An automated response system suitable for grass-eating animal wilding training
Patent Information
- Application Number
- CN202610815563.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-08
- Publication Date
- 2026-09-25
AI Technical Summary
[0022]与现有技术相比，本发明的有益效果是：
Smart Images
Figure CN122804706A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wildlife conservation technology, specifically to an automated response system suitable for rewilding training of herbivores. Background Technology
[0002] Herbivores, as a key functional group in terrestrial ecosystems, play vital ecological roles in vegetation regulation, nutrient cycling, and predator food web maintenance. Globally, due to habitat fragmentation, overhunting, and human disturbance, large herbivore populations have declined sharply, with some isolated small populations facing the risk of genetic diversity loss and local extinction. Captive breeding and reintroduction is an important means of restoring endangered herbivore wild populations; however, long-term captivity leads to decreased locomotion, insufficient foraging skills, degeneration of anti-predator behaviors, and a lack of adaptive alertness to human activities, resulting in extremely low success rates for "hard reintroduction." Traditional methods for reconstructing anti-predator behaviors during herbivore reintroduction training largely rely on manual observation and subjective judgment. This approach struggles to achieve real-time, standardized monitoring of behavioral responses and lacks scientific quantitative data to assess an individual's ability to survive in the wild, often leading to inaccurate assessments and failing to meet the decision-making requirements for reintroduction. In designing anti-predation stimuli, existing technologies often focus only on a single stimulus type, failing to consider the impact of different predator types and habitat structure on herbivore behavior. They also cannot dynamically adjust training intensity based on individual plasticity phenotypes, resulting in a mismatch between training effectiveness and the actual adaptation needs of herbivores, impacting the efficiency and success rate of rewilding training. For assessing group behavior coordination, traditional methods typically only observe and test individual individuals, ignoring differences in behavioral coordination between individuals and the influence of group size on individual responses. This leads to a lack of holistic and precise group behavior assessment, making it difficult to maintain stable group vigilance and collective escape capabilities, thus affecting the social integration and long-term survival of released individuals. Regarding habitat adaptation training, existing management measures are mostly single-environment training, conducting behavioral tests only in fixed locations. They cannot predict behavioral strategy transitions based on structural differences between open and concealed areas, nor can they adjust training programs in a timely manner by incorporating group effects and other behavioral factors, making it difficult to proactively mitigate the impact of adverse environmental conditions on released individuals. In existing technologies, the management of each training stage is independent, lacking consideration of the interrelationship between behavioral factors such as anti-predation behavior, habitat adaptability, group coordination, and tolerance to human activities. This makes it impossible to optimize the overall effect of rewilding training, resulting in an incomplete assessment of the rewilding release of herbivores and an unsystematic training approach. It is difficult to provide released individuals with suitable survival skills in the wild, thus restricting the stable growth of wild herbivore populations and the sustainable development of their habitats.
[0003] The purpose of this invention is to provide an automated response system suitable for the rewilding training of herbivores, in order to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides an automated response system suitable for rewilding training of herbivores, the system comprising:
[0005] A motion detection module specifically designed for herbivores acquires motion signals from herbivores, triggering the intelligent control module.
[0006] The rewilding-specific intelligent control module, based on the trigger signal of the motion detection module, dynamically configures stimulation parameters and outputs stimulation activation commands according to the rewilding training stage and individual plasticity phenotype.
[0007] A library of specific stimuli for anti-predation behavior in herbivores includes a library of ambush predators, a library of stalking predators, a library of human activity stimuli, and a library of symbiotic animal control stimuli. All stimuli were recorded in the field, acoustically standardized, and behaviorally validated.
[0008] The multimodal stimulus output module, based on the stimulus activation command, outputs standardized stimuli in single or combined forms through the sound stimulus submodule, odor stimulus submodule, and visual stimulus submodule;
[0009] A dual habitat type comparison test platform, including open area test units and concealed area test units, is used to simultaneously assess the strategy switching ability of herbivores in different habitat structures;
[0010] The synchronous video recording and AI analysis module records behavioral responses throughout the process based on the stimulus output of the stimulus output module, and automatically extracts behavioral variables through deep learning algorithms.
