluctuates within the evaluation window without exceeding the set threshold range, the evaluation window is defined as m consecutive complete behavioral path execution cycles. After each cycle, the output value of the behavioral cognition connection index is recorded to obtain m sets of continuous index values. Then, the difference between the maximum and minimum values ​​in these m sets of values ​​is calculated as the fluctuation range of the behavioral cognitive connection index. If the fluctuation range is less than or equal to the preset stability threshold, and the cognitive transition prediction score is not lower than the transition threshold for m consecutive times, then the current behavioral cognitive state is determined to have a continuous acceptance characteristic.

Images