Monitoring and early warning method and system for natural reserve of wild animals

Abstract

The invention provides a monitoring and early warning method and system for a natural protection area of wild animals. The method comprises the following steps: firstly, collecting a multi-modal ecological monitoring data stream containing animal activity track data, a habitat environment parameter sequence and a biological voiceprint signal set; then, dynamically matching the animal activity trajectory data with a reference threat mode to generate a first ecological association network feature; and extracting migration path fragments, combining with an environment parameter change gradient, and deducing second ecological association network features through the model. And then, performing cross-modal fusion on the two features to generate a multi-level threat coupling vector, constructing a map based on the biological voiceprint, performing overlay analysis on the map and the coupling vector, and calibrating and outputting a real-time threat level pedigree through an adaptive algorithm. And finally, according to the mapping relationship between the real-time threat level pedigree and a preset strategy library, activating a corresponding ecological protection intervention instruction set, realizing multi-modal data comprehensive analysis, accurate real-time monitoring and early warning and intelligent response, and effectively protecting the ecology of the natural reserve of wild animals.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to a monitoring and early warning method and system for wildlife nature reserves. Background Art

[0002] In the field of monitoring and early warning for wildlife nature reserves, traditional technologies mainly focus on single-dimensional monitoring.

[0003] Early monitoring methods mostly relied on manual regular patrols. This method not only consumed a large amount of manpower, material resources and time, but also had a limited monitoring range, making it difficult to comprehensively and timely grasp the ecological conditions within the nature reserve. With the development of technology, some automated monitoring devices have been gradually introduced, but most of them only monitor single factors.

[0004] For example, some monitoring systems only focus on measuring habitat environmental parameters, such as temperature and humidity sensors, light intensity sensors, etc. Although they can obtain certain environmental data, they cannot effectively associate this environmental information with the behavioral changes of wild animals. This results in that even if abnormal environmental parameters are detected, it is difficult to determine whether there is an actual threat to wild animals, as well as the degree and scope of the threat. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a monitoring and early warning method for wildlife nature reserves, the method comprising: Collecting a multi-modal ecological monitoring data stream within a target wildlife nature reserve, the multi-modal ecological monitoring data stream including animal activity trajectory data, a sequence of habitat environmental parameters, and a set of bioacoustic signals; Dynamically matching the animal activity trajectory data with a benchmark threat pattern in a predefined ecological threat index library to generate a first ecological association network feature, the first ecological association network feature characterizing the spatio-temporal coupling relationship between the behavior of a target animal individual and a potential threat scenario; Extracting migration path segments from the animal activity trajectory data, and combining with the change gradient of the sequence of habitat environmental parameters, to deduce a second ecological association network feature through a threat pattern evolution model, the second ecological association network feature reflecting the intensity of the synergistic effect between abnormal fluctuations in the migration path and mutations in environmental parameters; Performing cross-modal feature fusion on the first ecological association network feature and the second ecological association network feature to generate a multi-level threat coupling vector; Based on the frequency domain characteristic distribution of the biometric voiceprint signal set, a voiceprint anomaly detection map is constructed, and the voiceprint anomaly detection map is spatially superimposed and analyzed with the multi-level threat coupling vector. After a comprehensive threat situation matrix is generated, a dynamic threshold calibration is performed on the comprehensive threat situation matrix using an adaptive weight allocation algorithm to output a real-time threat level spectrum; According to the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, the ecological protection intervention instruction set of the corresponding level is activated.

[0006] On the other hand, an embodiment of the present invention also provides a monitoring and early warning system for a wildlife nature reserve, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0007] Based on the above aspects, the embodiments of the present application greatly improve the comprehensiveness, accuracy and timeliness of nature reserve monitoring and early warning. Specifically, the collection of multimodal ecological monitoring data streams covers animal activity trajectory data, habitat environmental parameter sequences and biological soundprint signal sets. This multi-dimensional data acquisition method comprehensively and meticulously reflects the operating status of the protected area ecosystem. Compared with the traditional monitoring method with a single data source, multimodal data provides a rich information foundation.

[0008] By dynamically matching animal activity trajectory data with a predefined ecological threat indicator library, the generated first ecological association network feature can accurately capture the spatiotemporal coupling relationship between the individual behavior of target animals and potential threat scenarios. It not only breaks the limitation of traditional monitoring that only focuses on a single factor, but also provides real-time insight into the dynamic changes of ecological threats in time and space dimensions, making monitoring and early warning more forward-looking and targeted.

[0009] By extracting migration path fragments and combining them with the gradient of changes in habitat environmental parameters, the characteristics of the second ecological association network are derived, which reflects the synergistic intensity of abnormal fluctuations in migration paths and sudden changes in environmental parameters, reveals the complex relationships between different elements in the ecosystem, and helps to deeply understand the intrinsic driving mechanism of ecological threats. Compared with traditional methods, it can better explore potential ecological risk factors.

[0010] Cross-modal feature fusion generates multi-level threat coupling vectors, which effectively integrates the key information of different modal data and avoids the one-sidedness and limitations that may be brought about by single modal data. Multi-level feature fusion provides a more comprehensive and representative feature expression for subsequent threat situation analysis, significantly improving the overall grasp of ecological threats.

[0011] Construct a voiceprint anomaly detection map based on the biological voiceprint signal set, and perform spatial superposition analysis with the multi-level threat coupling vector. The generated comprehensive threat situation matrix comprehensively presents the ecological threat situation within the protected area. The adaptive weight allocation algorithm further calibrates the dynamic threshold of the matrix, and the output real-time threat level spectrum is more accurate and flexible, capable of timely adapting to the dynamic changes of the ecological environment, providing a reliable basis for protection decisions.

[0012] Finally, activate the corresponding level of ecological protection intervention instruction set based on the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, realizing the seamless connection between monitoring and early warning and protection actions. Thereby, the efficiency and effect of nature reserves in responding to ecological threats are greatly improved. Brief Description of the Drawings

[0013] Figure 1 It is a schematic execution flowchart of the monitoring and early warning method for wildlife nature reserves provided by an embodiment of the present invention.

[0014] Figure 2 It is a schematic diagram of the hardware architecture of the monitoring and early warning system for wildlife nature reserves provided by an embodiment of the present invention. Detailed Embodiments

[0015] The present invention will be specifically described below in conjunction with the accompanying drawings of the specification. Figure 1 It is a schematic flowchart of the monitoring and early warning method for wildlife nature reserves provided by an embodiment of the present invention. The monitoring and early warning method for wildlife nature reserves will be introduced in detail below.

[0016] Step S110, collect the multi-modal ecological monitoring data stream within the target wildlife nature reserve. The multi-modal ecological monitoring data stream includes animal activity trajectory data, habitat environment parameter sequences, and biological voiceprint signal sets.

[0017] In this embodiment, in a large wildlife nature reserve, such as a giant panda nature reserve located in the mountains, various types of monitoring devices can be widely deployed within the reserve to collect the multi-modal ecological monitoring data stream. For the collection of animal activity trajectory data, collars or trackers with positioning functions can be installed in key areas of the reserve, and these devices are worn on giant pandas. These positioning devices can record the geographical location information of giant pandas at regular time intervals, such as every 15 minutes, and transmit the data to the data center through satellite communication or local wireless transmission networks. In this way, over time, continuous animal activity trajectory data can be formed, showing information such as the activity range and movement path of giant pandas within the reserve.

[0018] The collection of the habitat environmental parameter sequence is achieved through multiple sensor networks. In different ecological regions within the protected area, such as bamboo forests, areas near streams, slopes, etc., temperature and humidity sensors, light intensity sensors, soil humidity sensors, etc. are deployed. The temperature and humidity sensors can monitor the temperature and humidity of the surrounding environment in real time. For example, during the high-temperature period in summer, they can accurately record the temperature changes in the bamboo forest, from the relatively cool 20 degrees Celsius in the morning to possibly rising to 30 degrees Celsius in the afternoon. The light intensity sensors can detect the intensity of sunlight irradiation at different time periods, which is of great significance for understanding the light conditions of the habitats preferred by giant pandas. The soil humidity sensors can reflect the moisture content of the soil. For example, during the rainy season, the soil humidity will increase significantly, while during the dry season, it will decrease. These sensors arrange the collected data in chronological order to form the habitat environmental parameter sequence.

[0019] The collection of the bioacoustic signal set relies on highly sensitive sound collection devices. Within the protected area, in this embodiment, these sound collection devices are installed on trees or in concealed positions to ensure that the sounds emitted by organisms can be captured within the largest possible range. For example, giant pandas may emit various sounds for communication or to express emotions. These sounds are captured by the collection devices to form the bioacoustic signal set. At the same time, the calls of other wild animals such as birds and the calls of small mammals are also collected. These sound signals contain different characteristics such as frequencies, durations, intensities, etc., jointly constituting a rich bioacoustic signal set.

[0020] Step S120, dynamically match the animal activity trajectory data with the benchmark threat patterns in the predefined ecological threat index library to generate the first ecological association network feature, where the first ecological association network feature characterizes the spatio-temporal coupling relationship between the individual behavior of the target animal and potential threat scenarios.

