A wild animal intelligent monitoring system and method based on multi-source information fusion

Abstract

The application relates to the technical field of intelligent monitoring and ecological protection, and discloses a wild animal intelligent monitoring system and method based on multi-source information fusion, which comprises the following steps: an information perception module extracts event features at an edge node, generates a delay event record when resources are limited, and reports the record at an appropriate opportunity; a risk estimation module constructs an ecological patrol resource scheduling graph based on node basic data, establishes a double-mode compensation income matrix, and calculates a monitoring continuity risk value; a compensation decision module retrieves a compensation candidate node when a risk is triggered, generates an instruction to adjust the monitoring parameters of the candidate node according to the expected resource consumption; and a work order reconstruction module calculates a compensation consumption increment and superimposes the increment on an artificial maintenance priority to generate a secondary state patrol work order. The application can dynamically adjust the monitoring strategy of surrounding nodes to make up for local coverage loss, guarantee the monitoring network continuity in a resource-limited environment, and improve the accuracy of artificial operation and maintenance scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and ecological protection technology, specifically to an intelligent wildlife monitoring system and method based on multi-source information fusion. Background Technology

[0002] Currently, wildlife monitoring mainly relies on edge monitoring nodes and sensor networks deployed in the wild environment for data collection. However, because the target monitoring areas are typically located in remote areas, they generally face objective problems such as power supply difficulties and lack of broadband network coverage.

[0003] Existing wildlife monitoring systems often experience monitoring blind spots when individual nodes cease operation due to power depletion, insufficient storage space, or hardware failure. In such cases, the system typically relies on manual maintenance, disrupting the continuity of the entire monitoring network during the lull before personnel arrive. Furthermore, conventional field monitoring equipment often uses fixed sampling frequencies and trigger thresholds, lacking effective data collaboration and status awareness mechanisms between adjacent nodes. When coverage gaps occur in localized areas, surrounding normally functioning nodes cannot proactively adjust their monitoring parameters based on terrain resistance and their remaining available resources to fill these gaps.

[0004] Furthermore, existing manual maintenance scheduling methods largely rely on fixed inspection cycles or single fault alarm signals. When generating patrol tasks, they lack comprehensive consideration of regional ecological value, accessibility caused by natural terrain, and the additional resources consumed by nodes to compensate for blind spots. This leads to a disconnect between maintenance work order scheduling and changes in the actual environment, making it difficult to achieve the most efficient restoration of the monitoring network under conditions where both human and equipment resources are limited.

[0005] Therefore, this invention proposes an intelligent wildlife monitoring system and method based on multi-source information fusion to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an intelligent wildlife monitoring system and method based on multi-source information fusion. This solves the problems of monitoring continuity interruption caused by the lack of a collaborative compensation mechanism between adjacent nodes when local nodes fail in existing wildlife monitoring systems, as well as the lack of objective data basis for manual operation and maintenance scheduling.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent wildlife monitoring system based on multi-source information fusion, comprising: The information perception module, deployed on the edge monitoring node, is used to extract target event features and output an event evidence completeness score; when the edge monitoring node meets the resource-constrained conditions, it generates a delayed event record and writes it into the delay-tolerant event queue, and reports the delayed event record when the link recovery conditions are met; The risk prediction module, deployed in the cloud-based scheduling and management system, is used to construct an ecological patrol resource scheduling map based on the basic data of the edge monitoring nodes; establish a dual-mode compensation benefit matrix; and calculate the monitoring continuity risk value of the edge monitoring nodes. The compensation decision module, deployed in the cloud-based scheduling and management system, is used to retrieve candidate compensation nodes in the dual-mode compensation benefit matrix when the monitored continuity risk value reaches the risk trigger threshold; generate compensation level instructions based on the expected compensation resource consumption and send them to the candidate compensation nodes to adjust the monitoring parameters of the candidate compensation nodes. The work order reconstruction module, deployed in the cloud-based scheduling and management system, is used to record the node operating resources occupied by the candidate compensation nodes and calculate the compensation consumption increment; the compensation consumption increment is added to the basic manual maintenance priority, and a secondary patrol work order is generated based on the updated manual maintenance priority.

[0008] Preferably, the information perception module extracts target event features and outputs an event evidence completeness score, specifically used for: The classification confidence, temporal consistency, spatial prior matching degree, and physical trigger signal matching degree are extracted as multi-dimensional feature components; the data validity flag bits of the corresponding sensors for each multi-dimensional feature component are checked; if a feature component is found to be missing, the remaining valid feature components are redistributed according to the relative proportion of their original weights; the multi-dimensional feature components and the redistributed weights are weighted and summed to generate the event evidence completeness score.

[0009] Preferably, the resource-constrained conditions include weak network conditions (uplink network bandwidth is lower than the network bandwidth threshold), low power conditions (remaining available power is lower than the power alarm threshold), and substandard conditions (event evidence completeness score is lower than the immediate reporting threshold); the delayed event record includes an event occurrence timestamp and a validity period parameter. The information perception module performs data reporting when the data recycling conditions are met. Specifically, it is used to: obtain the difference between the event occurrence timestamp of each delayed event record in the delayed tolerance event queue and the current system time; delete delayed event records whose difference is greater than the validity period parameter; and sort the unexpired delayed event records in descending order according to the event evidence completeness score and report them in sequence.

[0010] Preferably, the basic data includes the spatial coordinates of the edge monitoring nodes, patrol road topology data, and node operating status parameters; the dual-mode compensation benefit matrix includes a spatial coverage compensation benefit sub-matrix and a corridor sampling compensation benefit sub-matrix. The risk prediction module establishes a spatial coverage compensation benefit sub-matrix, specifically used for: extracting a failed node and an adjacent node with connectivity in the ecological patrol resource scheduling graph; calculating the replaceable coverage ratio of the adjacent node relative to the failed node; extracting the remaining available power and local storage space of the adjacent node and weighted summing them to obtain the equipment availability score; obtaining the estimated compensation resource consumption for the adjacent node to perform compensation actions; linearly weighting the replaceable coverage ratio, the equipment availability score, and the estimated compensation resource consumption, and generating spatial coverage compensation benefit values ​​after interval pruning, thus constructing the spatial coverage compensation benefit sub-matrix.

[0011] Preferably, the risk prediction module establishes a corridor sampling compensation benefit sub-matrix, specifically used for: A comprehensive resistance surface is constructed by acquiring slope values ​​and surface cover type resistance values ​​from the digital elevation model; the cumulative passage resistance cost from the failure node to the adjacent node along the natural terrain is calculated using a minimum cost path search algorithm; the cumulative passage resistance cost is mapped to the probability of the target passing through the adjacent node using a negative exponential decay function; the passage probability, the equipment availability score, and the estimated compensation resource consumption are linearly weighted and calculated, and after interval pruning, the corridor sampling compensation benefit value is generated, thus constructing the corridor sampling compensation benefit submatrix.

[0012] Preferably, the risk prediction module calculates the monitoring continuity risk value of the edge monitoring node, specifically for: The comprehensive ecological value is determined by combining the historical occurrence frequency of species in the target area under normal conditions and the protection level weight of target species under the activation state of sudden events. The battery decay rate and memory bad block ratio of the edge monitoring nodes are extracted and weighted to obtain the decay weighted sum value. The node operation redundancy is obtained by subtracting the decay weighted sum value from the value. The maximum compensation benefit value in the dual-mode compensation benefit matrix is ​​extracted and the coverage gap coefficient is calculated by subtracting the maximum compensation benefit value from the value. The path slope integral value and rainfall parameters are obtained to generate the comprehensive environmental resistance value. The comprehensive environmental resistance value is input into the logistic function for mapping to obtain the probability of manual patrol interruption. The comprehensive ecological value, the difference term obtained by subtracting the node operation redundancy from the value, the coverage gap coefficient, and the probability of manual patrol interruption are jointly multiplied to output the monitoring continuity risk value.

[0013] Preferably, the compensation decision module generates compensation level instructions, specifically for: Obtain a first ratio of the energy consumption increment required to perform the compensation action to the remaining available power of the compensation candidate node, and a second ratio of the storage occupation increment to the remaining available local storage space. Add positive constant smoothing terms as denominators to the remaining available power and the remaining available local storage space to participate in the division operation to obtain the first ratio and the second ratio, and calculate the expected compensation resource consumption by weighting. When the expected compensation resource consumption is less than or equal to a first cost threshold, generate a first compensation level instruction. When the expected compensation resource consumption is greater than the first cost threshold and less than or equal to the second cost threshold, generate a second compensation level instruction.

