Wild animal multi-dimensional monitoring data acquisition method based on AI

Abstract

The invention belongs to the technical field of wild animal monitoring, and relates to an AI-based wild animal multi-dimensional monitoring data acquisition method, which comprises the following steps: S1, synchronously acquiring multi-dimensional monitoring data through a multi-source heterogeneous sensor network deployed in a monitoring area, the data including visual images, infrared thermal imaging, sound audio and environmental parameters; s2, carrying out preprocessing and fusion analysis on the collected data, and generating a comprehensive monitoring report containing species identification, position information, behavior categories and timestamps; according to the method, the proportion of an effective image (namely an image containing a clear target main body) is increased from less than 40% caused by delayed triggering and target blurring in a traditional mode to more than 70%, and unprecedented data support is provided for researching a complete behavior chain of animals; special behavior events can be automatically identified and labeled, and the workload of later manual screening is reduced; and meanwhile, multi-modal data are cooperatively acquired, so that more comprehensive scientific research information can be provided.

Description

Technical Field

[0001] This invention belongs to the field of wildlife monitoring technology, specifically an AI-based method for collecting multi-dimensional monitoring data of wildlife. Background Technology

[0002] With the increasing urgency of global biodiversity conservation needs and the convergence of IoT, edge computing, and artificial intelligence technologies, intelligent wildlife monitoring technology has become a key support for research in ecology, conservation biology, and animal behavior. Traditional monitoring methods, such as manual patrols and infrared-triggered cameras, are no longer sufficient to meet the modern scientific research demands for large-scale, all-weather, and highly detailed observations.

[0003] 1. Current status of technological development: Currently, mainstream automated monitoring systems typically employ a "multi-source heterogeneous sensor network" architecture. This involves deploying nodes within the monitoring area that include visible light cameras, infrared thermal imagers, environmental sensors (temperature, humidity, light intensity), and audio acquisition devices (pickups or microphone arrays). These systems collect multi-dimensional data, including visual, infrared, acoustic, and environmental data, and utilize deep learning-based image recognition (such as YOLO and ResNet) and sound classification models to automatically identify specific species, perform population statistics, and classify some basic behaviors (such as movement and feeding). This significantly reduces the burden of manual screening for researchers and increases the breadth of data collection.

[0004] 2. Core defects and bottlenecks in existing technologies: Although the above methods have achieved initial success, the following systemic defects have been exposed in actual large-scale deployment and long-term operation, which urgently need to be addressed: (1) High resource consumption and low efficiency: Most existing systems adopt a passive working mode of "24 / 7 operation" or "based on simple motion detection triggering". This extensive strategy leads to the sensor network being in a high power consumption state for a long time. A large amount of battery energy and network bandwidth are consumed in shooting empty scenes without animals and repeatedly recording the daily activities of common species. Research data shows that in typical fixed monitoring networks, more than 60% of energy consumption and storage space are occupied by "invalid data", while the capture rate of "key events" with high scientific research value, such as the behavior of rare species and interspecific interactions, is less than 10%. This causes serious waste of resources and directly limits the continuous deployment time of equipment in remote fields.

[0005] (2) Rigid perception and decision-making, lack of intelligent collaboration: The sensor nodes of the existing system usually operate as independent "data islands", and there is no information sharing and task collaboration between nodes. The system cannot predict the future activities of animals (such as movement trajectory and behavioral intentions) based on the captured information. For example, when the target leaves the monitoring area of ​​node A, the system cannot predict the direction it may go and wake up the downstream node B in advance to prepare, resulting in the interruption of valuable continuous observation sequence and difficulty in obtaining complete behavioral ecology data (such as complete predation chains and migration path fragments).

[0006] (3) Lack of a value-driven resource allocation mechanism: Current technologies generally treat monitoring data as a "homogeneous" information flow, failing to establish a quantitative "information value" evaluation system. When resources are limited (such as insufficient power or bandwidth congestion), the system cannot intelligently determine whether to prioritize high-definition recording of the courtship behavior of a critically endangered species or continue recording the ordinary activities of a large group of common animals. This lack of value judgment prevents valuable monitoring resources from being directed towards the most urgently needed and highest-output research goals, reducing the scientific effectiveness and application value of the entire monitoring network.