[0021] Taking the previously mentioned giant panda nature reserve as an example, there are various benchmark threat patterns in the predefined ecological threat index library. Suppose one of the benchmark threat patterns is a threat related to human activities, such as the illegal entry of humans into the core activity area of giant pandas. When the collected giant panda activity trajectory data shows that a giant panda suddenly changes its regular activity path, moves from the deep bamboo forest where it was originally safe towards the edge of the protected area, and during this process, the activity trajectory becomes irregular and the speed also changes. At this time, in this embodiment, the animal activity trajectory data is first processed. First, noise filtering is performed on it to exclude abnormal data points caused by occasional equipment failures or environmental interferences, and then trajectory interpolation processing is performed to ensure the continuity of the trajectory, generating a standardized activity trajectory sequence. For the habitat environmental parameter sequence, time alignment is performed to ensure consistency with the animal activity trajectory data on the time scale, and at the same time, abnormal values are corrected to generate a continuous environmental parameter surface.

[0022] Perform spatio-temporal grid mapping on the standardized activity trajectory sequence and the continuous environmental parameter surface. For example, divide the protected area into small grid cells, and the size of each cell can be determined according to actual needs and data accuracy. For example, one square kilometer is taken as a cell. Within each grid cell, extract the behavior density distribution and the environmental parameter gradient. For example, in the grid cells where giant pandas are highly active, the behavior density distribution is relatively high. And if the environmental parameter gradient in this area shows that the noise level suddenly increases due to surrounding human activities or abnormal changes occur in temperature and humidity.

[0023] Then, compare the behavior density distribution and the environmental parameter gradient with the spatio-temporal constraint conditions of the benchmark threat pattern through the threat pattern matching engine. If the temporal and spatial characteristics of the giant panda's activity trajectory near the area with frequent human activities match the spatio-temporal constraint conditions of the benchmark threat pattern where illegal human entry causes animals to change their behavior, the threat matching confidence of each grid cell will be generated. For example, in several grid cells near the edge of the protected area, the threat matching confidence is relatively high, while in the grid cells far from the human activity area, the threat matching confidence is relatively low. Construct a three-dimensional threat heat map based on the threat matching confidence of all grid cells as the first ecological association network feature. In this three-dimensional threat heat map, the height can represent the magnitude of the threat matching confidence, and different colors can represent different threat levels, thus intuitively showing the spatio-temporal coupling relationship between the individual behavior of giant pandas and potential threat scenarios, such as which areas of giant pandas are more vulnerable to human activities and the degree of threat.

[0024] Step S130, extract the migration path segments from the animal activity trajectory data, and combine with the change gradient of the habitat environmental parameter sequence, and deduce the second ecological association network feature through the threat pattern evolution model, where the second ecological association network feature reflects the synergistic effect intensity of the abnormal fluctuation of the migration path and the mutation of the environmental parameters.

[0025] Continuing with the example of the giant panda nature reserve, for the activity trajectory data of giant pandas, this embodiment identifies the path turning points and staying periods therein. For example, giant pandas may migrate from bamboo forests at lower altitudes to areas at higher altitudes in spring every year. During this process, there will be path turning points when it moves from one bamboo forest to another, and there may be staying periods near water sources. Through this information, this embodiment segments the migration path sub-segments with continuous motion vectors.

[0026] Perform curvature analysis and velocity variation detection on each sub - segment of the migration path. If, in a certain sub - segment of the migration path, the curvature of the giant panda's movement path suddenly increases, the originally relatively straight path becomes curved, or there are obvious variations in speed, such as suddenly accelerating or decelerating from normal slow movement, this embodiment marks these abnormal path morphological features. At the same time, combine the change gradient of the habitat environmental parameter sequence. For example, in the bamboo forest near the giant panda's migration path, due to abnormal climate, the temperature drops suddenly, which affects the growth of bamboo, and the soil humidity also mutates due to precipitation changes.

[0027] Conduct a causal correlation analysis between the abnormal path morphological features and the environmental parameter mutation events within the corresponding time window. If it is found that the abnormal changes in the giant panda's path are related to the environmental parameter mutations in terms of time and space, for example, in the area where the temperature drops suddenly, the giant panda changes its migration path or speed, this embodiment calculates the path - environment coupling coefficient. Suppose in a specific sub - segment of the migration path, the environmental parameter mutation is very severe, and the giant panda's path also shows obvious abnormal changes, then this path - environment coupling coefficient will be relatively high. Based on the path - environment coupling coefficients of all sub - segments of the migration path, construct a dynamic threat propagation chain as the second ecological association network feature. This dynamic threat propagation chain can show the degree of influence of environmental parameter mutations on the giant panda's migration on different migration paths, and how this influence spreads between different path segments, reflecting the intensity of the synergistic effect between abnormal fluctuations of the migration path and environmental parameter mutations.

[0028] Step S140: Perform cross - modal feature fusion on the first ecological association network feature and the second ecological association network feature to generate a multi - level threat coupling vector. Based on the frequency - domain feature distribution of the bio - acoustic fingerprint signal set, construct a bio - acoustic fingerprint anomaly detection map, and perform spatial superposition analysis on the bio - acoustic fingerprint anomaly detection map and the multi - level threat coupling vector. After generating a comprehensive threat situation matrix, use an adaptive weight allocation algorithm to dynamically calibrate the threshold of the comprehensive threat situation matrix and output a real - time threat level spectrum.

[0029] In the monitoring scenario of the giant panda nature reserve, first perform principal component dimensionality reduction on the first ecological association network feature (three - dimensional threat heat map). In this process, this embodiment extracts the core features that can best represent threat information from the three - dimensional threat heat map, such as the information of grid cells with high threat matching confidence and representativeness, and converts them into core threat space vectors.

[0030] For the second ecological association network feature (dynamic threat propagation chain), in this embodiment, it is converted into a spatio-temporal propagation probability matrix. This matrix can represent the probability of threat propagation along the migration path at different time and space positions. Then, a tensor product operation is performed on the spatio-temporal propagation probability matrix and the core threat space vector to generate a multi-level threat coupling vector. This vector integrates two different modalities of information: the spatio-temporal coupling relationship between animal individual behavior and potential threat scenarios, and the intensity of the synergistic effect between abnormal fluctuations in the migration path and environmental parameter mutations.

[0031] For the bioacoustic signal set, in this embodiment, it is subjected to frame addition and windowing processing. For example, continuous sound signals are segmented according to a certain time length (such as 0.1 seconds per frame), and an appropriate window function is added to improve the accuracy of spectral analysis. Then, the Mel frequency cepstral coefficients and harmonic energy ratios of each frame of the signal are extracted. These features are jointly discriminated by an abnormal acoustic fingerprint classifier. For example, if the Mel frequency cepstral coefficients and harmonic energy ratios of a certain sound are significantly different from the characteristics of normal giant panda calls or the normal calls of other organisms, it will be determined as an abnormal acoustic fingerprint. An acoustic fingerprint anomaly probability distribution map is output, and different regions in this map represent the probability of acoustic fingerprint anomalies. Then, based on the sound source localization algorithm, spatial clustering is performed on abnormal acoustic fingerprint events. For example, if multiple abnormal acoustic fingerprint events are detected in a certain area of the protected area, these events are classified into one category, and an acoustic fingerprint anomaly detection map with azimuth markers is generated.

[0032] The azimuth markers of the acoustic fingerprint anomaly detection map are registered with the spatial coordinates of the multi-level threat coupling vector. For example, if the acoustic fingerprint anomaly detection map shows that there are acoustic fingerprint anomalies in the northeast area of the protected area, and there is also a high threat coupling value in this area in the multi-level threat coupling vector, accurate spatial registration is performed. Then, the overlapping area ratio and spatial correlation index between the acoustic fingerprint anomaly area and the threat coupling area are calculated to generate a comprehensive threat situation matrix. This comprehensive threat situation matrix synthesizes threat information from multiple aspects such as animal activity trajectories, habitat environments, and bioacoustics.

[0033] Next, use the adaptive weight allocation algorithm to dynamically calibrate the comprehensive threat situation matrix. According to the threat level distribution law in the historical threat event database, establish the baseline threshold curves for each threat dimension (such as the animal behavior threat dimension, the environmental threat dimension, the voiceprint threat dimension, etc.). For example, for the animal behavior threat dimension, if the threat levels were mostly at the medium level in the past under similar abnormal animal behavior situations, then establish the baseline threshold curve based on this. Use the sliding time window algorithm to calculate the cumulative intensity values of each threat dimension in the current comprehensive threat situation matrix in real time. For example, within a short time window (such as 1 hour), calculate the cumulative intensities of dimensions such as animal behavior threat, environmental threat, and voiceprint threat. Map the cumulative intensity values to the baseline threshold curves through non-linear interpolation methods, and output the real-time threat level spectrum with dynamic adjustment. This spectrum can accurately reflect the real-time threat levels within the protected area according to the current actual situation, such as different levels like low threat, medium threat, or high threat.

[0034] Step S150, according to the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, activate the corresponding level of ecological protection intervention instruction set.

[0035] For example, in the context of a giant panda nature reserve, assume that the real-time threat level spectrum shows a high threat level. In the preset emergency response strategy library, there is a corresponding set of ecological protection intervention instruction sets for the high threat level. First, analyze the weight distribution of each threat dimension in the real-time threat level spectrum. For example, if the weight of the animal behavior threat dimension is high, it indicates that the animals may face direct danger, and if the weights of the environmental threat dimension and the voiceprint threat dimension are also high, it may mean that there are multiple threat factors acting together.