[0014] Preferably, the compensation decision module adjusts the monitoring parameters through the compensation level instruction, specifically for: When the candidate node receives the first compensation level instruction, it increases the image sampling frequency, decreases the analog voltage comparison threshold of the passive infrared sensor, and increases the reporting priority of the captured data. When the candidate node receives the second compensation level instruction, it adjusts the monitoring direction parameter to cover the missing area, maintains the basic sampling frequency, increases the analog voltage comparison threshold of the passive infrared sensor, and limits the reporting priority of the image file. It starts the built-in timer, and after the set execution time window expires, it resets the monitoring parameters back to the normal baseline value.

[0015] Preferably, the work order reconstruction module generates a secondary patrol work order, which is further used for: The system acquires the maximum transmit power, receive sensitivity parameters, and vegetation attenuation coefficient of the wireless LAN communication module. It then inputs the logarithmic distance path loss model to calculate the physical communication distance boundary to generate the effective wireless coverage area. The endpoint constraint of the secondary patrol route is set as the effective wireless coverage area. The system receives structured real-valued data submitted by manual patrols, including floating-point values ​​of latitude and longitude deviations, species occurrences, and vegetation shading conditions. It then uses a moving average algorithm to weight and fuse historical parameter values ​​with reference parameter values ​​calculated based on the structured real-valued data, and updates the feature aggregation weights and the initial parameters of the dual-mode compensation benefit matrix.

[0016] This invention also provides a method for intelligent monitoring of wild animals based on multi-source information fusion, comprising the following steps: Extract features of the target event and output an event evidence completeness score; When the edge monitoring node meets the resource-constrained conditions, a delay event record is generated and written into the delay-tolerant event queue, and the delay event record is reported when the link recovery conditions are met; An ecological patrol resource scheduling diagram is constructed based on the basic data of the edge monitoring nodes; Establish a dual-mode compensation benefit matrix and calculate the monitoring continuity risk value of the edge monitoring node; When the monitored continuity risk value reaches the risk trigger threshold, a compensation candidate node is retrieved in the dual-mode compensation benefit matrix, a compensation level instruction is generated based on the expected compensation resource consumption and sent to the compensation candidate node to adjust the monitoring parameters of the compensation candidate node. Record the node operation resources occupied by the candidate compensation nodes and calculate the compensation consumption increment. Add the compensation consumption increment to the basic manual maintenance priority and generate a secondary patrol work order based on the updated manual maintenance priority.

[0017] This invention provides an intelligent wildlife monitoring system and method based on multi-source information fusion. It has the following beneficial effects: 1. This invention constructs a closed-loop system from edge data acquisition and cloud-based dynamic scheduling to manual on-site maintenance through the coordinated operation of various modules. During system operation, the data reporting strategy can be dynamically adjusted based on the resource-constrained status of edge nodes, and the monitoring continuity risk is assessed in the cloud based on basic data and a dual-mode compensation benefit matrix. When a local node triggers a risk threshold, the system automatically issues instructions to adjust the monitoring parameters of candidate nodes to compensate for coverage gaps, and incorporates the additional resource consumption generated by compensation into the priority of manual maintenance. This collaborative processing mechanism ensures the overall continuity of the monitoring network in field environments with limited equipment resources and provides objective data-driven basis for manual operation and maintenance.

[0018] 2. The dual-mode compensation benefit matrix constructed in this invention comprehensively considers the compensation feasibility of two physical dimensions: spatial coverage and corridor sampling. By combining the alternative coverage ratio of adjacent nodes, equipment availability scores, estimated compensation resource consumption, and target passage probability calculated based on natural terrain resistance, the system can quantitatively evaluate the actual benefits of compensation actions under multi-dimensional environmental constraints. Combined with the monitoring continuity risk value calculated from environmental resistance and node operational redundancy, this mechanism enables the system to make scheduling decisions that conform to geographical characteristics and equipment status when executing node compensation tasks, reducing the risk of resource depletion of adjacent nodes due to blind deployment.

[0019] 3. The compensation decision module of this invention adopts a hierarchical parameter adjustment strategy based on estimated resource consumption when executing instructions. The system generates compensation instructions of different levels by calculating the incremental ratio of energy to storage usage, and introduces a positive constant smoothing term in the division operation to improve the computational stability of the underlying algorithm. After receiving instructions of different levels, the edge candidate nodes can adjust parameters such as image sampling frequency, analog voltage comparison threshold, or monitoring direction in a targeted manner, and automatically reset after a set time window. This hierarchical control and timed recovery mechanism ensures the temporary monitoring range while taking into account the long-term endurance and storage capacity of the compensation nodes themselves.

[0020] 4. This invention introduces physical communication boundary constraints and a closed-loop correction mechanism for field parameters when generating sub-state patrol work orders. The system combines wireless communication module parameters and vegetation attenuation coefficients to set the effective wireless coverage area as the patrol endpoint constraint, improving the success rate of field data retrieval. Simultaneously, the system can receive structured real-label data from manual patrol field feedback, and uses a moving average algorithm to weight and fuse historical parameters with reference parameters, achieving continuous updates to feature aggregation weights and initial parameters of the compensation matrix. This allows the monitoring system's operating model to continuously adapt to the real environment as the deployment time progresses. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system architecture according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the ecological patrol resource scheduling diagram and the dual-mode compensation benefit matrix construction in an embodiment of the present invention; Figure 4 This is a schematic diagram of the edge multi-source sensing fusion and latency-tolerant event queue management process according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the joint calculation of dynamic ecological value and monitoring continuity risk in an embodiment of the present invention. Figure 6 This is a schematic diagram of the closed loop of dual-mode compensation decision-making, maintenance priority recharge, and work order reconstruction in an embodiment of the present invention. Figure 7 This is a heat map showing the scheduling of ecological patrol resources and the distribution of risks according to an embodiment of the present invention; wherein, (a) is a schematic diagram of the global risk distribution in the early stage of local node anomalies; and (b) is a schematic diagram of the network coverage reconstruction status after cross-node compensation is performed. Figure 8 The graphs are comparison curves of monitoring continuity and energy consumption under continuous operation of the system according to an embodiment of the present invention; wherein, (a) is a comparison curve of system performance over a 6-month operating cycle; and (b) is a dynamic curve of monitoring continuity risk after a local anomaly occurs.

[0022] Among them, 100 is the information perception module; 200 is the risk prediction module; 300 is the compensation decision module; and 400 is the work order reconstruction module. Detailed Implementation

[0023] The technical solutions in 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Reference Figure 1 The present invention provides an intelligent wildlife monitoring system based on multi-source information fusion, comprising an information perception module 100, a risk prediction module 200, a compensation decision module 300, and a work order reconstruction module 400.

[0025] The information sensing module 100 is deployed in each edge monitoring node within the monitoring area. The information sensing module 100 initiates data fusion processing after the underlying environmental sensors capture physical trigger signals, extracts target event features, and outputs an event evidence completeness score. The information sensing module 100 is responsible for generating delayed event records with event timestamps and validity period parameters when the node status meets resource-constrained conditions, managing the latency-tolerant event queue configured in the local storage area, and performing data reporting when link recovery conditions are met.

[0026] The risk prediction module 200 is deployed in the cloud-based scheduling and management system. Based on the spatial coordinates of edge monitoring nodes, digital elevation model data, patrol road topology data, and operational status data, the risk prediction module 200 constructs a global ecological patrol resource scheduling map. The risk prediction module 200 is responsible for establishing a dual-mode compensation benefit matrix, including a spatial coverage compensation benefit sub-matrix and a corridor sampling compensation benefit sub-matrix, and calculating the current ecological value and monitoring continuity risk value of target nodes by combining the returned data.

[0027] The compensation decision module 300 is deployed in the cloud-based scheduling and management system. The compensation decision module 300 receives continuous risk values ​​and compares them with set risk trigger thresholds. The compensation decision module 300 is responsible for retrieving candidate compensation nodes in the dual-mode compensation benefit matrix, generating compensation level instructions of different levels based on the expected compensation resource consumption, and issuing control instructions to the candidate nodes to adjust monitoring parameters.

[0028] The work order reconstruction module 400 is deployed in the cloud-based scheduling and management system. This module records the node operating resources occupied by each candidate node executing compensation instructions and calculates the incremental compensation consumption. The work order reconstruction module 400 is responsible for proportionally adding the incremental compensation consumption to the node's basic manual maintenance priority, generating and issuing secondary patrol work orders based on the updated manual maintenance priority and patrol path edge weights, and simultaneously receiving on-site verification tag data to modify feature calculation weight parameters.