[0007] (4) Weak system adaptability and evolutionary capacity: Most existing system parameters (such as trigger sensitivity and sampling frequency) need to be preset manually. Once deployed, they are difficult to dynamically adjust according to environmental changes (such as seasonal changes and vegetation growth) or changes in animal behavior patterns. The system cannot learn and optimize its decision-making strategy from historical monitoring results, which may lead to a gradual decline in its long-term performance. Frequent manual intervention and parameter tuning are required, resulting in high operation and maintenance costs. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention proposes an AI-based method for multi-dimensional monitoring and data collection of wild animals. This invention primarily addresses the problems described in the background section.

[0009] The technical solution adopted by this invention to solve its technical problem is: an AI-based multi-dimensional monitoring data collection method for wild animals, comprising the following steps: S1: Multi-dimensional monitoring raw data of wild animals are collected synchronously or on demand through a multi-source heterogeneous sensor network deployed in the monitoring area; the multi-source heterogeneous sensor network includes, but is not limited to, visible light cameras, infrared thermal imagers, directional microphone arrays, temperature and humidity sensors, and GPS / BeiDou positioning modules; the multi-dimensional monitoring raw data includes at least visual images, infrared thermal images, sound audio, individual location, and environmental parameters. S2: The multidimensional monitoring raw data is preprocessed and feature-level fused using edge computing nodes or cloud servers; the preprocessing includes denoising, standardization, and timestamp alignment; the feature-level fusion is based on a deep learning model, extracting and associating cross-modal features to generate a structured comprehensive monitoring report; the report includes at least species identity, precise location, behavioral category, physiological state assessment, and corresponding timestamp; S3: Construct a spatiotemporal prediction model based on an attention mechanism, and predict the activity characteristics of the target individual or group within a future preset time window based on the historical activity sequence and current environmental context in the comprehensive monitoring report; the activity characteristics include the predicted trajectory spatial coordinates, possible behavioral states and their transition probabilities; S4: Establish a quantitative evaluation model. Based on the prediction results of step S3, dynamically evaluate the expected information value and estimated energy consumption of each sensor node in the network when performing various potential monitoring tasks, and calculate its value-energy consumption ratio as the core basis for scheduling decisions. S5: Construct a multi-objective optimization model with the core objective of maximizing the total monitoring value of the system and minimizing the total energy consumption of the system, and introduce bandwidth resource constraints and key area monitoring guarantee constraints; use a non-dominated sorting genetic algorithm with an elite strategy to solve the model and output a set of Pareto optimal sensor scheduling schemes under given resources. S6: Execute the selected optimal scheduling scheme and control the corresponding sensor nodes to work according to the scheme; at the same time, based on the difference between the actual results and the predictions of this round of monitoring, dynamically update the prediction model parameters and value assessment weights through the feedback optimization mechanism to realize the system's self-learning and adaptive optimization.

[0010] The activity feature prediction in step S3 specifically includes: S31: Constructing the spatiotemporal feature vector of animal activity:

[0011] Where s is the species feature vector. Let be the spatial coordinates at time t. For feature vectors, For environmental feature vectors, This represents the feature vector of historical activity patterns. S32: Using an attention-based time series prediction model for activity trajectory prediction:

[0012] Where T is the prediction time window and k is the historical observation window; S33: Calculate the probability of behavioral state transition:

[0013] in, This is a learnable reference matrix.

[0014] The value and energy consumption assessment in step S4 includes: S41: Define node n to execute tasks Expected information value function:

[0015] in: As a weight for species rarity,

[0016] Assign a score to the scarcity of behavior.

[0017] This is a data integrity indicator, reflecting the task. Data dimensional completeness that can be obtained To predict location confidence,

[0018] arrive These are adjustable weighting coefficients, and

[0019] S42: Energy consumption model for computational tasks:

[0020] in, Execute tasks for node n Power consumption Expected duration of the mission Energy consumption for data transmission S43: Value-to-Energy Ratio of Computing Nodes:

[0021] in, It is a small constant to prevent division by zero.

[0022] The multi-objective optimization model in step S5 is:

[0023] in: The binary decision variable represents whether node n executes the task. ; N is the total number of sensor nodes; T represents the total number of available tasks; For the task Bandwidth requirements; Total bandwidth constraint; C represents the set of key regions. Let c be the set of nodes covering the region. This represents the minimum monitoring and assurance level for region c.