[0036] According to this weight distribution, match the multi-level response protocols in the preset emergency response strategy library. For the situation of high threat level and high weight of animal behavior threat, response protocols including emergency rescue of animals, strengthening of perimeter patrols, etc. may be matched. According to the spatial coverage range and duration of the threat level, generate the drone patrol path planning, ecological corridor blockade instructions, and the priority queue for manual inspections. If the area covered by the high threat level is large and the duration is long, the drone patrol path in a larger range will be planned, the ecological corridors that may be dangerous will be blocked, and the priority of manual inspections will be increased, so that the inspection personnel will focus on the high threat areas.

[0037] Finally, the instruction set is synchronized to the protected area management terminal and mobile law enforcement devices through a low-latency communication network. After receiving the instructions, the protected area management terminal can schedule the management resources of the entire protected area, such as arranging more staff and equipment to high-threat areas. After receiving the instructions, the mobile law enforcement devices allow law enforcement officers to perform specific operations according to the instructions, such as setting up blockade signs in ecological corridors and patrolling along the planned routes, so as to respond to high-threat situations in the protected area in a timely and effective manner and protect the living environment and safety of wild animals such as giant pandas.

[0038] Based on the above steps, the embodiments of the present application greatly improve the comprehensiveness, accuracy, and timeliness of natural protected area monitoring and early warning. Specifically, the collection of multi-modal ecological monitoring data streams covers animal activity trajectory data, habitat environmental parameter sequences, and bioacoustic signal sets. This multi-dimensional data acquisition method comprehensively and meticulously reflects the operating state of the protected area ecosystem. Compared with the traditional monitoring method with a single data source, multi-modal data provides a rich information basis.

[0039] Dynamically matching the animal activity trajectory data with a predefined ecological threat index library, the generated first ecological association network feature can accurately capture the spatio-temporal coupling relationship between the individual behaviors of target animals and potential threat scenarios. It not only breaks the limitation of traditional monitoring that only focuses on a single factor but also can insight into the dynamic changes of ecological threats in the time and space dimensions in real time, making the monitoring and early warning more forward-looking and targeted.

[0040] By extracting migration path segments and combining them with the change gradient of habitat environmental parameters to derive the second ecological association network feature, it reflects the intensity of the synergistic effect between abnormal fluctuations in migration paths and sudden changes in environmental parameters, reveals the complex interrelationships between different elements in the ecosystem, helps to deeply understand the internal driving mechanism of ecological threats, and can more effectively detect potential ecological risk factors compared with traditional methods.

[0041] Cross-modal feature fusion generates multi-level threat coupling vectors, effectively integrating the key information of different modal data and avoiding the one-sidedness and limitations that may be brought by single-modal data. The multi-level feature fusion provides a more comprehensive and representative feature expression for subsequent threat situation analysis, significantly improving the overall grasp ability of ecological threats.

[0042] Based on the bioacoustic signal set, a bioacoustic anomaly detection map is constructed and spatially superimposed with the multi-level threat coupling vectors. The generated comprehensive threat situation matrix comprehensively presents the ecological threat situation in the protected area. The adaptive weight allocation algorithm further calibrates the dynamic threshold of the matrix, and the output real-time threat level spectrum is more accurate and flexible, capable of adapting to the dynamic changes of the ecological environment in a timely manner and providing a reliable basis for protection decisions.

[0043] Finally, based on the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, the corresponding-level ecological protection intervention instruction set is activated, realizing the seamless connection between monitoring and early warning and protection actions. Thereby, the efficiency and effectiveness of nature reserves in responding to ecological threats are greatly improved.

[0044] In a possible implementation manner, before step S120, the method further includes: Performing noise filtering and trajectory interpolation processing on the animal activity trajectory data to generate a standardized activity trajectory sequence.

[0045] Performing time alignment and outlier correction on the habitat environmental parameter sequence to generate a continuous environmental parameter surface.

[0046] In this embodiment, for the animal activity trajectory data, due to the influence of the device itself or environmental factors, the original data may contain noise. For example, the positioning device may be affected by electromagnetic interference in the complex terrain of the mountainous area, resulting in small jumps or deviations in the recorded position information. If this kind of noisy data is not processed, it will affect the accuracy of subsequent analysis. Therefore, noise filtering is required. Through specific algorithms, such as filtering algorithms based on statistical models, the data points that significantly deviate from the normal trajectory fluctuation range are identified and removed. At the same time, due to some inevitable factors, such as the device being briefly disconnected or the signal transmission being interrupted, the trajectory data may be discontinuous. At this time, trajectory interpolation processing is particularly important. In this embodiment, a suitable interpolation method, such as spline interpolation, is used to supplement reasonable values at the positions where the data is missing according to the trend of the trajectory data before and after, so as to generate a standardized activity trajectory sequence.

[0047] The habitat environmental parameter sequence also faces similar problems. The data collection frequencies of different sensors may vary, which requires time alignment operations. For example, the temperature and humidity sensor may collect data every 10 minutes, while the light intensity sensor may collect data every 15 minutes. To ensure that these data can accurately correspond on the time scale, in this embodiment, with a unified time reference, such as a 10-minute time interval, the data collected by the light intensity sensor is adjusted to align it with the data of the temperature and humidity sensor in time. In addition, the sensor may have outliers due to malfunctions or extreme environmental factors. For example, in rainy weather, the soil humidity sensor may give a temporarily too high humidity value because it is soaked in water. In this embodiment, these outliers are identified and corrected by comparing with the data of surrounding sensors or based on statistical analysis methods of historical data, and then a continuous environmental parameter surface is generated.

[0048] Step S120 includes: Step S121: Perform spatio-temporal grid mapping on the standardized activity trajectory sequence and the continuous environmental parameter surface, and extract the behavior density distribution and environmental parameter gradient within the grid cells.

[0049] Step S122: Compare the behavior density distribution and environmental parameter gradient with the spatio-temporal constraint conditions of the benchmark threat pattern through a threat pattern matching engine, and generate the threat matching confidence of each grid cell.

[0050] Step S123: Construct a three-dimensional threat heat map based on the threat matching confidence of all grid cells as the first ecological association network feature.

[0051] When generating the first ecological association network feature, first perform spatio-temporal grid mapping on the standardized activity trajectory sequence and the continuous environmental parameter surface. The giant panda nature reserve is divided into numerous grid cells, assuming each grid cell is a square area of 1 square kilometer. Within each grid cell, the behavior density distribution of animal activity trajectories is counted. For example, in a certain grid cell, if the giant panda passes by more frequently and stays for a longer time, then the behavior density of this cell is higher. At the same time, analyze the environmental parameter gradient, such as the change in temperature or humidity from one side of the grid cell to the other. If within a certain grid cell, the temperature on the side close to the human activity area is significantly higher than the other side far from the human activity area, this indicates the existence of a certain environmental parameter gradient.

[0052] Next, compare the behavior density distribution and environmental parameter gradient with the spatio-temporal constraint conditions of the benchmark threat pattern through a threat pattern matching engine. The spatio-temporal constraint conditions of the benchmark threat pattern are predefined. For example, when humans frequently appear within a certain range of the giant panda's activity area during a certain period, it is regarded as a threat pattern. If within a certain grid cell, the animal behavior density distribution shows that the giant panda frequently appears near the area with frequent human activities, and the environmental parameter gradient also shows environmental changes caused by human activities in this area, such as increased noise, then the threat matching confidence of this grid cell will be higher. Finally, construct a three-dimensional threat heat map based on the threat matching confidence of all grid cells. In this map, the x and y axes represent the geographical coordinates of the nature reserve, the z axis represents the magnitude of the threat matching confidence, and different colors can represent different threat levels, thus intuitively showing the spatio-temporal coupling relationship between the individual behaviors of animals and potential threat scenarios within the entire nature reserve.

[0053] In a possible implementation manner, step S130 includes: Step S131: Identify the path turning points and stay periods in the standardized activity trajectory sequence, and segment out the migration path sub-segments with continuous motion vectors.

[0054] Step S132: Conduct curvature analysis and velocity variation detection on each sub-segment of the migration path, and mark the abnormal path morphological features.

[0055] Step S133: Conduct causal association analysis between the abnormal path morphological features and the environmental parameter mutation events within the corresponding time window, and calculate the path-environment coupling coefficient.

[0056] Step S134: Construct a dynamic threat propagation chain based on the path-environment coupling coefficients of all sub-segments of the migration path as the second ecological association network feature.

[0057] When analyzing the activity trajectory data of giant pandas to obtain the second ecological association network feature, this embodiment starts from the standardized activity trajectory sequence. Identifying the path turning points and staying periods therein is the key first step. For example, during the migration of giant pandas, when it moves from one bamboo forest to another, its moving direction may change significantly, and this point is the path turning point. Near the water source, giant pandas may stay for a long time to drink water and rest, and this is the staying period. Based on this information, this embodiment can segment the sub-segments of the migration path with continuous motion vectors.

[0058] Conducting curvature analysis and velocity variation detection on each sub-segment of the migration path is an important means to deeply understand the migration behavior of giant pandas. Suppose in a certain sub-segment of the migration path, under normal circumstances, the moving path of giant pandas should be a relatively smooth curve, but if due to certain factors, such as encountering a steep hillside or human activity interference, the path curvature may suddenly increase and become curved. At the same time, velocity variation detection can also detect some abnormal situations. For example, when giant pandas are migrating normally, their speed is relatively stable, but if they suddenly speed up or slow down in a certain section, this may imply that there are abnormal situations in the surrounding environment. This embodiment marks these abnormal path morphological features for subsequent analysis.