[0011] The group behavior coordination and evaluation unit, based on the analysis results of the synchronous video recording and AI analysis module, quantifies the recorded group size, group alert synchronization rate, collective escape start delay, and consistency of escape direction.
[0012] The rewilding progress quantification tracking system, based on the evaluation results of the group behavior coordination evaluation unit, integrates three-dimensional data of behavioral buffering capacity, strategy differentiation capacity, and human activity appropriateness to calculate a "suitable for release" type rewilding progress index and generate herbivore rewilding training evaluation results.
[0013] Preferably, the rewilding-specific intelligent control module includes a rewilding stage identification unit, a stimulus intensity gradient adjustment unit, and a group coordination triggering unit. The rewilding stage identification unit is configured with stage one markers (emotional development period, 7-9 months old), stage two markers (anti-predation behavior reconstruction period, 10-17 months old), stage three markers (adaptation to the wild environment period, 18-24 months old), stage four markers (pre-release assessment period, 25 months old+), and stage five markers (post-release monitoring period, 26 months old+). The stimulus intensity gradient adjustment unit is based on the behavioral plasticity "suitable for release" theory and dynamically adjusts stimulus parameters according to the herbivore's plasticity phenotype: highly plastic individuals receive moderate stimulus intensity (sound pressure level reduced by 3-5 dB, odor release rate reduced by 30%), moderately plastic individuals receive standard stimulus intensity, and low-plasticity individuals receive high-intensity stimulus with auxiliary reinforcement (sound pressure level increased by 3-5 dB, visual stimulus components added, odor concentration increased by 50%). The group coordination triggering unit identifies the number of individuals in the scene through AI video analysis and automatically classifies them into a single individual (1 individual), a small group (2-3 individuals), and a large group (≥4 individuals). When ≥2 individuals are detected, the delay triggering time is extended to 5-8 seconds.
[0014] Preferably, the herbivore anti-predation behavior specific stimulus library includes a sub-library of ambush predators, a sub-library of pursuit predators, a sub-library of human activity, and a sub-library of companion animal controls. The ambush predator stimulus library includes leopard roars, tiger howls, and lynx calls, characterized by low frequency (<1kHz), short pulses, and high amplitude (85-90 dB SPL), simulating a "covert-outburst" type threat, used to test herbivores' long-range detection capabilities and rapid escape responses in open areas. The pursuit predator stimulus library includes wolf howls and jackal calls, characterized by mid-frequency (1-3kHz), continuous, and medium amplitude (80-85 dB SPL), simulating an "open-pursuit" type threat, used to test herbivores' sustained alertness and maneuver escape strategies against group predators. The human activity stimulus library includes footsteps, voices, tool operation sounds, and dog barks, with peak sound pressure levels of 75-80 dB SPL, simulating common human disturbances within the protected area. This is used to assess the intensity of the "superpredator" effect and the appropriate alertness response after training, with the training objective being to control the escape probability within the 30-45% range. The companion animal control stimulus library includes same-species alert calls, same-species communication calls, neutral bird calls, wind sounds, and flowing water sounds, used to distinguish between specific anti-predation responses and general startle responses.
[0015] Preferably, the multimodal stimulation output module includes an acoustic stimulation submodule, an odor stimulation submodule, and a visual stimulation submodule. The acoustic stimulation submodule is equipped with a standardized acoustic stimulation library, a remote waterproof speaker, and a sound calibration unit, with a frequency response range of 100Hz-20kHz and a maximum output sound pressure level ≥100dB (at 1m). It is used to play the standardized acoustic stimulation in a field environment. The sound calibration unit is used to automatically calibrate the output sound pressure level before each experiment to ensure that the actual output is consistent with the preset parameters (error ≤±1dB). The odor stimulation submodule is equipped with an odor release device, an odor sample library, and an odor diffusion control unit. Odor samples include predator urine, predator feces, and a neutral control. The odor release device is equipped with a wind speed and direction sensor; when the wind speed is >3m / s, the release amount increases, and when the wind direction is away from the target area, the release is paused. The visual stimulation submodule is equipped with a movable mechanical model, an LED flashing array, and a projection device. The movable mechanical model simulates a simplified form of a predator or a similar animal and is equipped with a servo motor to achieve a "crouch-pounce" movement mode.