[0024] The multi-objective optimization model is solved using a non-dominated sorting genetic algorithm with an elitist strategy. S51: Initialize the population, chromosome encoding is... ; S52: Calculate the Pareto frontier level and crowding distance for each individual; S53: Generate offspring through tournament selection, simulated binary crossover, and polynomial mutation; S54: Merge parent and child populations and perform fast non-dominated sorting; S55: Selecting a new generation of populations based on Pareto level and crowding; S56: Repeat S52-S55 until the maximum number of iterations is reached or convergence is achieved.

[0025] The feedback optimization mechanism in step S6 includes: S61: Define a measure of the difference between actual observed value and predicted value:

[0026] S62: Update the species activity pattern library:

[0027] in, Forgetting factor, For feature extraction functions; S63: Adjust the weight parameters of the valuation function:

[0028] in, This is the learning rate.

[0029] It also includes a dynamic adjustment strategy for the sensor's operating mode, including: when At this time, node n enters enhanced monitoring mode to improve sampling rate and resolution; when At that time, node n maintains the standard monitoring mode; when When this happens, node n enters energy-saving mode; in, and This is a preset threshold.

[0030] The beneficial effects of this invention are as follows: 1. In this invention, traditional threshold triggering or random sampling relies heavily on chance in capturing rare species and transient critical behaviors (such as courtship, parental care, and interspecies competition). This invention, through activity feature prediction in step S3, can predict "when and where high-value targets might appear"; through value assessment in step S4, it can identify "which target and behavior has the greatest scientific research value"; and finally, through optimized scheduling in step S5, it precisely delivers resources to the predicted high-value spatiotemporal points. The system is no longer "waiting for the rabbit to run into the tree stump" but "actively striking." For example, when the system learns that a rare bird often calls for mates in a specific valley at dawn, it will pre-schedule the visible light and audio sensors in that area to a high-sensitivity state, thereby systematically increasing the probability of capturing such critical events from less than 20% in traditional methods to over 70%; through trajectory tracking in step S3... The predictive model allows the system to estimate the possible path of a target after it leaves the current node's field of view. The optimization scheduling in step S5 uses this predictive information as a constraint to activate sensor nodes on potential downstream paths in advance, putting them into a ready or tracking state. This breaks down information silos between sensor nodes, forming a collaborative sensing network. The clarity of continuous imaging and the completeness of behavioral records for fast-moving targets (such as leopards hunting or migrating herds) can be improved by more than 60%. The proportion of effective images (i.e., images containing clear target subjects) has increased from less than 40% in the traditional mode due to delayed triggering and target blurring to more than 70%, providing unprecedented data support for studying the complete behavioral chain of animals. It can automatically identify and label special behavioral events, reducing the workload of manual screening in the later stage. At the same time, multimodal data collaborative acquisition can provide more comprehensive scientific research information.

[0031] 2. In this invention, the method significantly reduces overall energy consumption. Through predictive hibernation and task optimization, the overall system energy consumption can be reduced by 40%-60%. More resources are obtained in critical areas and during critical periods, while non-critical areas are intelligently energy-saving, extending the lifespan of field deployments. With the same battery capacity, the system's working time is extended by 1-1.5 times. Redundant data transmission is reduced, and bandwidth utilization is increased by 30%-50%. Edge computing reduces cloud transmission, lowers communication costs, and adaptively adjusts the data compression rate to balance quality and transmission overhead.

[0032] 3. In this invention, the system can continuously learn from actual monitoring results, thereby improving prediction accuracy over time. It can also adapt to environmental changes and shifts in animal behavior patterns, reducing the need for manual parameter tuning and lowering operation and maintenance costs. This invention provides a Pareto optimal solution set, allowing for the selection of different trade-off schemes based on actual conditions. Furthermore, the constraints are flexibly configurable, adapting to different deployment scenarios and research needs. The algorithm used has a fast convergence speed, making it suitable for sudden situations and online scheduling decisions. It does not require precise prediction of animal activity hotspots; the system can learn autonomously, thus reducing the requirements for sensor placement and lowering initial exploration costs. This invention supports the hybrid deployment of heterogeneous sensors, is compatible with existing equipment, provides visualized interpretations of scheduling decisions, facilitating ecologists' understanding of system behavior, supports remote strategy adjustments without on-site manual intervention, and possesses fault detection and self-recovery capabilities, improving system reliability. Attached Figure Description

[0033] The invention will now be further described with reference to the accompanying drawings.