[0059] Causal association analysis of abnormal path morphological features and environmental parameter mutation events within the corresponding time window is the core step in constructing the second ecological association network features. For example, when this embodiment discovers that the speed of a giant panda suddenly slows down and the path curvature increases in a certain sub-segment of the migration path, and at the same time, within this time period, the environmental parameters in this area have mutated, such as a sudden drop in temperature leading to changes in the growth condition of bamboo and a reduction in the food resources of the giant panda. This embodiment calculates the path-environment coupling coefficient by analyzing this temporal and spatial correlation. If, in a certain specific sub-segment of the migration path, the mutation of environmental parameters has a very large impact on the migration path and speed of the giant panda, then this path-environment coupling coefficient will be relatively high. Finally, a dynamic threat propagation chain is constructed based on the path-environment coupling coefficients of all migration path sub-segments. This dynamic threat propagation chain can show how environmental parameter mutations affect the migration behavior of giant pandas on different migration paths and how this impact spreads between different path segments, thereby reflecting the intensity of the synergistic effect between abnormal fluctuations in migration paths and environmental parameter mutations.

[0060] Moreover, step S140 includes: Step S141, perform principal component dimensionality reduction processing on the three-dimensional threat heat map to extract the core threat space vector.

[0061] Step S142, convert the dynamic threat propagation chain into a spatio-temporal propagation probability matrix and perform a tensor product operation with the core threat space vector to generate the multi-level threat coupling vector.

[0062] For the three-dimensional threat heat map in the first ecological association network features, since its data dimension is relatively high and there may be some redundant information, this embodiment performs principal component dimensionality reduction processing. Through the principal component analysis algorithm, the principal components that can represent the threat information to the greatest extent are found, so as to extract the core threat space vector. For example, in the three-dimensional threat heat map, there may be multiple factors affecting the threat matching confidence, such as the animal behavior density in different regions, environmental parameter changes, etc. However, through principal component dimensionality reduction processing, this embodiment can find the key factor combination that can best represent the overall threat characteristics and convert it into the core threat space vector.

[0063] Regarding the dynamic threat propagation chain in the second ecological association network feature, in this embodiment, it is converted into a spatio-temporal propagation probability matrix. The construction of this matrix is based on the analysis of the path-environment coupling coefficients of each sub-segment of the migration path in the dynamic threat propagation chain. For example, in this embodiment, according to the path-environment coupling coefficients of each sub-segment of the migration path at different time and space positions, the probability of threat propagation between these positions is calculated. Suppose that in a certain sub-segment of the migration path, the path-environment coupling coefficient is relatively high, then the probability of threat propagation from this sub-segment to the adjacent sub-segment is relatively large, and there will be a relatively high probability value at the corresponding position in the spatio-temporal propagation probability matrix.

[0064] Finally, a tensor product operation is performed on the spatio-temporal propagation probability matrix and the core threat space vector. This operation process integrates two different modalities of information, namely, the spatio-temporal coupling relationship between the individual animal behavior and the potential threat scenario in the first ecological association network feature, and the synergy intensity between the abnormal fluctuations of the migration path and the mutation of environmental parameters in the second ecological association network feature. Through the tensor product operation, a multi-level threat coupling vector is generated. This multi-level threat coupling vector can more comprehensively and accurately reflect the comprehensive threat situation within the giant panda nature reserve, providing a more powerful basis for subsequent threat assessment and formulation of response measures.

[0065] In a possible implementation manner, step S140 further includes: Step S143, perform frame addition and windowing processing on the bioacoustic fingerprint signal set, and extract the Mel frequency cepstral coefficients and harmonic energy ratios of each frame of the signal.

[0066] Step S144, jointly discriminate the Mel frequency cepstral coefficients and harmonic energy ratios through an abnormal acoustic fingerprint classifier, and output an acoustic fingerprint abnormality probability distribution map.

[0067] Step S145, perform spatial clustering on the abnormal acoustic fingerprint events based on the sound source localization algorithm, and generate an acoustic fingerprint abnormality detection map with azimuth markings.

[0068] Step S146, register the azimuth markings of the acoustic fingerprint abnormality detection map with the spatial coordinates of the multi-level threat coupling vector.

[0069] Step S147, calculate the overlapping area ratio and spatial correlation index between the acoustic fingerprint abnormal area and the threat coupling area, and generate the comprehensive threat situation matrix.

[0070] In this embodiment, the bio-acoustic fingerprint signal is a continuous time-series signal. For example, various sound signals emitted by giant pandas, including sounds during foraging, sounds for communicating with companions, etc., as well as calls of other wild animals. The frame addition and windowing process is to segment this continuous signal according to a certain time length to form signal segments frame by frame. Assuming that the length of each frame is set to 0.1 seconds, this time length is determined according to the vocal characteristics of wild animals such as giant pandas and the requirements of signal analysis. The windowing process is to multiply each frame of the signal by a specific window function, such as the Hanning window function, with the aim of reducing spectral leakage and improving the accuracy of spectral analysis.

[0071] After completing the frame addition and windowing process, the Mel-frequency cepstral coefficients and harmonic energy ratios of each frame of the signal are extracted. The Mel-frequency cepstral coefficients are spectral features based on the auditory characteristics of the human ear, and they can effectively describe the frequency characteristics of sound signals. For the call signals of giant pandas, different vocal behaviors will correspond to different Mel-frequency cepstral coefficients. For example, the Mel-frequency cepstral coefficients of the calls of giant pandas during friendly communication and when frightened will have obvious differences. The harmonic energy ratio reflects the energy ratio relationship between the harmonic components and the fundamental component in the sound signal. When different organisms vocalize, their harmonic energy ratios also have unique characteristics. Through a specific algorithm, the Mel-frequency cepstral coefficients and harmonic energy ratios of each frame of the signal are accurately extracted.

[0072] Next, the Mel-frequency cepstral coefficients and harmonic energy ratios are jointly discriminated by an abnormal acoustic fingerprint classifier. The abnormal acoustic fingerprint classifier is a model trained based on a large number of known normal and abnormal bio-acoustic fingerprint sample data. It can judge whether the sound signal of the current frame is an abnormal acoustic fingerprint according to the input Mel-frequency cepstral coefficients and harmonic energy ratio features. For example, when a giant panda is in a normal living state, the Mel-frequency cepstral coefficients and harmonic energy ratios of its calls fluctuate within a certain range. If the corresponding eigenvalue of the sound signal of a certain frame exceeds the normal range, the abnormal acoustic fingerprint classifier will determine it as an abnormal acoustic fingerprint. After discriminating all frames in the bio-acoustic fingerprint signal set, a probability distribution map of acoustic fingerprint abnormality is output. This map is based on the geographical area of the protected area, and each area corresponds to an acoustic fingerprint abnormality probability value. The higher the probability value, the greater the possibility of abnormal acoustic fingerprints in that area.

[0073] Then, spatial clustering is performed on the abnormal voiceprint events based on the sound source localization algorithm. The sound source localization algorithm determines the occurrence location of the abnormal voiceprint events by analyzing information such as the time difference and intensity difference of the abnormal voiceprint signals received by different sensors. For example, multiple voice collection devices are distributed within the protected area. When an abnormal voiceprint event occurs in a certain area, the time and intensity of the signals received by different collection devices will vary. The approximate location of the abnormal voiceprint event is calculated through these difference information. Then, spatial clustering is performed on the abnormal voiceprint events with similar locations and characteristics, and they are grouped into the same type of abnormal voiceprint events. Finally, a voiceprint anomaly detection map with azimuth markings is generated. This map not only shows which areas within the protected area have abnormal voiceprint events but also marks the azimuth information of these events. For example, there is a high-probability abnormal voiceprint event in a certain area in the northeast direction of the protected area.

[0074] After constructing the voiceprint anomaly detection map, spatial overlay analysis is performed on it and the multi-level threat coupling vector to generate a comprehensive threat situation matrix. First, the azimuth markings of the voiceprint anomaly detection map are registered with the spatial coordinates of the multi-level threat coupling vector. The multi-level threat coupling vector contains threat information obtained through the analysis of animal activity trajectory data and habitat environment parameter sequences, and its spatial coordinates correspond to different areas within the protected area. The azimuth markings of the voiceprint anomaly detection map also point to specific areas within the protected area. Through methods such as precise geographical coordinate matching, ensure their accurate spatial correspondence. For example, the abnormal voiceprint area marked in the middle of the protected area in the voiceprint anomaly detection map should be accurately matched with the threat information corresponding to the middle area in the multi-level threat coupling vector.

[0075] After completing the registration, calculate the overlapping area ratio and spatial correlation index between the voiceprint abnormal area and the threat coupling area. The voiceprint abnormal area refers to the area with a relatively high probability of abnormal voiceprint determined in the voiceprint anomaly detection map, and the threat coupling area is the area in the multi-level threat coupling vector that represents a relatively high threat coupling value. The overlapping area ratio is the ratio of the area of the overlapping part of these two areas to their total area, and this ratio reflects the degree of spatial coincidence between the voiceprint anomaly and other threat factors. The spatial correlation index is calculated through a more complex mathematical model, which takes into account factors such as the similarity of the spatial distribution and distance relationship between the two areas. For example, if the voiceprint abnormal area and the threat coupling area mostly overlap, the overlapping area ratio will be relatively high. At the same time, if their spatial distribution shapes and position relationships are also very similar, the spatial correlation index will also be relatively high. By calculating these two indicators, a comprehensive threat situation matrix is generated. This matrix synthesizes the abnormal situations reflected by the biological voiceprint signals and the threat situations analyzed based on animal activity trajectories and habitat environments, and can more comprehensively reflect the comprehensive threat situation within the protected area.