[0029] Reference Figure 2 This invention provides a method for intelligent monitoring of wild animals based on multi-source information fusion, comprising the following steps: S100, construct a global ecological patrol resource scheduling map, and set a vertex set, patrol path edge set, node collaboration edge set and node operation status set in the ecological patrol resource scheduling map, and establish a dual-mode compensation benefit matrix including spatial coverage compensation benefit sub-matrix and corridor sampling compensation benefit sub-matrix; S200: After capturing the physical trigger signal at the edge, it starts data fusion processing, extracts the multi-dimensional feature components of the target event and calculates the event evidence completeness score, generates delayed event records based on the event evidence completeness score and node state parameters and writes them into the delay tolerance event queue, and performs data reporting when the triggering conditions are met. S300 receives the returned data and operation status messages, calculates the current ecological value, operation redundancy, coverage gap coefficient and probability of manual patrol interruption of the target node by combining the node attributes, and performs joint calculation on the above parameters to output the monitoring continuity risk value of the target node. S400, when the monitored continuous risk value reaches the risk trigger threshold, the candidate compensation node is retrieved in the dual-mode compensation benefit matrix, and a corresponding compensation level instruction is generated and issued based on the expected compensation resource consumption, so that the candidate compensation node adjusts the monitoring parameters according to the compensation level instruction. S500 records the node operation resources occupied by the candidate compensation nodes and calculates the compensation consumption increment. It adds the compensation consumption increment to the basic manual maintenance priority, generates and issues the secondary patrol work order based on the updated manual maintenance priority, and receives on-site verification tag data to perform closed-loop correction of system parameters.

[0030] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0031] Reference Figure 3 Step S100 is executed by the risk prediction module 200, which abstracts the ecological patrol and monitoring network into a computable graph topology model. Specifically, it includes the following sub-steps: S101, acquire basic data of the ecological patrol and monitoring network. Basic data includes the spatial coordinates of edge monitoring nodes, patrol road topology data, and node operating status parameters. To ensure time alignment of multi-source heterogeneous data and avoid time-series misalignment leading to graph state calculation errors, the risk prediction module 200 uses the network time protocol to perform timestamp synchronization and unified truncation processing on all types of received messages.

[0032] After synchronization is complete, the risk prediction module 200 receives latitude, longitude, and altitude data obtained by the built-in positioning components reported by each edge monitoring node through an interface, using these as spatial coordinates. Simultaneously, the system retrieves patrol road topology data from road network files in the geographic information system.

[0033] In addition, by analyzing the heartbeat messages periodically sent by each edge monitoring node, the remaining available power, local storage space, and current working / sleep duty cycle of the node are obtained. Among them, the working / sleep duty cycle refers to the proportion of time a node is in an active state within one working cycle, and its value is usually in the range of [0.05, 0.3] to balance energy consumption and sensing sensitivity.

[0034] S102, construct an ecological patrol resource scheduling graph based on spatial coordinates and patrol road topology data. As a preferred graph structure mapping method, the risk prediction module 200 sets a vertex set, a patrol path edge set, a node collaboration edge set, and a node operation status set in the ecological patrol resource scheduling graph.

[0035] Specifically, each edge monitoring node is instantiated as a node vertex element in the vertex set, and the current location of the patrol personnel is instantiated as a personnel location vertex element in the vertex set. When there is a reported ecological event area, the ecological event area is instantiated as an event area vertex element. The physical access roads connecting each vertex are mapped to path edge elements in the patrol path edge set. The actual physical travel distance between two nodes is set as the edge weight of the corresponding path edge, which reflects the travel cost for the patrol personnel to reach the corresponding node or event area.

[0036] Next, the wireless communication link capability between adjacent edge monitoring nodes is evaluated. If the wireless signal strength between two nodes is greater than a preset signal strength threshold, a cooperative edge element for the corresponding vertex is established in the node cooperative edge set. This signal strength threshold is calibrated by the receiving sensitivity parameter of the edge monitoring node's wireless module, typically set between -85dBm and -75dBm, and represents the minimum received signal power limit required to maintain stable handshake transmission of basic data packets between nodes. The extracted node operating state parameters are then used as attribute parameters for the corresponding vertex and written into the node operating state set.

[0037] S103, Calculate the spatial coverage compensation benefit submatrix. The technical purpose of this matrix is ​​to measure the ability of adjacent nodes to supplement or replace the original monitoring gaps of failed nodes by adjusting their own monitoring parameters. To eliminate errors caused by different physical dimensions, all parameters involved in the calculation are pre-normalized to the dimensionless interval [0,1] through maximum and minimum values.

[0038] For any two edge monitoring nodes connected in the node collaboration edge set, calculate their replaceable coverage ratio, device availability score, and estimated compensation resource consumption. The replaceable coverage ratio is the ratio of the spatial overlap area between the adjacent nodes after they are adjusted to their maximum monitoring field of view and the original monitoring area of ​​the failed node. The device availability score is obtained by weighted summing the normalized remaining available power and local storage space of the adjacent nodes. The estimated compensation resource consumption is the calculated value of the increased power loss and storage or processing resource consumption expected when the adjacent nodes perform compensation actions.

[0039] Therefore, the initial value of the spatial coverage compensation benefit of adjacent nodes to the failed node is obtained according to the following formula. After the initial value of the spatial coverage compensation benefit is processed by interval pruning, the spatial coverage compensation benefit value is obtained. The interval pruning process means that when the calculated value is less than 0, it is taken as 0, and when the calculated value is greater than 1, it is taken as 1, so that the spatial coverage compensation benefit value is kept within the interval [0,1]. ; In the formula, Indicates adjacent nodes For failed nodes Spatial coverage compensation benefit value; This represents the normalized percentage of substitutable coverage. Indicates adjacent nodes Equipment availability rating; This represents the normalized estimated consumption of compensatory resources.

[0040] These are the weighting coefficients representing coverage gain, health constraint, and energy consumption penalty, respectively, and their values ​​satisfy... and .

[0041] If adjacent nodes are extracted If the remaining battery power is below a preset safety threshold (e.g., set to 5% to 10% of the full battery capacity), the device's availability score will be directly applied. Set it to 0 to block the assignment of compensation in that direction. All the calculated spatial cover compensation benefit values ​​together constitute the spatial cover compensation benefit submatrix.

[0042] S104, calculates the corridor sampling compensation benefit sub-matrix. When the proportion of directly replaceable spatial coverage is limited or the field of view is obstructed, relying solely on spatial relationships cannot effectively replace the monitoring function. In this case, the system introduces the spatiotemporal movement characteristics of species and evaluates the temporal sampling compensation capability in the time dimension by calculating this sub-matrix.

[0043] When calculating the probability of a target reaching an adjacent node by traversing a monitored but missing area, the system establishes a probability estimation model based on physical terrain parameters. Specifically, the process involves: extracting slope values ​​and surface cover type resistance values ​​from the digital elevation model to construct a comprehensive resistance surface; the surface cover type resistance values ​​are determined by a preset resistance table, with resistance values ​​for different cover types ranging from [0,1] after normalization, where open areas, roads, or low grasslands correspond to lower resistance values, while dense shrublands, steep woodlands, or water edges correspond to higher resistance values. Then, using the area where the failed node is located as the starting point, a minimum cost path search algorithm is used to calculate the path to adjacent nodes along the natural terrain. The cumulative road resistance cost; ultimately, a negative exponential decay function is adopted (its mapping form is...). ,in, To accumulate traffic resistance costs, To characterize the attenuation coefficient of a species’ sensitivity to terrain resistance (usually set between 0.1 and 0.5), the cumulative passage resistance cost is mapped to the passage probability in the interval [0,1].

[0044] Based on this, the initial value of the corridor sampling compensation benefit is calculated according to the following formula. After interval pruning, the corridor sampling compensation benefit value is obtained. Interval pruning means that when the calculated value is less than 0, it is taken as 0, and when the calculated value is greater than 1, it is taken as 1, so that the corridor sampling compensation benefit value is kept within the interval [0,1]. ; In the formula, Indicates adjacent nodes For failed nodes The corridor sampling compensation benefit value; This indicates that the calculated target passes through adjacent nodes. The probability of passage. , , These are the weighting coefficients participating in the corridor sampling compensation calculation, and their values ​​satisfy... ,and The revenue values ​​of nodes with connectivity in the node collaboration edge set collectively constitute the corridor sampling compensation revenue submatrix.

[0045] S105, Combine and generate the dual-mode compensation benefit matrix. The risk prediction module 200 combines the previously calculated spatial coverage compensation benefit sub-matrix with the corridor sampling compensation benefit sub-matrix to generate the final dual-mode compensation benefit matrix. This matrix is ​​persistently stored as a graph attribute parameter of the ecological patrol resource scheduling map and provides a data call interface to the backend management service, providing underlying data support for the system's subsequent dynamic decision-making, monitoring of continuous risks, and work order assignment.

[0046] Reference Figure 4 Step S200 is executed by the information sensing module 100, which is used to complete the initial data purification and reduce the power consumption of invalid communication at the edge. The specific implementation process includes the following sub-steps: S201: After the underlying environmental sensors capture the physical trigger signal, data fusion processing is initiated to extract multi-dimensional feature components of the target event. In wildlife monitoring scenarios, the underlying environmental sensors mainly include passive infrared detectors and audio acquisition components. When the physical trigger signal exceeds the hardware noise floor level, the information perception module 100 wakes up the dormant image acquisition device to acquire the on-site image stream. To comprehensively evaluate the authenticity of the trigger event, the system simultaneously extracts feature components in four dimensions: classification confidence, temporal consistency, spatial prior matching degree, and physical trigger signal matching degree.