[0034] Figure 1 This is a schematic diagram of the steps of the data acquisition method in this invention; Figure 2 This is a schematic diagram of step S3 in this invention; Figure 3 This is a schematic diagram of step S6 in this invention. Detailed Implementation

[0035] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0036] like Figures 1 to 3 As shown, the AI-based multi-dimensional monitoring data collection method for wild animals includes the following steps: S1: Multi-dimensional monitoring raw data of wild animals are collected synchronously or on demand through a multi-source heterogeneous sensor network deployed in the monitoring area; the multi-source heterogeneous sensor network includes, but is not limited to, visible light cameras, infrared thermal imagers, directional microphone arrays, temperature and humidity sensors, and GPS / BeiDou positioning modules; the multi-dimensional monitoring raw data includes at least visual images, infrared thermal images, sound audio, individual location, and environmental parameters. All data is precisely timestamped using a unified network time protocol. The advantage of this design is that it can reconstruct the "holographic" state of an animal at a specific moment. For example, by aligning the audio of a roar at a certain moment with an infrared image of an open-mouthed attack, it can be accurately identified as predatory behavior, greatly improving the accuracy and scientific rigor of behavioral analysis.

[0037] S2: The multidimensional monitoring raw data is preprocessed and feature-level fused using edge computing nodes or cloud servers; the preprocessing includes denoising, standardization, and timestamp alignment; the feature-level fusion is based on a deep learning model, extracting and associating cross-modal features to generate a structured comprehensive monitoring report; the report includes at least species identity, precise location, behavioral category, physiological state assessment, and corresponding timestamp; Implementing a model similar to the Trajectory Transformer offers the following advantages: 1) High prediction accuracy: It can learn complex long-distance dependencies, such as the fixed daily paths animals take from their dens to water sources; 2) Interpretability: By visualizing attention weights, ecologists can understand what historical information the model is based on for its predictions, increasing their trust in AI decision-making.

[0038] S3: Construct a spatiotemporal prediction model based on an attention mechanism, and predict the activity characteristics of the target individual or group within a future preset time window based on the historical activity sequence and current environmental context in the comprehensive monitoring report; the activity characteristics include the predicted trajectory spatial coordinates, possible behavioral states and their transition probabilities; S4: Establish a quantitative evaluation model. Based on the prediction results of step S3, dynamically evaluate the expected information value and estimated energy consumption of each sensor node in the network when performing various potential monitoring tasks, and calculate its value-energy consumption ratio as the core basis for scheduling decisions. The weights in the value function (such as the species rarity weight) can be initialized by ecologists. The core benefit of implementing this step is that it establishes a unified decision-making benchmark, enabling the system to automatically and objectively compare the current value of "monitoring the courtship dance of a rare bird" and "monitoring the feeding of a herd of common deer," thereby guiding resources toward high-value events.

[0039] S5: Construct a multi-objective optimization model with the core objective of maximizing the total monitoring value of the system and minimizing the total energy consumption of the system, and introduce bandwidth resource constraints and key area monitoring guarantee constraints; use a non-dominated sorting genetic algorithm with an elite strategy to solve the model and output a set of Pareto optimal sensor scheduling schemes under given resources. The NSGA-II algorithm is used for implementation. Its advantages are: 1) Providing flexibility: Decision-makers can choose a solution that prioritizes value or energy efficiency from the Pareto set, based on current battery capacity or the urgency of the research task. 2) Strong global search capability: It can effectively avoid getting trapped in local optima and find truly excellent scheduling solutions.

[0040] S6: Execute the selected optimal scheduling scheme and control the corresponding sensor nodes to work according to the scheme; at the same time, based on the difference between the actual results and the predictions of this round of monitoring, dynamically update the prediction model parameters and value assessment weights through the feedback optimization mechanism to realize the system's self-learning and adaptive optimization.

[0041] This is implemented using online learning or periodic fine-tuning strategies. The fundamental advantage lies in enabling the system to adapt and evolve. For example, when an animal changes its migration route due to seasonal changes, the system can quickly adjust its prediction model based on feedback from several prediction errors, adapting to the new activity patterns and thus maintaining high monitoring efficiency over the long term, reducing the costs of manual maintenance and remodeling.