[0076] In a possible implementation, step S140 may further include: Establish a baseline threshold curve for each threat dimension according to the threat level distribution law in the historical threat event database.

[0077] Use the sliding time window algorithm to calculate the cumulative intensity value of each threat dimension in the current comprehensive threat situation matrix in real time.

[0078] Map the cumulative intensity value to the baseline threshold curve through the non-linear interpolation method, and output a dynamically adjusted real-time threat level spectrum.

[0079] Moreover, step S150 may include: Step S151, analyze the weight distribution of each threat dimension in the real-time threat level spectrum, and match the multi-level response protocols in the preset emergency response policy library.

[0080] Step S152, generate a drone patrol path plan, an ecological corridor blockade instruction, and an artificial inspection priority queue according to the spatial coverage range and duration of the threat level.

[0081] Step S153, synchronize the instruction set to the protected area management terminal and the mobile law enforcement device through a low-latency communication network.

[0082] In this embodiment, the historical threat event database stores data related to various threat events that occurred in the giant panda nature reserve in the past. These data detail the threat situations and corresponding threat levels under different threat dimensions. The threat dimensions cover those obtained from previous analyses, such as the animal behavior threat dimension, the habitat environment threat dimension, and the bioacoustic threat dimension. For example, in terms of the animal behavior threat dimension, the database records events where giant pandas change their normal activity patterns due to human activity interference. When giant pandas frequently approach the edge of the protected area or move away from their regular habitat range, they are judged to have different threat levels according to past experience, which may correspond to different degrees of behavioral abnormality and frequency from low to high. For the habitat environment threat dimension, the performance of factors such as abnormal temperature fluctuations, sudden changes in humidity, or vegetation damage in past events and their corresponding threat levels are also recorded. The bioacoustic threat dimension records the association between abnormal bioacoustic events and actual threats, such as the threat levels corresponding to certain abnormal giant panda calls or panicked calls of other wild animals in different situations.

[0083] By deeply analyzing these historical data, a baseline threshold curve is established for each threat dimension. Taking the threat dimension of animal behavior as an example, in this embodiment, it may be found that when the number of times a giant panda approaches the edge of the protected area within a month exceeds a certain amount, the threat level will gradually increase. According to this rule, a curve is drawn, with the abscissa representing the number of times approaching the edge of the protected area and the ordinate representing the threat level. This is the baseline threshold curve for this threat dimension. Similarly, similar curves are constructed for other threat dimensions.

[0084] Next, the sliding time window algorithm is used to calculate the cumulative intensity values of each threat dimension in the current comprehensive threat situation matrix in real time. The sliding time window algorithm sets a specific time window length, for example, 24 hours as a time window. Within this time window, the data in the comprehensive threat situation matrix is continuously updated, and the cumulative intensity value of each threat dimension is calculated. For the threat dimension of animal behavior, if within this 24 hours, the giant panda shows multiple abnormal activity paths or the activity range abnormally shrinks, etc., the threat intensities corresponding to these behaviors will be cumulatively calculated. In terms of the threat dimension of the habitat environment, if the temperature continuously rises or the humidity continuously decreases beyond the normal range within this time window, the threat intensities brought about by these environmental changes will also be cumulatively calculated. The threat dimension of bioacoustic fingerprint is no exception. If the frequency of detecting abnormal bioacoustic fingerprint events increases or the abnormality degree of abnormal bioacoustic fingerprints intensifies within 24 hours, its threat intensity will also be cumulatively calculated.

[0085] Then, the cumulative intensity values are mapped to the baseline threshold curve through the non - linear interpolation method, and a real - time threat level spectrum with dynamic adjustment is output. The non - linear interpolation method takes into account that the relationship between threat intensity and threat level is not a simple linear one. For example, when the cumulative intensity value in the threat dimension of animal behavior is within a certain range, it may correspond to a slow increase in the threat level, but when the cumulative intensity value exceeds a certain critical value, the threat level may increase sharply. The calculated cumulative intensity values are mapped to the previously established baseline threshold curve according to this non - linear relationship, so as to obtain the real - time threat level of each threat dimension. Combining the real - time threat levels of all threat dimensions forms a real - time threat level spectrum. This spectrum can reflect in real time the overall threat level under the combined action of various threat factors within the current giant panda nature reserve, such as being in a low - threat, medium - threat or high - threat level overall.

[0086] When activating the corresponding-level ecological protection intervention instruction set according to the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, it is necessary to first analyze the weight distribution of each threat dimension in the real-time threat level spectrum. The weights of different threat dimensions in the real-time threat level spectrum reflect the relative importance of each dimension in the current overall threat situation. For example, if at a certain moment, the weight of the animal behavior threat dimension is relatively high, this may mean that the abnormal behavior of the giant panda is the current main threat factor, perhaps because the giant panda has undergone serious behavioral changes due to significant external interference. A relatively low weight of the habitat environment threat dimension may indicate that the current environmental factors are relatively stable, and a medium weight of the bioacoustic threat dimension may imply that although there are some abnormal bioacoustic events, their contribution to the overall threat is not the most significant.

[0087] According to this weight distribution, match the multi-level response protocols in the preset emergency response strategy library. The preset emergency response strategy library contains various response strategies for different threat situations. If the weight of the animal behavior threat dimension is high and the overall threat level is high threat, the possible response protocols that may be matched include immediately dispatching professional personnel to conduct close observations on the giant panda to check for injuries or other abnormalities, and at the same time increasing the intensity of the investigation of surrounding human activities to prevent further interference. If the weight of the habitat environment threat dimension suddenly increases, for example, due to extreme weather causing serious deterioration of the habitat environment, the corresponding response protocol may be to initiate an emergency habitat restoration plan, such as urgently replenishing water sources or providing temporary food resources.

[0088] Then, according to the spatial coverage and duration of the threat level, generate the drone patrol path planning, ecological corridor blockade instructions, and the priority queue for manual inspections. If the spatial coverage of the threat level is large, for example, covering most of the protected area, then the drone patrol path planning needs to cover a wider area to ensure comprehensive monitoring of the entire threatened area. For ecological corridors, if the threat level remains high and involves the ecological corridor area, an ecological corridor blockade instruction needs to be issued to prevent external interference factors from entering the core protected area through the corridor. At the same time, determine the priority queue for manual inspections according to the threat level and spatial coverage, list the areas with high threat levels and large coverage as the priority areas for manual inspections, and arrange more inspection personnel and resources for key inspections.

[0089] Finally, the instruction set is synchronized to the protected area management terminal and mobile law enforcement devices through a low-latency communication network. The low-latency communication network ensures that instructions can be transmitted quickly and accurately. After receiving the instruction set, the protected area management terminal can uniformly schedule the resources within the protected area and coordinate all parties to execute corresponding protection intervention measures. After receiving the instruction, the mobile law enforcement device enables law enforcement officers to quickly take actions according to the requirements of the instruction, such as performing the ecological corridor blockade task or conducting manual inspections according to the priority queue, etc., so as to timely and effectively respond to various threat situations within the protected area and ensure the living environment and ecological security of wild animals such as giant pandas.

[0090] In a possible implementation manner, the method further includes: Establish a digital twin model of the ecosystem of the target wildlife nature reserve, and receive and map the multi-modal ecological monitoring data stream in real time.

[0091] Embed a threat evolution simulator in the digital twin model, and predict the threat diffusion path within a future time window based on the current comprehensive threat situation matrix.

[0092] Optimize the execution timing and resource allocation plan of the ecological protection intervention instruction set according to the prediction results.

[0093] Among them, the update step of the digital twin model includes: Collect the ecological status feedback data after the actual intervention measures are executed, and calculate the threat suppression efficiency coefficient.

[0094] Input the threat suppression efficiency coefficient into the threat evolution simulator for backpropagation training, and update the prediction parameters of the threat diffusion path.

[0095] When the threat suppression efficiency coefficient is lower than the preset threshold, trigger a secondary response mechanism and regenerate the cross-regional collaborative intervention instruction set.

[0096] In this embodiment, the digital twin model is a comprehensive digital mapping of the ecosystem of the giant panda nature reserve, covering numerous elements such as the topography, vegetation distribution, water source conditions, and the distribution and activity patterns of various wild animals within the reserve. The multi-modal ecological monitoring data stream contains rich information such as the previously mentioned animal activity trajectory data, habitat environment parameter sequences, and bioacoustic signal sets. For example, for the animal activity trajectory data, the corresponding module in the digital twin model will accurately depict the activity paths of giant pandas within the reserve based on the received data, including their migration routes between bamboo forests in different seasons and the residence time at water sources. Data such as temperature, humidity, light intensity, and soil humidity in the habitat environment parameter sequences will be mapped to the corresponding geographical areas in the digital twin model, thus reflecting the environmental status of different regions. The bioacoustic signal set will also have corresponding manifestations in the digital twin model. For example, when a specific call signal of a giant panda is received, the model can determine the approximate location where the call is emitted and its possible meaning (such as indicating warning or courtship, etc.) based on pre-analysis.