[0047] In the process of extracting classification confidence, as a preferred method, the acquired image frames are scaled to a 224×224 three-channel RGB format and then input into the built-in lightweight convolutional neural network. This network consists of an input layer, multiple depthwise separable convolutional layers for extracting local edge and texture features, a global average pooling layer, and a normalized exponential function classification layer for outputting an N-dimensional species probability distribution vector, where N is the total number of target species categories preset by the system. In this embodiment, the number of depthwise separable convolutional layers is 3 to 6, with each layer connected sequentially. A batch normalization layer and a non-linear activation layer are set after at least some of the depthwise separable convolutional layers. The network outputs the probability values ​​for each candidate species classification and takes the maximum probability value as the classification confidence. To ensure that those skilled in the art can reproduce this model, the network undergoes offline supervised learning using publicly available wildlife image datasets and manually labeled category labels before system deployment. During the training phase, the cross-entropy loss function is used to calculate the error between the predicted value and the true label (which is a one-hot encoded vector that is pre-annotated by humans and represents a clear category of species), and the network layer weights are updated through the backpropagation algorithm.

[0048] For temporal consistency, the system obtains bounding boxes of target objects in multiple consecutive frames using object detection or image segmentation algorithms, extracts the coordinates of the center points of the bounding boxes, and calculates the variance of the center point displacement vectors between adjacent frames. To avoid target occlusion leading to bounding box loss and affecting temporal consistency calculation, if no target is detected in a frame, a Kalman filter algorithm is used to perform trajectory prediction interpolation on the center point of that frame to maintain temporal continuity. Then, a negative exponential function (whose mapping form is...) is used... ,in, The variance of the displacement vector of the center point between adjacent frames. The variance adjustment coefficient (with a value range of 0.5 to 2.0) maps the displacement variance to a time consistency parameter with a value in the interval [0,1]. This value reflects the motion pattern of the target; the smaller the variance, the more continuous the motion trajectory.

[0049] The spatial prior matching degree is obtained by reading the historical species distribution probability map of the monitoring area pre-loaded locally. Based on the spatial coordinates of the current edge monitoring node, the historical occurrence probability of the target species in that coordinate area is retrieved, and this value is directly used as the spatial prior matching degree.

[0050] When calculating the physical trigger signal matching degree, the system performs an inner product operation between the currently captured audio frequency distribution characteristics or infrared pyroelectric waveform and the locally stored reference waveform feature vector, and then calculates the cosine similarity between the two. To prevent calculation errors, when performing division to calculate the similarity, if the magnitude of any feature vector is detected to be close to 0, the system introduces a very small positive number into the denominator. As a smoothing term to avoid division by zero errors. The feature components of the above four dimensions have all been normalized, and their values ​​are strictly limited to the interval [0,1].

[0051] S202, the extracted feature components from each dimension are aggregated into an event evidence completeness score. To avoid affecting the score calculation due to the loss of specific feature components caused by individual sensor malfunctions, the system checks the data validity flag of each sensor before performing weighted aggregation. If a feature component is found to be missing, the weight reallocation logic is triggered. Specifically, the remaining valid feature components are redistributed according to their relative proportions of their original weights, that is, the adjusted weight is equal to the original weight divided by the sum of the weights of all valid feature components, so that the total weight remains at 1.

[0052] After weight allocation, the information sensing module 100 aggregates information according to the following formula: ; In the formula, This represents the score indicating the completeness of the event evidence in the calculated output. Indicates classification confidence level; Indicates timing consistency; Indicates the spatial prior matching degree; This indicates the physical trigger signal matching degree.

[0053] , , , These are the empirical weight coefficients corresponding to each feature. In this embodiment, the values ​​of each coefficient are all in the range of (0,1) and satisfy the following conditions: The specific values ​​are optimized by the system during the pre-deployment phase using a grid search method combined with historical test data, and then fixed in the local configuration table of the node.

[0054] S203, determine whether the node state parameters meet the boundary conditions for delayed event generation. After calculating the event evidence completeness score, the device does not immediately initiate a wireless communication request, but first extracts the current node state parameters to check network and hardware resources. The boundary conditions specifically include weak network conditions, low battery conditions, and conditions where the score does not meet the standard.

[0055] In actual operation, the system estimates the current network bandwidth by sending fixed-length probe packets to the gateway and calculating the round-trip time. When it detects that the current uplink network bandwidth is lower than the network bandwidth threshold, or the remaining available power of the device is lower than the power alarm threshold, or the event evidence completeness score is less than the preset instant reporting threshold, the device is determined to meet the resource-constrained conditions.

[0056] The network bandwidth threshold here is set between 100Kbps and 500Kbps, determined based on the minimum throughput required for a single lossless transmission of a basic image frame. The power alarm threshold is set to 15% to 20% of the node's full-load battery capacity, providing a minimum reserve power to maintain the basic sensing components for 3 days. The real-time reporting threshold is typically set between 0.75 and 0.85, serving as an empirical dividing line between high-confidence events and suspected false alarms.

[0057] S204, Generate a delayed event record and write it to the delayed tolerance event queue. After determining that the node status parameter meets any one of the resource-constrained conditions, the information perception module 100 intercepts the immediate reporting process and encapsulates the associated data of the currently triggered event into a delayed event record.

[0058] The data structure for this delayed event record is defined as a tuple containing multiple fixed fields, including: an event timestamp recording the absolute time of the event, coordinates representing the physical location of the device, an event evidence integrity score calculated in the aforementioned steps, and a local resource redundancy snapshot representing the current resource status of the node.

[0059] The local resource redundancy snapshot is calculated based on a weighted combination of temperature safety and the proportion of remaining local storage space. Specifically, the temperature safety is first calculated based on the deviation of the current operating temperature from the preset safe operating temperature range, and then mapped to a standard interval of (0,1). When the current operating temperature is within the preset safe operating temperature range, the temperature safety takes a higher value; when the current operating temperature deviates from the preset safe operating temperature range, the temperature safety decreases as the deviation increases. Then, the temperature safety and the proportion of remaining local storage space are multiplied by a third weighting coefficient and a fourth weighting coefficient, respectively, and then summed in a weighted manner. The values ​​of these two weighting coefficients are both limited to the interval (0,1), and their sum equals 1.

[0060] Additionally, the tuple contains a validity period parameter calculated based on the current device battery level. This tuple, along with the corresponding image data or text summary data, is written to a dedicated latency-tolerant event queue in the edge monitoring node's local storage for caching.

[0061] S205 manages the flow of the latency-tolerant queue and executes data reporting when the triggering conditions are met. During the silent period of the data buffer, the information sensing module 100 maintains a low-power listening state to continuously monitor the communication link and the status of adjacent nodes. When it detects a cooperative communication message sent by an adjacent node, thus meeting the link recovery condition, or successfully parses a broadcast beacon containing a valid identity verification code emitted by the patrol device through a near-field communication protocol such as Bluetooth Low Energy, thus meeting the data retrieval condition, the system triggers a batch dequeue operation.

[0062] During dequeueing, the system iterates through all records in the delay tolerance event queue and extracts the difference between the event's timestamp and the current system time. If the difference exceeds the record's set expiration date, the data is deemed to have lost its ecological monitoring timeliness and is deleted, thereby freeing up local storage space. For records that have not expired, they are sorted in descending order based on their event evidence completeness score, and data is sent sequentially to the cloud-based dispatch management system or nearby patrol terminals.

[0063] During data reporting transmission, if the system detects a sudden increase in the bit error rate (BER) of the wireless channel (i.e., the BER exceeds the set BER tolerance threshold, which is typically set to 10), -3 Up to 10 -2 If there is a risk of disconnection (between the two), the image file is removed from the data to be reported, and only the text tuple containing the recognition results and coordinates is retained for downgraded reporting to reduce the amount of reported data.

[0064] Reference Figure 5Step S300 is executed by the risk prediction module 200, which is used to quantify the impact of local node failure on the monitoring continuity of the entire ecological monitoring network from a global perspective. The specific implementation process includes the following sub-steps: S301, Extract the basic ecological value of the target node under normal conditions. To ensure statistical alignment of multi-source data over time windows, the risk prediction module 200 selects the past 12 months as a fixed historical data statistical period. Based on the spatial coordinates of the target node's location, the historical occurrence frequency of species in that coordinate region within the statistical period is retrieved. In this embodiment, the historical occurrence frequency of species after maximum-minimum value normalization is directly used as the basic ecological value under normal conditions.