[0016] Preferably, the dual habitat type comparison testing platform includes an open area testing unit and a concealed area testing unit. The open area testing unit is located in forest gaps, floodplain meadows, or artificially cleared areas, with a field of view >100m, herbaceous vegetation height <30cm, and no shrub obstruction. It is used to test the "precision defense" strategy—utilizing the open field of view for long-distance surveillance and adopting differentiated responses to different types of threats (ambush response > pursuit response). The concealed area testing unit is located in dense shrublands, bamboo forests, or artificially planted tall grass areas, with shrub height 1.5-3.0m, coverage >60%, and a field of view <30m. It is used to test the "generalized fear" strategy—increasing overall alertness, shortening reaction time, and adopting a conservative response to all potential threats. The habitat switching test mode is triggered simultaneously at adjacent open area-concealed area paired sites (50-100m apart), recording the differences in responses of the same group in the two habitats and calculating the strategy switching index.
[0017] Preferably, the synchronous video recording and AI analysis module includes a herbivore-specific behavior recognition model, an individual re-identification model, and an automatic behavior quantification tool. The herbivore-specific behavior recognition model is based on the Mask R-CNN architecture and is trained on typical herbivore postures (head down for grazing, head up for alertness, running, jumping, freezing), achieving an accuracy rate of ≥95%. The individual re-identification model, for herbivores without significant markings (such as roe deer and deer), uses body shape, antler type (males), gait characteristics, and spatial location memory to distinguish individuals, supporting the removal of repetitive events within a 30-minute cooldown period. The automatic behavior quantification tool automatically extracts the reaction latency (from stimulus onset to initial alertness), alert duration, escape speed, escape trajectory curvature, grazing interruption time, and resumption of grazing time.
[0018] Preferably, the group behavior coordination assessment unit quantifies and records the following variables: group size (number of individuals visible in the frame, categorized as solid, small group, or large group); group alert synchronization rate (proportion of group members entering alert state and time difference, target: <2s when N=2, <4s when N=4); collective escape initiation delay (time from the start of the stimulus to the first individual starting to escape, and the time span from the first individual escaping to the last individual escaping, target: 2-3s when N=1, 4-6s when N=4); escape direction consistency (vector composite consistency index of the escape directions of group members, 1=completely consistent, 0=completely dispersed, target: >0.7); group splitting index (proportion of events where the group splits into ≥2 subgroups during the escape process, target: <10%).
[0019] Preferably, the feralization progress quantification tracking system includes an individual plasticity profile, a "suitable for release" type feralization progress index calculation unit, and a release achievement judgment engine. The "suitable for release" type feralization progress index constructs a three-dimensional evaluation space based on Botero's (2026) behavioral plasticity evolution theory: X-axis (behavioral buffer capacity) = (open area escape probability + concealed area escape probability) / 2, range 0-100%; Y-axis (strategy differentiation capacity) = |open area ambush / pursuit response ratio - concealed area ambush / pursuit response ratio|, range 0-1; Z-axis (human activity suitability) = 1 - |human sound escape probability - 37.5%| / 37.5%, range 0-1. The comprehensive feralization progress index (GPI) = (X × 0.4 + Y × 0.3 + Z × 0.3) × 100%, range 0-100. The release criteria assessment engine is based on a four-level GPI output: Excellent (GPI≥80, approved for immediate release without further support), Good (60≤GPI<80, approved for release, but requires a food supplementation station for 3-6 months for soft release support), Pass (40≤GPI<60, training extended for 30-60 days, strengthening weak dimensions, re-evaluation after retesting), and Fail (GPI<40, removed from the release candidate list, converted to captive breeding individuals or long-term captive display).
[0020] Preferably, the system further includes a historical behavior data retrieval module, a behavior network simulation module, an evaluation model configuration module, and a behavior data verification module. The historical behavior data retrieval module retrieves historical behavior records and partitions them according to the monitored behavior type and the ecological modality of herbivore rewilding training to obtain behavior measurement points. The behavior network simulation module performs graph neural network simulation based on the topology of behavior factor distribution during herbivore rewilding training, constructing a baseline value configuration unit. The input nodes are behavior factor distribution nodes, the input is the duration of behavior factor changes, the output nodes are behavior measurement points, and the output is the baseline value of the behavior index. The evaluation model configuration module integrates the average output of multiple baseline value configuration units based on the variance of the duration of changes of multiple behavior factors to construct a baseline value configuration model, processes the duration of changes of the multiple behavior factors, and obtains the baseline value of the behavior measurement point. The behavior data verification module performs delayed prompt rule data verification when the monitored behavior of the first measurement point is inconsistent with the baseline value of the behavior measurement point.