[0097] Next, a threat evolution simulator is embedded in the digital twin model to predict the threat diffusion path within the future time window based on the current comprehensive threat situation matrix. The comprehensive threat situation matrix integrates threat information from multiple aspects such as animal activity trajectories, habitat environments, and bioacoustics. The threat evolution simulator uses this information and combines the interaction relationships between various elements in the ecosystem for simulation and prediction. For example, if the current comprehensive threat situation matrix shows a high threat of human activities (such as illegal human entry or construction activities) near a certain habitat of giant pandas, and the habitat environment has also deteriorated to a certain extent (such as the reduction of bamboo forests due to pollution), the threat evolution simulator will, based on these factors and the interaction relationships within the ecosystem, predict the possible diffusion path of this threat to the surrounding areas within a future period (such as within the next week). It will take into account the activity habits of giant pandas. For example, giant pandas may migrate to other areas due to the threat to their habitat, and this migration may further exert pressure on the new regional ecology, thus affecting the threat diffusion path. At the same time, the status of other wild animals reflected by the bioacoustic signals will also be taken into consideration. If other animals change their activity ranges or behavior patterns due to the threat, this will also have an impact on the spread of the threat.

[0098] Then, optimize the execution timing and resource allocation plan of the ecological protection intervention instruction set according to the prediction results. If the prediction results show that the threat in a certain area will increase sharply in the next few days, then the execution timing of the ecological protection intervention instruction set for that area needs to be advanced. For example, if it was originally planned to inspect and maintain a potentially threatened ecological corridor in a week, according to the prediction results, it needs to be adjusted to be carried out immediately or within a shorter time. In terms of resource allocation, if the threat in a certain area spreads quickly and has a large impact range, more resources need to be allocated to that area. For example, increase the patrol frequency of drones in that area, dispatch more human inspectors, or invest more equipment for environmental restoration, etc.

[0099] In terms of the update step of the digital twin model, first collect the ecological status feedback data after the actual intervention measures are implemented. When a series of ecological protection intervention instruction sets are executed for the threat situation in the protected area, such as carrying out habitat restoration work, strengthening the control of human activities, or treating injured giant pandas, etc., it is necessary to collect the feedback data of all aspects of the ecosystem after these intervention measures are implemented. These data include the vegetation restoration situation of the restored habitat, whether the behavior of the giant pandas has returned to normal, whether the illegal human activities have been effectively curbed, etc.

[0100] Then, calculate the threat suppression efficiency coefficient. This coefficient is an important indicator to measure the threat suppression effect of ecological protection intervention measures. For example, if a series of measures are implemented in a certain area to reduce the threat of human activities to giant pandas, by comparing various factors such as the frequency of illegal human activities in that area before and after the intervention measures, the recovery of the behavior of giant pandas, and the improvement of the habitat environment, a quantified threat suppression efficiency coefficient is calculated. If, after the intervention, illegal human activities have almost disappeared, the activities of giant pandas have returned to the normal mode, and the habitat environment has also been significantly improved, then the threat suppression efficiency coefficient will be higher; on the contrary, if there are still many illegal human activities after the intervention, the behavior of giant pandas has not improved significantly, and the environment has not been effectively improved, then the threat suppression efficiency coefficient will be lower.

[0101] Input the threat suppression efficiency coefficient into the threat evolution simulator for backpropagation training to update the prediction parameters of the threat diffusion path. When the threat evolution simulator is initially constructed, there may be certain errors or inaccuracies in its prediction parameters. By inputting the calculated threat suppression efficiency coefficient for backpropagation training, the parameters in the simulator can be adjusted to make its prediction more accurate. For example, if the threat suppression efficiency coefficient in a certain area is low, it indicates that there are biases in the simulator's previous understanding of the relationships between ecosystem elements and the threat propagation mechanism in this area. Through backpropagation training, parameters such as the association weights between ecological elements related to this area and the threat propagation speed are adjusted to improve the prediction accuracy of the threat diffusion path.

[0102] When the threat suppression efficiency coefficient is lower than the preset threshold, trigger the secondary response mechanism and regenerate the cross-regional collaborative intervention instruction set. The preset threshold is a critical value set based on the ecological protection objectives of the protected area and past experience. If the calculated threat suppression efficiency coefficient is lower than this preset threshold, it indicates that the current intervention measures are not effective and more powerful measures need to be taken. The secondary response mechanism involves cross-regional collaborative actions. For example, if the threat suppression effect is not good in a certain local area of the giant panda nature reserve, more resources may need to be allocated from the entire nature reserve and even surrounding related areas. The regenerated cross-regional collaborative intervention instruction set may include coordinating surrounding protected areas to jointly strengthen the protection of the giant panda migration route, jointly carrying out large-scale habitat restoration projects, and uniformly dispatching more law enforcement forces to crack down on human illegal activities to ensure the effective protection of the ecosystem of the giant panda nature reserve.

[0103] In a possible implementation manner, the method further includes: Construct a human activity monitoring network around the target wildlife nature reserve, and collect traffic flow data, nighttime light intensity distribution, and illegal intrusion alarm signals.

[0104] Cross-validate the human activity monitoring data with the comprehensive threat situation matrix to identify compound ecological threats induced by human factors.

[0105] Automatically generate law enforcement evidence collection clue packages and public warning information push strategies according to the compound threat characteristics.

[0106] Among them, the step of cross-validating the human activity monitoring data with the comprehensive threat situation matrix to identify compound ecological threats induced by human factors includes: Analyze the spatio-temporal coupling degree between the spatio-temporal trajectories of human activities and animal abnormal behavior events, and calculate the human interference contribution factor.

[0107] Evaluate the long-term cumulative effects of human activities on ecological threats in combination with the persistent change trends of habitat environmental parameters.

[0108] When the anthropogenic disturbance contribution factor exceeds the critical value, activate the directional tracking and image acquisition functions of the high-precision video surveillance device.

[0109] In this embodiment, traffic flow monitoring devices, such as induction coils or video monitoring devices, are set on the roads around the giant panda nature reserve. These devices can accurately record the vehicle passing numbers and types at different times. Traffic flow data is crucial for understanding the potential impacts of human activities on the reserve because vehicle passing may bring impacts such as noise and exhaust emissions, and may also imply the movement of people. The monitoring of the nocturnal light intensity distribution is achieved by light intensity sensors distributed at different positions around the reserve. For example, in some villages or human activity areas near the edge of the reserve, if there is abnormal high-intensity light at night, it may imply the existence of illegal nocturnal activities. For example, poachers may use strong light for searching or lighting, or some construction activities produce strong light interference at night. The illegal intrusion alarm signal is obtained through fence sensors, infrared monitoring devices, etc. set at the boundary of the reserve. When someone or a large animal crosses the fence or triggers the infrared monitoring, an illegal intrusion alarm signal will be generated, which directly reflects the possible illegal entry behavior into the reserve.

[0110] Cross-validate the human activity monitoring data with the comprehensive threat situation matrix to identify the compound ecological threats induced by human factors. The comprehensive threat situation matrix integrates threat information from multiple aspects, including animal activity trajectories, habitat environments, and biological sound patterns. First, analyze the spatio-temporal coupling degree between the spatio-temporal trajectories of human activities and animal abnormal behavior events, and calculate the anthropogenic disturbance contribution factor. For example, when the traffic flow monitoring device shows that the number of vehicles leading to the reserve on a certain road section increases during a certain period, and at the same time, the activity trajectory of the giant pandas in the reserve shows that they move away from that road section and their behavior becomes abnormal, such as the activity frequency increases and the original regular foraging and resting patterns are disrupted. Through precise time and space coordinate matching, analyze the correlation degree between this spatio-temporal trajectory of human activities and the animal abnormal behavior event in terms of time and space, and calculate the anthropogenic disturbance contribution factor. If this correlation is very close, for example, the giant pandas show obvious abnormal behavior shortly after the increase in vehicle numbers, then the anthropogenic disturbance contribution factor will be relatively high.

[0111] Next, combined with the persistent change trend of habitat environmental parameters, evaluate the long-term cumulative effects of human activities on ecological threats. For example, the exhaust emissions brought about by long-term traffic flow may lead to a decline in air quality around protected areas. By monitoring the persistent change trend of habitat environmental parameters such as pollutant concentrations in the air, the long-term impact of such human activities on the ecosystem can be evaluated. If it is found in years of monitoring that as the traffic flow increases year by year, the vegetation growth around the protected area is inhibited and the soil quality deteriorates, etc., this indicates that there is a long-term cumulative effect of human activities on ecological threats. If within a certain period, not only does the traffic flow increase, but also the night light intensity abnormally rises, and the habitat environment of giant pandas continues to deteriorate and the abnormal animal behavior increases, this indicates that there may be a compound ecological threat induced by human factors.