[0065] The above-mentioned maximum and minimum value normalization process involves subtracting the global statistical minimum frequency from the historical frequency of the current coordinate region, and then dividing by the difference between the global maximum and minimum frequencies. To avoid the error of the denominator being 0 due to the extreme uniformity of species distribution leading to the highest and lowest frequencies being equal, a very small positive constant (usually 0.001) is added to the denominator for smoothing correction. The value of this parameter is constrained to the interval [0,1], and a higher value indicates a stronger ecological observation activity in the region under normal conditions.

[0066] When the target node lacks complete historical statistical data, the risk prediction module 200 uses the average historical occurrence frequency of similar nodes in adjacent areas, or the prior value of regional ecological value configured during system initialization, as a substitute input for the basic ecological value.

[0067] S302, combining the activation status of sudden events with segmented logic to calculate the comprehensive ecological value. The system determines in real time whether the target node is in the activation state of a sudden event. When a node does not detect a valid target event extracted in the previous steps, it is determined to be in a normal background, and the comprehensive ecological value is directly equal to the basic ecological value mentioned above. When a node captures a target event and is in the activation state, the current sudden ecological value is calculated by combining the protection level of the target species and the completeness score of the event evidence.

[0068] Before calculating the ecological value of an emergency, a mapping table between species protection levels and numerical weights is established. As a preferred method, Class I protected animals are mapped to weight values ​​ranging from 0.9 to 1.0, and common species are mapped to weight values ​​ranging from 0.1 to 0.3. The protection level weights obtained from the table are multiplied by the previously calculated event evidence completeness score using pure numerical multiplication to obtain the ecological value of the emergency. Since both parameters involved in the multiplication are mapped to the dimensionless interval [0,1], this operation ensures dimensional consistency while also constraining the output result within the standard numerical interval [0,1].

[0069] To ensure a smooth transition of value assessment over time, when a node is active, the maximum value between the basic ecological value and the emergency ecological value is selected as the current comprehensive ecological value through comparative logic. This operation of selecting the larger value is used to retain a higher evaluation value between normal value and emergency event value, so as to characterize the importance of the target area to the ecological monitoring network at the current moment. This comprehensive ecological value is also constrained within the range of [0,1].

[0070] S303, extract multidimensional risk factors for risk measurement. These multidimensional risk factors specifically include three independent variable factors: node operational redundancy, coverage gap coefficient, and probability of manual patrol interruption.

[0071] For node operational redundancy, the system extracts the current battery degradation rate and memory bad block ratio of the device through the underlying hardware interface. The battery degradation rate is defined as the deviation between the node's current maximum rechargeable capacity and its factory rated capacity; the memory bad block ratio is defined as the physical proportion of non-writable sectors in the local storage medium. The battery degradation rate and memory bad block ratio are multiplied by their respective first and second hardware weighting coefficients, linearly weighted, and then the weighted sum is subtracted from 1 to obtain the node operational redundancy. The values ​​of the two hardware weighting coefficients are both limited to the range of (0,1) and their sum equals 1. The specific values ​​are configured during system initialization based on the actual dependence of the monitored node on power supply and storage resources. To avoid hardware interface communication anomalies affecting computation, if a hardware interface communication anomaly causes a read failure, the system extracts historical average data from the past 72 hours as a replacement.

[0072] The purpose of the coverage gap coefficient is to quantify the scale of the monitoring gap area that cannot be compensated by the surrounding topology network after a node failure. Using the bimodal compensation benefit matrix persistently stored in the previous steps, the compensation benefit values ​​that all neighboring nodes can provide to the target node are extracted, and the maximum compensation benefit value is selected. Subtracting this maximum compensation benefit value from 1 yields the coverage gap coefficient. If the target node is in a geographically isolated state, i.e., the set of neighboring nodes is empty, the system triggers the anomaly handling boundary logic, directly setting the coverage gap coefficient to 1, indicating that surrounding nodes cannot provide effective compensation after the node's failure.

[0073] To determine the probability of manual patrol interruption, the system extracts the path topology from the target node to the nearest manual patrol station from the geographic information system. It then integrates and sums the terrain slope values ​​along the route to obtain the path slope integral value, while simultaneously acquiring real-time rainfall environmental parameters from the local meteorological service interface. The path slope integral value and rainfall parameters are normalized and then weighted and summed to obtain the comprehensive environmental resistance value. This comprehensive environmental resistance value is then input into a logistic function for nonlinear mapping to obtain the final probability of manual patrol interruption. This nonlinear mapping calculation is performed using the following formula: ; In the formula, This represents the mapped probability of manual patrol interception. This represents the comprehensive environmental resistance value obtained from previous calculations; The slope coefficient of the mapping curve, which characterizes the steepness of probability growth, is set between [5, 15] to adjust the sensitivity of environmental resistance to the impact of manual patrols. The set environmental resistance threshold center point is typically taken as an empirical value of 0.5. By introducing this mapping model, the system can map the comprehensive environmental resistance value to the standard probability interval of (0,1), thereby characterizing the nonlinear change trend of the probability of manual patrol blocking when environmental resistance increases.

[0074] S304 performs a joint calculation of dynamic ecological value and multidimensional risk factors, aggregating and outputting continuous risk values. All factors involved in the calculation have been dimensionless. The risk prediction module 200 performs a joint multiplication operation based on the continuous risk aggregation formula: ; In the formula, This indicates the monitoring continuity risk value of the target node; This represents the normalized comprehensive ecological value; Indicates node redundancy, using The complement terms are used to characterize the risk of node malfunction; This represents the coverage gap coefficient obtained from the aforementioned calculation; This indicates the probability of interruption by manual patrols.

[0075] In this embodiment, the multiplicative joint solution model processes each risk factor concurrently under mutually independent approximation conditions. Its actual physical meaning is as follows: When the target area has high ecological observation value, equipment operational redundancy is reduced, external monitoring networks cannot provide effective compensation, and on-site conditions hinder manual maintenance, the combined effect of these conditions increases the monitoring continuity risk value. This joint solution logic based on multi-dimensional factors effectively avoids one-sided scheduling judgments caused by relying solely on a single extreme condition (e.g., equipment is in a low-battery alarm state but the area itself has no recent high-frequency observation value, or the area has high value but is surrounded by overlapping compensation nodes with full coverage). The calculated monitoring continuity risk value serves as a quantitative indicator to measure the impact of local node failure on the global network, providing data basis for the system to dynamically assign maintenance work orders in the future.

[0076] Reference Figure 6 Step S400 is executed by the compensation decision module 300, which is used to schedule monitoring node resources across nodes to supplement the coverage of the missing area when the risk exceeds the limit. The specific implementation process includes the following sub-steps: S401, Cross-node candidate optimization and estimated compensation resource consumption assessment. When the continuity risk value calculated by the risk prediction module 200 exceeds the set risk tolerance threshold, the compensation decision module 300 initiates the cross-node compensation process. As a preferred approach, this risk tolerance threshold is typically set between 0.65 and 0.75, configured by the system during initialization based on the species rarity of the involved ecological area and the grid fault tolerance requirements.

[0077] In the dual-mode compensation benefit matrix, for the coverage gap area where the failed node is located, the system first searches for adjacent nodes in the spatial coverage compensation benefit submatrix in descending order of benefit value. If no spatial compensation candidate node meets the benefit threshold and operating status conditions, then adjacent nodes are searched in the corridor sampling compensation benefit submatrix in descending order of benefit value, and the highest-benefit adjacent node that meets the benefit threshold and is in normal operating status is identified as the compensation candidate node. If the compensation benefit value corresponding to the highest-benefit adjacent node is lower than the preset compensation benefit threshold, then it is determined that there is no available compensation candidate node; the preset compensation benefit threshold ranges from 0.4 to 0.6. If the search finds that the set of adjacent nodes is empty or all adjacent nodes are in an abnormal state, the system directly marks the coverage gap area as a manual intervention state and reports a maintenance work order to avoid continuously executing invalid candidate searches.

[0078] After identifying candidate nodes, the system calculates the estimated compensatory resource consumption for that node to take over the monitoring task. This indicator is quantified based on the additional energy consumption and storage usage resulting from expanding the monitoring area. Before calculation, the compensation decision module 300 introduces a proportional calculation model based on the percentage of remaining resources to ensure that the evaluation logic does not lead to a biased judgment due to a single resource surplus. The specific formula is as follows: ; In the formula, This indicates the estimated amount of resources to be consumed in compensation. This represents the increase in average hourly energy consumption (in mAh) required for a candidate node to perform compensation. This indicates the current remaining available power (in mAh) of the candidate node. This indicates the estimated increase in storage usage due to the expanded monitoring area (in MB). This represents the remaining available local storage space (in MB). Since the numerators and denominators of the factors in the formula use the same physical units, the division naturally eliminates the dimensions, ensuring dimensional consistency in the addition operation.

[0079] To prevent calculation errors caused by insufficient battery power or local storage space resulting in a denominator of 0, the system introduces a very small positive number into the denominator term. (The value is usually set to 0.01 to 0.05) as a smoothing term. and The resource preference weight coefficients for energy consumption and storage consumption are respectively limited to the interval (0,1) and their sum is equal to 1.