[0042] The activity feature prediction in step S3 specifically includes: S31: Constructing the spatiotemporal feature vector of animal activity:

[0043] Where s is the species feature vector. Let be the spatial coordinates at time t. For feature vectors, For environmental feature vectors, This represents the feature vector of historical activity patterns. S32: Using an attention-based time series prediction model for activity trajectory prediction:

[0044] Where T is the prediction time window and k is the historical observation window; S33: Calculate the probability of behavioral state transition:

[0045] in, This is a learnable reference matrix.

[0046] The value and energy consumption assessment in step S4 includes: S41: Define node n to execute tasks Expected information value function:

[0047] in: As a weight for species rarity,

[0048] Assign a score to the scarcity of behavior.

[0049] This is a data integrity indicator, reflecting the task. Data dimensional completeness that can be obtained To predict location confidence,

[0050] arrive These are adjustable weighting coefficients, and

[0051] S42: Energy consumption model for computational tasks:

[0052] in, Execute tasks for node n Power consumption Expected duration of the mission Energy consumption for data transmission S43: Value-to-Energy Ratio of Computing Nodes:

[0053] in, It is a small constant to prevent division by zero.

[0054] The multi-objective optimization model in step S5 is:

[0055] in: The binary decision variable represents whether node n executes the task. ; N is the total number of sensor nodes; T represents the total number of available tasks; For the task Bandwidth requirements; Total bandwidth constraint; C represents the set of key regions. Let c be the set of nodes covering the region. This represents the minimum monitoring and assurance level for region c.

[0056] The multi-objective optimization model is solved using a non-dominated sorting genetic algorithm with an elitist strategy. S51: Initialize the population, chromosome encoding is... ; S52: Calculate the Pareto frontier level and crowding distance for each individual; S53: Generate offspring through tournament selection, simulated binary crossover, and polynomial mutation; S54: Merge parent and child populations and perform fast non-dominated sorting; S55: Selecting a new generation of populations based on Pareto level and crowding; S56: Repeat S52-S55 until the maximum number of iterations is reached or convergence is achieved.

[0057] The feedback optimization mechanism in step S6 includes: S61: Define a measure of the difference between actual observed value and predicted value:

[0058] S62: Update the species activity pattern library:

[0059] in, Forgetting factor, For feature extraction functions; S63: Adjust the weight parameters of the valuation function:

[0060] in, This is the learning rate.

[0061] It also includes a dynamic adjustment strategy for the sensor's operating mode, including: when At this time, node n enters enhanced monitoring mode to improve sampling rate and resolution; when At that time, node n maintains the standard monitoring mode; when When this happens, node n enters energy-saving mode; in, and This is a preset threshold.

[0062] It achieves "on-demand allocation" of energy. When a high-value target appears, the relevant nodes work at full capacity to capture high-quality data; when the monitoring value is average, it operates in standard mode; and when there is no target, it saves energy deeply. This eliminates energy waste at the micro level and is the direct technical reason for the system's overall energy consumption being reduced by 40%-60%.

[0063] Predictive scheduling is expected to increase the effective capture rate of rare species and special behaviors by 50%-80%; it enables "relay" continuous tracking, increasing the probability of obtaining complete behavioral sequences by more than 60%; it improves the imaging clarity of fast-moving targets, increasing the effective image ratio from less than 40% to more than 70%; and it establishes a quantitative value assessment system to ensure that high-value targets receive priority monitoring resources. It can automatically identify and label special behavioral events, reducing the workload of manual screening in the later stage; at the same time, multimodal data collaborative acquisition can provide more comprehensive scientific research information. Using this method, overall energy consumption can be significantly reduced. Through predictive hibernation and task optimization, the overall system energy consumption can be reduced by 40%-60%. More resources are obtained in critical areas and during critical periods, while intelligent energy saving is achieved in non-critical areas, extending the lifespan of field deployments. With the same battery capacity, the system working time is extended by 1-1.5 times. Redundant data transmission is reduced, and bandwidth utilization is improved by 30%-50%. Cloud transmission is reduced through edge computing, communication costs are lowered, and data compression rate is adaptively adjusted to balance quality and transmission overhead. The system can continuously learn from actual monitoring results, thereby improving prediction accuracy over time. It can also adapt to environmental changes and shifts in animal behavior patterns, reducing the need for manual parameter tuning and lowering operation and maintenance costs. This invention provides a Pareto optimal solution set, allowing the selection of different trade-off schemes based on actual conditions. Furthermore, the constraints are flexibly configurable, adapting to different deployment scenarios and research needs. The algorithm used has a fast convergence speed, making it suitable for sudden situations and online scheduling decisions. It does not require precise prediction of animal activity hotspots, as the system can learn autonomously, thus reducing the requirements for sensor deployment locations and lowering initial exploration costs. This invention supports the hybrid deployment of heterogeneous sensors, is compatible with existing equipment, provides a visualized interpretation of scheduling decisions, facilitates ecologists' understanding of system behavior, supports remote strategy adjustments without on-site manual intervention, and possesses fault detection and self-recovery capabilities, improving system reliability.