[0112] When the anthropogenic disturbance contribution factor exceeds the critical value, activate the directional tracking and image acquisition functions of high-precision video surveillance devices. The critical value is set based on long-term research on the protected area ecosystem and past experience. For example, if past research shows that when the anthropogenic disturbance contribution factor reaches 0.8, it may pose a serious threat to the survival of giant pandas and the ecosystem. When the calculated anthropogenic disturbance contribution factor exceeds this critical value, the high-precision video surveillance devices distributed at key locations within the protected area will be activated. These devices have high resolution and precise directional tracking capabilities, and can track and photograph specific areas or targets. For example, if it is found that human activities are concentrated near the core habitat of giant pandas and the anthropogenic disturbance contribution factor exceeds the critical value, the video surveillance devices will conduct directional tracking on this area, record image information such as the movement trajectories and appearance characteristics of the people entering this area, and provide strong evidence for subsequent law enforcement and analysis.

[0113] Automatically generate law enforcement evidence collection clue packages and public warning information push strategies based on the characteristics of complex threats. When a complex ecological threat induced by human factors is identified, law enforcement evidence collection clue packages are automatically generated according to the specific characteristics of the threat. For example, if it is found that the area with abnormally increased night light intensity overlaps with the activity area of giant pandas, and combined with the illegal intrusion alarm signal to determine the existence of illegal entry behavior, the law enforcement evidence collection clue package will include the light intensity data of this area, the time and location information of the illegal intrusion alarm, and relevant video surveillance images. These information can be directly provided to law enforcement officers so that they can conduct law enforcement investigations quickly and accurately. At the same time, a public warning information push strategy is generated according to the characteristics of complex threats. If the complex threat mainly comes from human activities in surrounding villages, such as excessive tourism development or agricultural activities posing a threat to the protected area, then the public warning information push strategy may be to push information about the importance of protecting giant pandas and the harm their activities may cause to giant pandas to the residents of surrounding villages, which can be pushed through mobile phone text messages, community bulletin boards or special environmental protection publicity platforms to improve the public's environmental awareness and reduce the threat of human factors to the ecological system of the protected area.

[0114] In a possible implementation manner, the method further includes: Deploy edge computing devices at key ecological nodes of the target wildlife nature reserve to perform real-time feature extraction on the multi-modal ecological monitoring data stream collected locally.

[0115] Encrypt and aggregate the feature extraction results of each node through a federated learning framework to update the global threat detection model parameters.

[0116] Dynamically distribute the updated model parameters to each edge computing device to achieve the co-evolution of distributed threat perception capabilities.

[0117] Among them, the update steps of the federated learning framework include: Design a differential privacy protection strategy to add geographical mask noise to sensitive animal location information.

[0118] Adopt the model parameter differential aggregation technology to eliminate the influence of the data distribution skew of edge nodes on the global model.

[0119] When it is detected that a node device is abnormally offline, start the model parameter rollback mechanism and reallocate the computing tasks to neighboring nodes.

[0120] In this embodiment, the key ecological nodes include the main habitats of giant pandas, such as the core bamboo forest areas, water sources, and important migration corridors. Edge computing devices are deployed at these locations. For example, small, low-power edge computing devices are installed on multiple bamboo trees in the core bamboo forest area. These devices can perform real-time processing on the multi-modal ecological monitoring data streams collected locally. For animal activity trajectory data, the edge computing devices can directly perform preliminary analysis on the raw location information transmitted by the positioning devices, and extract features such as the activity range boundary and frequently active areas of giant pandas in this area. Taking the activities of giant pandas in the bamboo forest as an example, the edge computing devices can determine the sub-areas of the bamboo forest where giant pandas often move based on the location data over a period of time, as well as features such as whether their activity range has a tendency to expand or contract towards the edge of the bamboo forest. For the sequence of habitat environmental parameters, the edge computing devices can perform real-time analysis on the data collected by local temperature and humidity sensors, soil humidity sensors, etc., and extract the change trend features of the environmental parameters. For example, the edge computing devices near the water source can quickly determine the short-term fluctuation features of the soil humidity around the water source, as well as the long-term humidity change trend related to seasonal changes. For the biological acoustic fingerprint signal set, the edge computing devices can perform preliminary processing on the sound signals collected locally, and extract features such as the frequency distribution features of biological acoustic fingerprints and the change features of the vocalization frequency within a specific time period. For example, during the breeding season of giant pandas, the edge computing devices can analyze whether the change in the vocalization frequency of giant pandas in this area is related to breeding behavior.

[0121] The feature extraction results of each node are encrypted and aggregated through the federated learning framework to update the parameters of the global threat detection model. After edge computing devices for feature extraction are available at multiple key ecological nodes within the protected area, the federated learning framework is used to integrate these scattered feature extraction results. The federated learning framework uses encryption technology to ensure the security and privacy of data. For example, each edge computing device encrypts the extracted feature results and sends them to a central node (the data remains encrypted throughout this process to ensure that sensitive information is not leaked). At the central node, aggregation operations are performed on these encrypted feature results. Suppose one node extracts the feature that the activity range of giant pandas in a certain area has shrunk, and another node extracts the feature that the environmental temperature in this area has risen abnormally. Through the encrypted aggregation of the federated learning framework, these features from different aspects are combined to update the parameters of the global threat detection model. The global threat detection model is a model for comprehensively evaluating ecological threats within the protected area, and the update of its parameters enables it to more accurately identify and evaluate threats. For example, as the features of each node are continuously aggregated and updated, the model can more precisely determine whether there is a higher ecological threat and the type and degree of the threat when the activity range of giant pandas shrinks and the environmental temperature rises abnormally.

[0122] Dynamically distribute the updated model parameters to each edge computing device to achieve the co-evolution of distributed threat perception capabilities. After updating the parameters of the global threat detection model, these new parameters are dynamically distributed to each edge computing device. For example, when the global threat detection model adjusts the recognition parameters for a certain threat pattern (such as the shrinking of the giant panda's habitat caused by human activities) based on new aggregated features, these new parameters are sent to each edge computing device. On each edge computing device, using these new model parameters, they can more accurately identify threats when analyzing locally collected data. Taking the edge computing device in the bamboo forest area as an example, when it receives the new model parameters, it can more precisely determine whether there are threats caused by human activities (such as the logging of surrounding bamboo forests) when analyzing the activity trajectory data, environmental parameter data, and bioacoustic data of giant pandas in the bamboo forest, thus achieving the co-evolution of distributed threat perception capabilities. The threat perception ability of each edge computing device is continuously improved with the update of the global model parameters, and this improvement will in turn feedback to the next round of feature extraction and global model update, forming a virtuous cycle.

[0123] In terms of the update step of the federated learning framework, first design a differential privacy protection strategy to add geographical masking noise to sensitive animal location information. Since the location information of wild animals such as giant pandas is very sensitive data, in order to protect their privacy while enabling effective data processing, a differential privacy protection strategy is designed. For example, for the precise location information of giant pandas in the core habitat, geographical masking noise is added when transmitting its data to other nodes or the central node in the federated learning framework. This geographical masking noise is not simple random noise, but is designed based on the geographical characteristics of the protected area and the animal's activity habits. For instance, according to the topography and landforms of the giant panda's habitat, its actual location is blurred within a certain range. If a giant panda is at a precise coordinate position in a valley, after adding geographical masking noise, during data transmission, other nodes receive a blurred location information within that valley range, which not only protects the precise location privacy of the giant panda but also does not affect the overall data analysis and model training.

[0124] Adopt the model parameter difference aggregation technology to eliminate the influence of the data distribution skew of edge nodes on the global model. At different key ecological nodes within the protected area, due to differences in geographical environment, ecological conditions and other factors, the data distribution collected by edge computing devices may be skewed. For example, edge computing devices near the water source may collect more data on changes in water-related environmental parameters, while edge computing devices in the bamboo forest area may collect more data on environmental parameters related to bamboo growth. This skew in data distribution may affect the accuracy of the global threat detection model. Through the model parameter difference aggregation technology, when aggregating the feature extraction results of each node, this difference in data distribution is compensated. Specifically, when calculating the global model parameters, instead of simply averaging the parameters of each node, weighted processing is performed according to the distribution characteristics of the data of each node. For example, for nodes with less collected data but more critical data features, a higher weight is given, thereby eliminating the influence of the data distribution skew of edge nodes on the global model and ensuring that the global threat detection model can accurately reflect the ecological status of the entire protected area.

[0125] When a node device is detected to be abnormally offline, start the model parameter rollback mechanism and reallocate the computing tasks to neighboring nodes. During the actual operation of the protected area, due to equipment failures, power supply problems or other unexpected situations, node devices may be abnormally offline. For example, an edge computing device in the bamboo forest suddenly goes offline due to a power outage caused by bad weather. When this situation is detected, the model parameter rollback mechanism is started. The model parameter rollback mechanism refers to rolling back the parameters of the global threat detection model to a previous stable version to avoid errors in model parameter updates caused by the offline of this node. At the same time, reallocate the computing tasks to neighboring nodes. For example, reallocate some of the computing tasks in the bamboo forest area originally responsible by the offline device to other edge computing devices in the surrounding bamboo forest area. After receiving the new computing tasks, these neighboring nodes will continue to perform feature extraction and data processing work using the data they collect and some relevant data stored previously, ensuring that the ecological monitoring and threat detection work of the entire protected area can proceed continuously and stably.

[0126] In a possible implementation manner, the method further includes: Integrate the historical ecological restoration records and species reproduction data of the target wildlife nature reserve to construct an ecological system resilience assessment matrix.

[0127] Perform multi-dimensional correlation analysis on the real-time threat level spectrum and the ecological system resilience assessment matrix to predict the ecological restoration potential under different intervention strategies.