[0080] The specific values ​​of the two coefficients mentioned above are set based on the actual dependence of the node on power supply and storage resources. Specifically, when a candidate node is equipped with solar panels and is continuously charging, but its local storage chip capacity is small, The value is set to be greater than When the device is powered by a pure battery without external energy supplementation, The value is set between 0.7 and 0.9 to increase the weight of power constraints in the expected consumption of compensatory resources.

[0081] S402, Multi-level cost determination based on expected compensation resource consumption. After calculating the expected compensation resource consumption, the compensation decision module 300 compares it with internally set first and second cost thresholds. As a preferred approach, the first cost threshold is set to a range of 0.3 to 0.4, and the second cost threshold is set to a range of 0.7 to 0.8. These two thresholds are determined by extrapolating from historical experimental data based on the discharge decay curve characteristics of the batteries equipped at the nodes and the minimum continuous operating days required by the system.

[0082] When executing specific judgment rules, if the expected compensation resource consumption is less than or equal to the first cost threshold, the system determines that it is currently in a low-cost range and that the candidate nodes have sufficient redundancy, and then generates a first-level compensation instruction. If the expected compensation resource consumption is greater than the first cost threshold but less than or equal to the second cost threshold, the system determines that it is in a medium-cost range, indicating that resources are relatively scarce, and then generates a second-level compensation instruction to limit excessive energy consumption. If the expected compensation resource consumption is greater than the second cost threshold, the system determines that the candidate node resources are insufficient to support cross-regional compensation actions. To reduce the risk of candidate nodes experiencing operational anomalies due to insufficient resources and expanding the coverage gap, the system terminates the node compensation process and stops issuing compensation instructions.

[0083] S403 triggers the wake-up mechanism and issues tiered compensation instructions. Candidate nodes meeting the compensation conditions are typically in low-power sleep mode under normal circumstances when no event is triggered. The compensation decision module 300 needs to send instructions through the wake-up mechanism at the bottom layer of the wireless sensor network.

[0084] In this embodiment, the system employs a low-power wake-up communication method to enable candidate nodes to receive compensation instructions. During the sleep period, candidate nodes maintain a low-power listening state to receive wake-up messages sent by the cloud scheduling management system or gateway. The compensation decision module 300 sends a wake-up message containing the candidate node's identifier to the candidate node through the gateway. After recognizing the wake-up message and completing identity verification, the candidate node switches from the low-power listening state to the instruction receiving state and sends a confirmation message back to the cloud scheduling management system or gateway. Upon receiving the confirmation message, the compensation decision module 300 issues the corresponding first compensation level instruction or second compensation level instruction.

[0085] As a preferred implementation, when the uplink wide area communication link is blocked or the wide area communication latency exceeds a preset latency threshold, the system uses an edge direct-connect wake-up method to transmit compensation trigger information. Specifically, when the cloud scheduling and management system is available, the compensation decision module 300 pre-issues a topology authorization table to the edge monitoring nodes. The topology authorization table includes at least the identifiers of the wake-up candidate compensation nodes, the allowed compensation level, the authorization validity period, and identity verification information.

[0086] When active nodes near a failed node determine, based on their locally cached topology authorization table, that a candidate node meets the compensation criteria, the active nodes send a wake-up beacon containing node identifier and identity verification information to the candidate node via their local low-power communication module. Upon receiving the wake-up beacon and completing identity verification, the candidate node switches from low-power monitoring to instruction receiving and receives or executes corresponding monitoring parameter adjustment instructions according to the compensation level and execution time window specified in the topology authorization table.

[0087] Through the aforementioned edge direct connection wake-up method, the system can maintain the continuity of the compensation scheduling process when the wide area communication link is limited, and provide the execution status basis for subsequent maintenance priority feedback and secondary patrol work order reconstruction.

[0088] S404 configures monitoring parameters and constrains the execution time window according to the compensation level instruction. After receiving the instruction, the candidate node dynamically adjusts the specific monitoring parameters within itself. The monitoring parameters here explicitly cover five dimensions: sampling frequency, trigger threshold, reporting priority, monitoring area configuration, and execution time window.

[0089] Upon receiving the first compensation level instruction, the candidate node executes the monitoring parameter configuration strategy corresponding to the first compensation level. The system writes the coordinate range of the target failed node into the current monitoring area configuration table to increase the monitoring intensity of the target area, and simultaneously increases the image sampling frequency, for example, from the normal 15fps to 30fps. In addition, the system lowers the analog voltage comparison threshold of the passive infrared sensor, preferably from the conventional 1.5V to 1.2V, thereby improving the physical sensing sensitivity to weak biological heat sources, and elevates the reporting priority of the captured data to the highest level.

[0090] Upon receiving the second-level compensation instruction, the candidate node implements a resource-constrained balanced compensation strategy. The candidate node only adjusts the monitoring direction parameter to the direction of the missing region, maintaining the basic sampling frequency unchanged. The system then moderately increases the analog voltage comparison threshold of the passive infrared sensor, preferably from the conventional 1.5V to 1.8V, thereby filtering out small interfering heat sources or low-value false alarms at the physical trigger level. Simultaneously, at this level, only the category text of key targets is extracted and reported with high priority, while large accompanying image files are downgraded to the lowest priority, thus achieving a balance between supplementing coverage and minimizing its own energy consumption.

[0091] To prevent the device battery from being depleted indefinitely due to uplink network anomalies causing a prolonged compensatory state, all monitoring parameter adjustment commands include a set execution time window. This time window is typically set between 24 and 48 hours. The candidate node's built-in timer starts counting down synchronously upon receiving the command, executing the corresponding monitoring strategy within the preset time window. Once the time window expires and no extension confirmation packet is received from the network side, the candidate node will automatically clear the temporary monitoring area configuration, and all monitoring parameters will be reset to their normal baseline values.

[0092] Reference Figure 6 Step S500 is executed by the work order reconstruction module 400, which is used to realize the cross-domain conversion of node compensation consumption into manual scheduling instructions and system closed-loop correction. The specific implementation process includes the following sub-steps: S501, Node Consumption Mapping and Maintenance Priority Feedback. After the compensation decision module 300 completes the cross-node compensation action, the candidate nodes participating in the compensation will incur additional system resource consumption. The work order reconstruction module 400 calculates the node's power consumption integral and storage occupancy ratio within the compensation execution time window.

[0093] The power consumption integral is calculated by dividing the additional charge consumed during the compensation period by the battery's rated capacity; the storage occupancy ratio is calculated by dividing the number of bytes in the monitoring media file generated during the compensation action by the total local storage capacity. To avoid division by zero due to communication anomalies in the underlying hardware driver causing the read rated capacity or total storage capacity value to be 0, a default capacity value is preset in the calculation logic. When the read denominator value is less than or equal to 0, the system uses the default capacity value from the factory configuration list as the denominator in the calculation.

[0094] Since the numerators and denominators of the above two calculations use the same physical units, the division naturally results in a dimensionless ratio. Based on this, the system calculates the compensatory consumption increment according to the node consumption mapping formula: ; In the formula, Indicates the increase in compensatory consumption; Indicates points for electricity consumption; Indicates the storage usage ratio; and These represent the power consumption weighting coefficient and the storage consumption weighting coefficient, respectively, both limited to the interval (0,1) and their sum equal to 1. When the node uses solar float charging for power supply, it is preferred to... Take 0.2, Take 0.8; when the node is powered by a pure battery without external energy supplementation, the preferred value is... Take 0.8, Take 0.2.

[0095] After obtaining the incremental compensation consumption, the system adds it to the basic manual maintenance priority of the candidate nodes as a weighted coefficient. Specifically, the basic manual maintenance priority is multiplied by... This yields the final manual maintenance priority value. This priority feedback mechanism transforms the resource consumption generated by nodes performing compensatory actions into quantifiable manual scheduling indicators. When generating a maintenance work order, the system prioritizes assigning personnel to nodes with high resource consumption to perform battery replacements or memory cleanup, thereby reducing the probability of these nodes experiencing operational anomalies due to previous compensatory actions.

[0096] S502, a data recovery path planning system based on the effective wireless coverage area. For target nodes with normal hardware operation but obstructed uplink wide-area communication links, on-site physical disassembly and repair are usually unnecessary; only manual approach is required to recover locally cached data. In such cases, the work order reconstruction module 400 adjusts the patrol path planning target from the node's physical coordinates to the effective wireless coverage area. The system extracts the maximum transmit power and receive sensitivity parameters of the target node's built-in wireless LAN communication module, as well as the vegetation attenuation coefficient provided by the on-site geographic information system.