[0021] Preferably, the behavior data verification module includes a delay prompt rule unit, an anomaly judgment unit, and a continuous monitoring unit. The delay prompt rule unit is configured with a threshold for the duration of inconsistent states (e.g., 0.5 hours) and a threshold for the proportion of inconsistent state data (e.g., 20%). The anomaly judgment unit outputs an anomaly execution command when the monitored data meets either the duration threshold or the proportion threshold. The continuous monitoring unit continues to collect behavior monitoring data when the monitored data does not meet either the duration threshold or the proportion threshold.
[0022] Compared with the prior art, the beneficial effects of the present invention are: This automated response system for herbivore rewilding training, through the integration of multiple functional modules, enables precise monitoring, dynamic regulation, and coordinated optimization of various behavioral factors during herbivore rewilding training, creating a suitable environment for herbivores to survive in the wild. Among these modules, a dedicated herbivore motion detection module acquires herbivore motion signals in real time and triggers the intelligent control module based on actual activity levels. This makes behavioral testing more targeted and effectively maintains stable conditions that align with the natural behavioral state of herbivores, avoiding adverse effects on their behavioral responses due to human interference.
[0023] The rewilding-specific intelligent control module, based on motion detection signals and combined with the rewilding training stage and individual plasticity phenotype, dynamically configures current stimulus parameters and outputs stimulus initiation commands. This ensures that the stimulus intensity is adapted to the herbivore's adaptive needs, preventing overtraining and habituation in highly plastic individuals while maintaining training effectiveness for less plastic individuals, thus providing differentiated training programs for herbivores. The herbivore anti-predation behavior-specific stimulus library is comprehensively designed based on training objectives, incorporating various predator types and human activity sounds. The library structure is optimized according to the matching degree between stimulus types and herbivore response patterns, enabling holistic and precise control of stimulus types, maintaining the ecological authenticity of the stimulus, providing favorable conditions for behavioral response reconstruction, and creating a suitable stimulus environment for herbivore anti-predation behavior training.
[0024] The dual-habitat type comparative testing platform, by configuring open-area and concealed-area testing units, pre-calculates the differences in behavioral strategies under different habitat structures and adjusts the training focus based on the group behavior coordination assessment results. This proactively addresses potential adverse environmental conditions, preventing habitat homogenization or rigid strategies from undermining the effectiveness of rewilding training and ensuring the comprehensiveness of herbivore rewilding training. The rewilding progress quantitative tracking system extracts three-dimensional data on behavioral buffering capacity, strategy differentiation capacity, and human activity tolerance from the group coordination assessment results. It calculates a "suitable for release" type rewilding progress index and outputs a pass / fail judgment. This achieves synergistic optimization of multiple behavioral factors, including anti-predation behavior, habitat adaptability, group coordination, and human activity tolerance, breaking the limitations of traditional independent management of each factor and improving the overall suitability of rewilding training.
[0025] Through the collaborative work of its various modules, the entire system forms a complete system for assessing and regulating rewilding training. It can respond to changes in various behavioral factors in real time, dynamically adjust training strategies, and ensure that the rewilding training effect always meets the needs of herbivores for survival and reproduction in the wild. This helps to promote the stable growth of wild herbivore populations and provides a feasible technical solution for the intelligent management of the rewilding and release of endangered herbivores, thus promoting the development of wildlife conservation work in a more scientific and efficient direction.