[0064] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An AI-based method for multi-dimensional monitoring and data collection of wild animals, characterized in that, Includes the following steps: S1: Multi-dimensional monitoring data is collected synchronously through a multi-source heterogeneous sensor network deployed in the monitoring area. The data includes visual images, infrared thermal imaging, sound audio, and environmental parameters. S2: Preprocess and fuse the collected data to generate a comprehensive monitoring report that includes species identification, location information, behavioral categories, and timestamps; S3: Based on historical activity patterns and current environmental context, predict the activity characteristics of target individuals or groups in the future time period; S4: Based on the prediction results, evaluate the value and energy consumption of each sensor node in performing different monitoring tasks; S5: Under system resource constraints, generate the optimal sensor scheduling scheme through multi-objective optimization; S6: Execute the scheduling scheme and optimize it based on the actual monitoring results.

2. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 1, characterized in that: The activity feature prediction in step S3 specifically includes: S31: Constructing the spatiotemporal feature vector of animal activity: Where s is the species feature vector. Let be the spatial coordinates at time t. For feature vectors, For environmental feature vectors, This represents the feature vector of historical activity patterns. S32: Using an attention-based time series prediction model for activity trajectory prediction: Where T is the prediction time window and k is the historical observation window; S33: Calculate the probability of behavioral state transition: in, This is a learnable reference matrix.

3. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 2, characterized in that: The value and energy consumption assessment in step S4 includes: S41: Define node n to execute tasks Expected information value function: in: As a weight for species rarity, Scoring for behavioral scarcity This is a data integrity indicator, reflecting the task. Data dimensional completeness that can be obtained To predict location confidence, arrive These are adjustable weighting coefficients, and S42: Energy consumption model for computational tasks: in, Execute tasks for node n Power consumption Expected duration of the mission Energy consumption for data transmission S43: Value-to-Energy Ratio of Computing Nodes: in, It is a small constant to prevent division by zero.

4. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 3, characterized in that: The multi-objective optimization model in step S5 is: in: The binary decision variable represents whether node n executes the task. ; N is the total number of sensor nodes; T represents the total number of available tasks; For the task Bandwidth requirements; Total bandwidth constraint; C represents the set of key regions. Let c be the set of nodes covering the region. This represents the minimum monitoring and assurance level for region c.

5. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 4, characterized in that: The multi-objective optimization model is solved using a non-dominated sorting genetic algorithm with an elitist strategy. S51: Initialize the population, chromosome encoding is... ; S52: Calculate the Pareto frontier level and crowding distance for each individual; S53: Generate offspring through tournament selection, simulated binary crossover, and polynomial mutation; S54: Merge parent and child populations and perform fast non-dominated sorting; S55: Selecting a new generation of populations based on Pareto level and crowding; S56: Repeat S52-S55 until the maximum number of iterations is reached or convergence is achieved.

6. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 5, characterized in that: The feedback optimization mechanism in step S6 includes: S61: Define a measure of the difference between actual observed value and predicted value: S62: Update the species activity pattern library: in, Forgetting factor, For feature extraction functions; S63: Adjust the weight parameters of the valuation function: in, This is the learning rate.

7. The AI-based multi-dimensional monitoring data collection method for wild animals according to claim 6, characterized in that: It also includes a dynamic adjustment strategy for the sensor's operating mode, including: when At this time, node n enters enhanced monitoring mode to improve sampling rate and resolution; when At that time, node n maintains the standard monitoring mode; when When this happens, node n enters energy-saving mode; in, and This is a preset threshold.

Images