[0097] At this point, a logarithmic distance path loss model is introduced for distance estimation. This model combines the reference path loss value at 1m with the path loss index of the field environment (typically 3.0 to 4.5 in dense forest environments) to calculate the physical communication distance boundary that satisfies the minimum signal-to-noise ratio for signal demodulation. The effective communication radius obtained from this is typically between 50m and 200m.

[0098] The system then generates a circular effective wireless coverage area with the target node's physical coordinates as the center and the effective communication radius as the radius. When scheduling manual patrol work orders, the work order reconstruction module 400 relaxes the endpoint constraint of the patrol route from specific node center coordinates to any spatial point within this effective wireless coverage area. When patrol personnel carrying mobile terminals move along the planned path, and the terminal's satellite positioning module detects that the current coordinates have entered the coverage area, the terminal automatically initiates a data retrieval task.

[0099] Regarding the communication establishment mechanism for this data recycling, after the mobile terminal enters the effective wireless coverage area, it establishes a near-field wireless data connection with the target node. After the connection is established, the mobile terminal sends a data recycling request to the target node, and the target node uploads the locally cached data to the mobile terminal according to the preset data slicing and verification rules.

[0100] S503, model adaptive correction based on on-site verification tags. After arriving at the field area, patrol personnel use mobile terminals to record the actual boundaries of the missing coverage areas and the actual frequency of species occurrence in the target area, and upload the on-site verification results. The work order reconstruction module 400 receives the verification results, performs data cleaning, and transforms them into structured real tag data.

[0101] In this embodiment, real label data is defined as a structured data record containing multidimensional fields, specifically including floating-point values ​​of latitude and longitude deviation of physical coordinates, integer values ​​of the number of species appearing within a specific time window, and classification enumeration values ​​of on-site vegetation occlusion status.

[0102] The system utilizes a moving average algorithm to perform closed-loop feedback correction on the model parameters of preceding stages based on extracted real label data. The correction scope covers the feature aggregation weights used by the information perception module 100 to calculate the event evidence completeness score and the initial parameters of the dual-mode compensation benefit matrix. The parameter update formula is as follows: ; In the formula, This indicates the new parameter value after correction; This represents the system's original historical parameter values; This refers to the reference parameter value calculated independently based on the actual label data submitted manually on-site; This represents the learning rate coefficient that controls the update step size.

[0103] The above formula represents a weighted fusion of historical parameter values ​​and reference parameter values ​​obtained based on on-site verification results, where the weight of the historical parameter values ​​is... The weight of the reference parameter value is .

[0104] Learning rate coefficient The value is set between 0.05 and 0.15. If the value is too large, the system parameters are easily affected by single local observation errors, resulting in numerical oscillations; if the value is too small, the model's tracking speed of the evolution of the real physical environment will lag significantly. Through this closed-loop feedback mechanism, the system uses manually verified on-site verification results to correct parameter deviations formed during long-term operation.

[0105] The present invention also provides a computer device, including: a processor and a memory, the memory storing a computer program executable by the processor, the computer program performing the method described above when executed by the processor.

[0106] The present invention also provides a storage medium storing a computer program, which is executed by a processor to perform the method described above.

[0107] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0108] To illustrate the operation of the technical solution of the present invention in a real-world ecological patrol scenario, the following is an application example deployed in a mountainous forest monitoring area of ​​a nature reserve, and the technical effects are explained in detail with reference to the accompanying drawings and comparative test data.

[0109] Within this gridded monitoring area, edge monitoring nodes A, B, and C are deployed. Node A is located at the intersection of animal passageways and is powered entirely by batteries; Node B is located in an adjacent open slope area, approximately 120 meters from Node A, and is equipped with solar power supplementation components; Node C is located in the riverside corridor area downstream of Node A, approximately 200 meters from Node A, and is powered entirely by batteries.

[0110] During a rainstorm, Node A detected a physical trigger signal indicating the presence of a critically protected species. The Information Sensing Module 100 was activated and extracted on-site images and infrared trigger data, outputting a high event evidence completeness score after feature fusion. At this time, due to the severe weather, Node A detected a sharp drop in uplink network bandwidth and its remaining available power was below the alarm threshold. The Information Sensing Module 100 determined that the current node status parameters met resource constraints, immediately intercepted the immediate reporting process, encapsulated the trigger data and a snapshot of local resource redundancy as a delayed event record, wrote it to the delay-tolerant event queue, and entered a low-power listening state. Before entering hibernation, Node A sent a lightweight event summary to surrounding active nodes via an edge direct connection wake-up message.

[0111] Subsequently, the cloud-based risk prediction module 200 failed to receive a normal heartbeat message from node A during periodic polling. Combining the previously persistently stored basic data, the risk prediction module 200 extracted the basic ecological value of node A's coordinates over the past 12 months. Since it received event summaries forwarded by surrounding nodes, the risk prediction module 200 determined that node A was in a state of sudden event activation. Combining the protection level weights corresponding to key protected species and the event evidence completeness score, it calculated the comprehensive ecological value. Simultaneously, given that the comprehensive environmental resistance value of the area increased due to increased rainfall, the system calculated a high probability of obstruction by manual patrols. After joint calculation of various factors, the monitoring continuity risk value of target node A increased and reached the set risk trigger threshold. Figure 7 As shown in Figure (a), the high-risk status of the area where node A is located is marked by a dark thermal area, indicating that the anomaly of this local node affects the monitoring continuity of the ecological monitoring network.

[0112] Upon triggering the over-limit alarm, the compensation decision module 300 initiates optimization. The system searches for neighboring nodes of node A in the dual-mode compensation benefit matrix. Node B has a high spatial overlap rate, meeting the spatial coverage compensation condition; Node C has a low spatial overlap rate, but its probability of passage through the corridor calculated based on the digital elevation model is high, and its corridor sampling compensation benefit value meets the candidate node screening criteria. When assessing the expected compensation resource consumption, because node B is equipped with solar panels and is charging, its remaining available power is sufficient, and its expected compensation resource consumption is lower than the set first cost threshold; while node C is purely battery powered, and if compensation is performed, it will face the risk of rapid power depletion. Therefore, the compensation decision module 300 determines node B as a compensation candidate node and issues it a first-level compensation instruction containing a wake-up flag.

[0113] Node B receives instructions in low-power listening mode and then adjusts the monitoring parameters: it writes the missing coordinate range of Node A into the monitoring area configuration table, increases the image sampling frequency from 15fps to 30fps, lowers the analog voltage comparison threshold of the passive infrared sensor from 1.5V to 1.2V, and initiates a 48-hour execution time window. Combined with... Figure 7 As shown in (b), the sensing coverage area of ​​node B has significantly expanded, providing alternative coverage for the areas previously lacking monitoring by node A. Figure 7 The original dark-colored risk thermal patches in (a) then decayed to within the safe threshold.

[0114] During the compensation execution, the work order reconstruction module 400 records the power consumption integral and storage occupancy ratio generated by node B, calculates the compensation consumption increment, and adds it to the basic manual maintenance priority of node B. For node A with uplink obstruction, the work order reconstruction module 400 calculates an effective communication radius of approximately 80 meters based on the parameters of node A's built-in wireless communication module and the on-site vegetation attenuation coefficient, generating a circular effective wireless coverage area. After receiving the secondary patrol work order, patrol personnel do not need to go to the physical installation location of node A; they only need to bring their mobile terminals into the 80-meter effective wireless coverage area to trigger the batch dequeue operation of node A's delay tolerance event queue through near-field communication and retrieve the delay event records related to the target species. After the task is completed, the patrol personnel upload the on-site verification tag data, and the system adaptively corrects the initial parameters of the information perception module 100 and the dual-mode compensation benefit matrix accordingly.

[0115] To verify the technical effectiveness of this invention, a comparative test was conducted based on the operating environment of the aforementioned monitoring area for six consecutive months. The benchmark was a conventional wildlife monitoring system employing a fixed trigger threshold and a real-time wide-area communication backhaul strategy. Combined with... Figure 8 The curves shown indicate that, under the test conditions of this embodiment, the solution of the present invention has the following technical advantages over the comparative benchmark system: This invention reduces the number of invalid retransmissions in weak network environments and improves the effective recovery capability of delay event records by using a delay-tolerant event queue and data recovery path planning in the effective wireless coverage area. Combined with... Figure 8 Data shows that, in scenarios where communication quality deteriorates due to heavy rain and strong winds, the effective data recovery rate of this invention reaches over 94%, which is significantly better than the performance of the benchmark system, which suffers from frequent packet loss and disconnection.

[0116] This invention limits the high resource consumption state of compensation nodes by assessing the expected resource consumption, determining costs at multiple levels, and imposing execution time window constraints. Based on... Figure 8 As can be seen from the energy consumption comparison curve, during the 6-month test period, the average node power consumption level of the monitoring area using the present invention was reduced by 41.3% compared with the comparison benchmark system, which reduced the risk of surrounding nodes falling into disordered compensation due to local failure and eventually causing large-scale battery depletion.