Claims
1. An automated response system suitable for rewilding training of herbivores, characterized in that, The system includes: A herbivore-specific motion detection module acquires herbivore motion signals and triggers an intelligent control module; a rewilding-specific intelligent control module, based on the trigger signals from the motion detection module, dynamically configures stimulus parameters and outputs stimulus activation commands according to the rewilding training stage and individual plasticity phenotype; a herbivore anti-predation behavior-specific stimulus library includes a sub-library of ambush predators, a sub-library of pursuit predators, a sub-library of human activity, and a sub-library of companion animal control stimuli, all of which have been recorded in the wild, acoustically standardized, and behaviorally verified; a multimodal stimulus output module, based on the stimulus activation commands, outputs standardized stimuli in single or combined forms through sound stimulus sub-modules, odor stimulus sub-modules, and visual stimulus sub-modules; A dual-habitat type comparative testing platform includes an open-field testing unit and a concealed-field testing unit, used to simultaneously assess the strategy switching ability of herbivores in different habitat structures; a synchronous video recording and AI analysis module, based on the stimulus output of the stimulus output module, records behavioral responses throughout the process and automatically extracts behavioral variables through deep learning algorithms; a group behavior coordination assessment unit, based on the analysis results of the synchronous video recording and AI analysis module, quantifies and records group size, group alert synchronization rate, collective escape initiation delay, and consistency of escape direction; and a rewilding progress quantification tracking system, based on the assessment results of the group behavior coordination assessment unit, integrates three-dimensional data on behavioral buffering capacity, strategy differentiation capacity, and human activity appropriateness to calculate a "suitable for release" type rewilding progress index and generate herbivore rewilding training assessment results.
2. The automated response system for rewilding training of herbivores according to claim 1, characterized in that, The rewilding-specific intelligent control module includes: a rewilding stage identification unit, equipped with stage one, stage two, stage three, stage four, and stage five identifiers, corresponding to the emotional development period, anti-predation behavior reconstruction period, wild environment adaptation period, pre-release assessment period, and post-release monitoring period, respectively; and a stimulus intensity gradient adjustment unit, which, based on the behavioral plasticity "suitable for release" theory, dynamically adjusts stimulus parameters according to the plasticity phenotype of herbivores. Highly plastic individuals are given medium stimulus intensity, moderately plastic individuals are given standard stimulus intensity, and low-plasticity individuals are given high-intensity stimulus plus auxiliary reinforcement. The group coordination triggering unit identifies the number of individuals in the scene through AI video analysis and automatically classifies them into single individuals, small groups, and large groups. When ≥2 individuals are detected, the delay triggering time is extended, and the group alert synchronization rate and the consistency of escape direction are recorded.
3. The automated response system for rewilding training of herbivores according to claim 2, characterized in that, The herbivore anti-predation behavior specific stimulus library includes: an ambush predator stimulus sub-library, including leopard roars, tiger howls, and lynx calls, characterized by low frequency, short pulses, and high amplitude, simulating a "covert-outburst" type threat, used to test herbivores' long-range detection capabilities and rapid escape response in open areas; and a pursuit predator stimulus sub-library, including wolf howls and jackal calls, characterized by mid-frequency, continuous, and medium amplitude, simulating an "open-pursuit" type threat, used to test herbivores'... The study aimed to assess the intensity of the "superpredator" effect and the appropriate alertness response after training by evaluating the sustained vigilance and maneuvering strategies against group predators. A human activity stimulus library, including footsteps, voices, tool operation sounds, and dog barks with peak sound pressure levels of 75-80 dB, was also included to simulate common human disturbances within the protected area. A companion animal control stimulus library, including same-species alert calls, same-species communication calls, neutral bird calls, wind sounds, and flowing water sounds, was used to distinguish between specific anti-predation responses and general startle responses.
4. The automated response system for rewilding training of herbivores according to claim 3, characterized in that, The multimodal stimulation output module includes: a sound stimulation submodule, equipped with a standardized sound stimulation library, a remote waterproof speaker, and a sound calibration unit, with a frequency response range of 100Hz-20kHz and a maximum output sound pressure level ≥100dB, used to play the standardized sound stimulation in a field environment; an odor stimulation submodule, equipped with an odor release device, an odor sample library, and an odor diffusion control unit, used to release trace amounts of odor samples at preset time points and adjust the release amount and timing according to wind speed and wind direction sensor data; and a visual stimulation submodule, equipped with a movable mechanical model, an LED flashing array, and a projection device, used to simulate the simplified morphology and visual changes of predators or animals of the same species.