[0117] This invention determines candidate nodes for compensation through a dual-mode compensation benefit matrix and dynamically adjusts monitoring parameters via commands, reducing the duration of uncompensated areas with missing coverage to less than 2 minutes. Furthermore, because the work order reconstruction module 400 feeds back the incremental compensation consumption to the priority of manual maintenance and relaxes the constraints on on-site patrol coordinates, it reduces the frequency of patrol personnel reaching physically challenging installation locations. Test results show that the cost of a single manual maintenance operation is reduced by more than 50% compared to the baseline system, effectively improving the operational stability of the ecological monitoring network under resource-constrained conditions.

[0118] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wildlife intelligent monitoring system based on multi-source information fusion, characterized in that, include: The information perception module, deployed at the edge monitoring node, is used to extract target event features and output an event evidence completeness score. When the edge monitoring node meets the resource-constrained conditions, a delay event record is generated and written into the delay-tolerant event queue, and the delay event record is reported when the link recovery conditions are met; The risk prediction module, deployed in the cloud-based scheduling and management system, is used to construct an ecological patrol resource scheduling map based on the basic data of the edge monitoring nodes; establish a dual-mode compensation benefit matrix; and calculate the monitoring continuity risk value of the edge monitoring nodes. The compensation decision module, deployed in the cloud-based scheduling and management system, is used to retrieve candidate compensation nodes in the dual-mode compensation benefit matrix when the monitored continuous risk value reaches the risk trigger threshold. Based on the expected consumption of compensation resources, a compensation level instruction is generated and sent to the compensation candidate node to adjust the monitoring parameters of the compensation candidate node; The work order reconstruction module, deployed in the cloud-based scheduling and management system, is used to record the node operating resources occupied by the candidate compensation nodes and calculate the compensation consumption increment; the compensation consumption increment is added to the basic manual maintenance priority, and a secondary patrol work order is generated based on the updated manual maintenance priority.

2. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The information perception module extracts the features of the target event and outputs an event evidence completeness score, specifically used for: Classification confidence, temporal consistency, spatial prior matching degree, and physical trigger signal matching degree are extracted as multidimensional feature components. Verify the data validity flag bits of the sensors corresponding to each multidimensional feature component; If a feature component is found to be missing, the remaining valid feature components will be redistributed according to their original relative weights. The multidimensional feature components are weighted and summed with the reassigned weights to generate the event evidence completeness score.

3. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The resource-constrained conditions include weak network conditions (uplink network bandwidth is lower than the network bandwidth threshold), low power conditions (remaining available power is lower than the power alarm threshold), and substandard conditions (event evidence completeness score is lower than the immediate reporting threshold); the delayed event record includes the event occurrence timestamp and validity period parameter. The information perception module performs data reporting when the data recycling conditions are met. Specifically, it is used to: obtain the difference between the event occurrence timestamp of each delayed event record in the delayed tolerance event queue and the current system time; delete delayed event records whose difference is greater than the validity period parameter; and sort the unexpired delayed event records in descending order according to the event evidence completeness score and report them in sequence.

4. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The basic data includes the spatial coordinates of the edge monitoring nodes, patrol road topology data, and node operation status parameters; the dual-mode compensation benefit matrix includes a spatial coverage compensation benefit sub-matrix and a corridor sampling compensation benefit sub-matrix. The risk prediction module establishes a spatial coverage compensation benefit sub-matrix, specifically used for: extracting a failed node and an adjacent node with connectivity in the ecological patrol resource scheduling graph; calculating the replaceable coverage ratio of the adjacent node relative to the failed node; extracting the remaining available power and local storage space of the adjacent node and weighted summing them to obtain the equipment availability score; obtaining the estimated compensation resource consumption for the adjacent node to perform compensation actions; linearly weighting the replaceable coverage ratio, the equipment availability score, and the estimated compensation resource consumption, and generating spatial coverage compensation benefit values ​​after interval pruning, thus constructing the spatial coverage compensation benefit sub-matrix.

5. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 4, characterized in that, The risk prediction module establishes a corridor sampling compensation benefit sub-matrix, specifically used for: A comprehensive resistance surface is constructed by obtaining slope values ​​and surface cover type resistance values ​​from the digital elevation model; the cumulative travel resistance cost from the failure node to the adjacent node along the natural terrain is calculated using a minimum cost path search algorithm. The cumulative passage resistance cost is mapped to the passage probability of the target passing through the adjacent node using a negative exponential decay function; the passage probability, the equipment availability score and the estimated compensation resource consumption are linearly weighted and calculated, and after interval pruning, the corridor sampling compensation revenue value is generated, and the corridor sampling compensation revenue sub-matrix is ​​constructed.

6. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The risk prediction module calculates the monitoring continuity risk value of the edge monitoring node, specifically for: The comprehensive ecological value is determined by combining the historical occurrence frequency of species in the target area under normal conditions and the protection level weight of target species under the activation state of sudden events; the battery decay rate and memory bad block ratio of the edge monitoring nodes are extracted, and a weighted sum is obtained to obtain the decay weighted sum value. The node operation redundancy is obtained by subtracting the decay weighted sum value from the value; the maximum compensation benefit value in the dual-mode compensation benefit matrix is ​​extracted, and the coverage gap coefficient is calculated by subtracting the maximum compensation benefit value from the value; the path slope integral value and rainfall parameters are obtained to generate the comprehensive environmental resistance value, and the comprehensive environmental resistance value is input into the logistic function for mapping to obtain the probability of manual patrol blockage; The monitoring continuity risk value is output by performing a joint multiplication operation on the comprehensive ecological value, the difference obtained by subtracting the node operation redundancy from the numerical value, the coverage gap coefficient, and the probability of manual patrol interruption.

7. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The compensation decision module generates compensation level instructions, specifically used for: Obtain a first ratio of the energy consumption increment required to perform the compensation action to the remaining available power of the compensation candidate node, and a second ratio of the storage occupation increment to the remaining available local storage space. Add positive constant smoothing terms as denominators to the remaining available power and the remaining available local storage space to participate in the division operation to obtain the first ratio and the second ratio, and calculate the expected compensation resource consumption by weighting. When the expected compensation resource consumption is less than or equal to a first cost threshold, generate a first compensation level instruction. When the expected compensation resource consumption is greater than the first cost threshold and less than or equal to the second cost threshold, generate a second compensation level instruction.

8. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 7, characterized in that, The compensation decision module adjusts the monitoring parameters through the compensation level instruction, specifically for: When the candidate node receives the first compensation level instruction, it increases the image sampling frequency, decreases the analog voltage comparison threshold of the passive infrared sensor, and increases the reporting priority of the captured data; when the candidate node receives the second compensation level instruction, it adjusts the monitoring direction parameter to cover the missing area, maintains the basic sampling frequency unchanged, increases the analog voltage comparison threshold of the passive infrared sensor, and limits the reporting priority of the image file. Start the built-in timer, and after the set execution time window expires, reset the monitoring parameters back to the normal baseline value.

9. The intelligent wildlife monitoring system based on multi-source information fusion according to claim 1, characterized in that, The work order reconstruction module generates a secondary patrol work order, which is further used for: The system acquires the maximum transmit power, receive sensitivity parameters, and vegetation attenuation coefficient of the wireless LAN communication module. It then inputs the logarithmic distance path loss model to calculate the physical communication distance boundary to generate the effective wireless coverage area. The endpoint constraint of the secondary patrol route is set as the effective wireless coverage area. The system receives structured real-valued data submitted by manual patrols, including floating-point values ​​of latitude and longitude deviations, species occurrences, and vegetation shading conditions. It then uses a moving average algorithm to weight and fuse historical parameter values ​​with reference parameter values ​​calculated based on the structured real-valued data, and updates the feature aggregation weights and the initial parameters of the dual-mode compensation benefit matrix.

10. A method for intelligent monitoring of wild animals based on multi-source information fusion, applied to the system described in any one of claims 1-9, characterized in that, Includes the following steps: Extract features of the target event and output an event evidence completeness score; When the edge monitoring node meets the resource-constrained conditions, a delay event record is generated and written into the delay-tolerant event queue, and the delay event record is reported when the link recovery conditions are met; An ecological patrol resource scheduling diagram is constructed based on the basic data of the edge monitoring nodes; Establish a dual-mode compensation benefit matrix and calculate the monitoring continuity risk value of the edge monitoring node; When the monitored continuity risk value reaches the risk trigger threshold, a compensation candidate node is retrieved in the dual-mode compensation benefit matrix, a compensation level instruction is generated based on the expected compensation resource consumption and sent to the compensation candidate node to adjust the monitoring parameters of the compensation candidate node. Record the node operation resources occupied by the candidate compensation nodes and calculate the compensation consumption increment. Add the compensation consumption increment to the basic manual maintenance priority and generate a secondary patrol work order based on the updated manual maintenance priority.

Images