5. The automated response system for rewilding training of herbivores according to claim 4, characterized in that, The dual habitat type comparison test platform includes: an open area test unit, configured in forest gaps, floodplain meadows, or artificially cleared areas, with a field of view >100m, herbaceous vegetation height <30cm, and no shrub obstruction, used to test the "precision defense" strategy; a concealed area test unit, configured in dense shrublands, bamboo forests, or artificially planted tall grass areas, with shrub height 1.5-3.0m, coverage >60%, and a field of view <30m, used to test the "generalized fear" strategy; and a habitat switching test mode, which is synchronously triggered at adjacent open area-concealed area paired sites to record the differences in the responses of the same group in the two habitats and calculate the strategy switching index.
6. The automated response system for rewilding training of herbivores according to claim 5, characterized in that, The synchronous video recording and AI analysis module includes: a herbivore-specific behavior recognition model, based on the "Mask R-CNN" architecture, trained on typical herbivore postures, with a recognition accuracy of ≥95%; an individual re-identification model, for herbivores without significant markings, using body shape outline, angular shape, gait features, and spatial location memory to distinguish individuals, supporting the removal of recurring events during the cooldown period; and an automatic behavior quantification tool, which automatically extracts reaction latency, alert duration, escape speed, escape trajectory curvature, feeding interruption time, and resumption of feeding time.
7. The automated response system for rewilding training of herbivores according to claim 6, characterized in that, The group behavior coordination assessment unit quantifies and records the following variables: group size (number of visible individuals in the frame, categorized as solid, small group, and large group); group alert synchronization rate (proportion of individuals entering alert state and time difference); collective escape initiation delay (time from stimulus onset to the first individual starting to escape, and time span from the first individual escaping to the last individual escaping); escape direction consistency (vector composite consistency index of escape directions of group members); and group splitting index (proportion of events where the group splits into ≥2 subgroups during the escape process).
8. The automated response system for rewilding training of herbivores according to claim 7, characterized in that, The rewilding progress quantification tracking system includes: an individual plasticity profile, recording the basic behavioral assessment results, training history, test records, and assessment results of each released individual; a "suitable for release" type rewilding progress index calculation unit, which constructs a three-dimensional assessment space based on the Botero behavioral plasticity evolution theory, with the X-axis representing behavioral buffering capacity, the Y-axis representing strategy differentiation capacity, and the Z-axis representing the appropriateness of human activities, and comprehensively calculates the rewilding progress index; and a release compliance judgment engine, which outputs four levels of judgment—excellent, good, passable, and fail—based on the rewilding progress index, corresponding to immediate release, soft release support, extended training, and removal from the release list, respectively.
9. The automated response system for rewilding training of herbivores according to claim 8, characterized in that, The system also includes: a historical behavior data retrieval module, which retrieves historical behavior records and partitions them according to the monitored behavior type and the ecological modality of herbivore rewilding training to obtain behavior measurement points; and a behavior network simulation module, which performs graph neural network simulation based on the topology of behavior factor distribution in herbivore rewilding training, constructs a baseline value configuration unit, with the input node being the behavior factor distribution node, the input being the duration of behavior factor change, and the output node being the behavior measurement point, and the output being the baseline value of the behavior index. The evaluation model configuration module integrates the output mean of multiple benchmark configuration units based on the variance of the change duration of multiple behavioral factors, constructs a benchmark configuration model, processes the change duration of the multiple behavioral factors, and obtains the benchmark values of behavioral measurement points. The behavior data verification module performs delay prompt rule data verification when the monitored behavior of the first measurement point is inconsistent with the benchmark value of the behavior measurement point.
10. The automated response system for rewilding training of herbivores according to claim 9, characterized in that, The behavior data verification module includes: a delay prompt rule unit, which configures a threshold for the duration of inconsistent states and a threshold for the proportion of inconsistent state data; an anomaly judgment unit, which outputs an anomaly execution instruction when the monitored data meets the threshold for the duration of inconsistent states or the threshold for the proportion of inconsistent state data; and a continuous monitoring unit, which continues to collect behavior monitoring data when the monitored data does not meet the threshold for the duration of inconsistent states or the threshold for the proportion of inconsistent state data.