[0128] Dynamically adjust the implementation intensity and action period of the ecological protection intervention instruction set according to the prediction results.

[0129] Among them, the step of performing multi-dimensional correlation analysis on the real-time threat level pedigree and the ecosystem resilience assessment matrix to predict the ecological restoration potential under different intervention strategies includes: Establish a threat intensity - restoration efficiency response surface model to quantify the restoration rate of key ecosystem indicators under different intervention measures.

[0130] Simulate the ecological resilience transition path under the superposition effect of consecutive rounds of intervention measures to identify the optimal combination of intervention strategies.

[0131] When it is predicted that the restoration rate is lower than the degradation rate, trigger an expert consultation request and activate the zoning control plan for protected areas.

[0132] In this embodiment, the historical ecological restoration records cover the detailed information of various ecological restoration projects implemented in the giant panda nature reserve in the past. For example, the restoration work on damaged bamboo forests, including the time, scope, restoration methods (such as the variety and quantity of replanted bamboo, measures to improve soil conditions, etc.) and the effect evaluation after restoration. The species reproduction data includes the reproduction situation of giant pandas and other related species, such as the reproduction cycle, reproduction success rate, survival rate of cubs of giant pandas, and the population quantity change and reproduction-related data of other wild animals. By integrating these data, an ecosystem resilience assessment matrix is constructed. This matrix evaluates the resilience of the protected area ecosystem from multiple dimensions. For example, it quantitatively evaluates from aspects such as vegetation restoration ability, species reproduction stability, and the resistance ability of the ecosystem to external disturbances. Taking the bamboo forest in the giant panda habitat as an example, in the ecosystem resilience assessment matrix, the self-renewal ability of the bamboo forest, the stability of the food supply for giant pandas, and the recovery ability in the face of natural disasters (such as fires, pests and diseases) will all be quantified as indicators in the matrix.

[0133] Perform multi-dimensional correlation analysis on the real-time threat level pedigree and the ecosystem resilience assessment matrix to predict the ecological restoration potential under different intervention strategies. The real-time threat level pedigree reflects the various threat situations in the current protected area, including the threat level situation after integrating multiple factors such as animal behavior threats, habitat environment threats, and bioacoustic threats. Multi-dimensional correlation analysis is to explore the relationship between the threat level and ecosystem resilience from multiple dimensions. For example, when the habitat environment threat level in the real-time threat level pedigree is relatively high, such as there is water pollution and bamboo forest degradation, and at the same time, in the ecosystem resilience assessment matrix, the self-renewal ability of the bamboo forest is weak, it is necessary to analyze the improvement effect of different intervention strategies on this situation.

[0134] A threat intensity - restoration efficiency response surface model is established to quantify the restoration rates of key ecosystem indicators under different intervention measures. For various ecological problems in giant panda nature reserves, different intervention measures will have different effects. For example, regarding the problem of bamboo forest degradation, if the intervention measure is to massively replant bamboo varieties suitable for the local environment and improve soil fertility at the same time, then the threat intensity - restoration efficiency response surface model can be used to quantify the restoration rates of key ecosystem indicators such as the restoration of bamboo forest area and the improvement of bamboo quality for this intervention measure. This model comprehensively considers the relationship between the intensity of the threat (such as the degree of bamboo forest degradation, including the proportion of reduced bamboo quantity, the proportion of poorly growing bamboos, etc.) and the restoration efficiency (such as the increase in bamboo forest area within a certain period of time, the health status of new bamboos, etc.). Suppose the degree of bamboo forest degradation reaches 30%. After one year of taking the above - mentioned intervention measures, through model calculation, it is obtained that the bamboo forest area has increased by 10% and the proportion of healthy - growing bamboos has increased by 15%. This is the quantification of the restoration rate of this intervention measure. Similarly, for other ecological problems, such as water pollution treatment and the restoration of habitats for other wild animals in the nature reserve, the restoration rates of different intervention measures for the corresponding key ecosystem indicators can be quantified through this model.

[0135] Simulate the ecological resilience transition path under the superposition effect of consecutive rounds of intervention measures to identify the optimal combination of intervention strategies. In actual conservation work, multiple rounds of intervention measures are often required. For example, for the overall restoration of giant panda habitats, the first round may be to treat water pollution first, the second round is to improve the growth environment of bamboo forests, and the third round is to strengthen the protection measures for giant panda reproduction, etc. By simulating the ecological resilience transition path under the superposition effect of these consecutive rounds of intervention measures, we can see how the resilience of the entire ecosystem gradually changes. For example, after the first - round water pollution treatment, although the direct impact on giant pandas may not be obvious, it has a certain improvement on other water - dependent organisms and the health of the entire ecosystem, and the ecological system resilience has a small increase; after the second - round improvement of the bamboo forest growth environment, the food resources of giant pandas increase, the habitat quality improves, and the ecological system resilience is further enhanced; after the third - round strengthening of reproduction protection measures, the reproduction of the giant panda population is more guaranteed, and the stability and resilience of the entire ecosystem are greatly improved. Through this simulation, it can be identified which combination of intervention strategies can achieve the maximum increase in ecological system resilience in the shortest time with the least resource input, thereby determining the optimal combination of intervention strategies.

[0136] When it is predicted that the restoration rate is lower than the degradation rate, an expert consultation request is triggered and the zoning control plan for protected areas is initiated. If through the above analysis and simulation, it is found that for a certain ecological problem, such as the continuous degradation of bamboo forests in some areas, although some intervention measures have been taken, the predicted restoration rate is lower than the degradation rate, this indicates that the current situation is very serious. At this time, an expert consultation request is triggered, and experts in related fields such as giant panda protection, ecological restoration, and forestry are convened for consultation. These experts will conduct in-depth analysis of the current situation based on their professional knowledge and experience. At the same time, the zoning control plan for protected areas is initiated. For example, the natural protected areas for giant pandas are divided into different regions according to factors such as ecological conditions and threat levels. For severely degraded regions, more stringent control measures may be taken, such as restricting human access, increasing the monitoring frequency of the regional ecosystem, and increasing resource investment for special restoration, to ensure the stability of the entire protected area ecosystem and effectively protect the living environment of wild animals such as giant pandas.

[0137] Dynamically adjust the implementation intensity and action cycle of the ecological protection intervention instruction set according to the prediction results. If the prediction results show that a certain intervention measure has a good effect on ecological restoration, for example, the intervention measure for bamboo forest restoration makes the restoration rate of the bamboo forest higher than the degradation rate and can achieve the expected ecological restoration goal in a short time, then the implementation intensity of this intervention measure in the ecological protection intervention instruction set can be appropriately reduced, or its action cycle can be shortened, and the resources can be transferred to other places where they are needed. On the contrary, if the prediction results indicate that the effect of a certain intervention measure is not good, such as the intervention measure for water pollution control has been implemented for some time, but the water pollution situation has not improved significantly, then it is necessary to increase the implementation intensity of this intervention measure, extend its action cycle, or adjust the specific content of the intervention measure, such as replacing the technology or method for treating pollution, etc., to improve the protection and restoration effect on the ecosystem.

[0138] For example, in a possible implementation manner, the method further includes: Deploy intelligent acoustic and light deterrence devices at the boundary of the target wildlife natural protected area, and the triggering logic of the intelligent acoustic and light deterrence devices is dynamically bound to the real-time threat level spectrum.

[0139] When illegal intrusion or high-risk animal behavior is detected, start the directional sound wave interference and laser warning mode according to the threat direction and level.

[0140] Record the spatio-temporal distribution characteristics of the deterrence device triggering events, which are used to optimize the sensitivity parameters of the subsequent early warning algorithm.

[0141] Among them, the control steps of the intelligent acoustic and light deterrence device include: Establish a species adaptation deterrence strategy library according to the sound-sensitive frequency bands and light response characteristics of different animal species.

[0142] Combine real-time meteorological conditions and terrain shielding coefficients to dynamically adjust the acoustic wave propagation angle and light intensity attenuation compensation parameters.

[0143] When the deterrence trigger frequency in the same area exceeds the set frequency threshold, control the intelligent acoustic and optical deterrence device to switch to a continuous active monitoring mode and increase the image sampling rate.

[0144] In this embodiment, the boundary of the target wildlife nature reserve is the first line of defense against external interference and illegal intrusion. Intelligent acoustic and optical deterrence devices are deployed along the boundary. These intelligent acoustic and optical deterrence devices can obtain threat information in the reserve in real time, and this information comes from a real-time threat level spectrum. The real-time threat level spectrum integrates threat situations in multiple aspects such as animal activity trajectories, habitat environments, and biological soundprints, and can accurately reflect the overall threat level in the reserve. For example, when the animal activity trajectory in the giant panda nature reserve shows that a large number of unknown animals are approaching the boundary, or the habitat environment parameters indicate that there are abnormal interference sources near the boundary, or the biological soundprint signal set detects abnormal animal calls in the boundary area, the real-time threat level spectrum will comprehensively evaluate these threats and give corresponding levels. The intelligent acoustic and optical deterrence device adjusts its trigger sensitivity according to this dynamically changing threat level. If the threat level is low, the device may be in a standby state with relatively low sensitivity; when the threat level increases, the device is ready to trigger countermeasures at any time.

[0145] When illegal intrusion or high-risk animal behavior is detected, the directional acoustic interference and laser warning modes are activated according to the threat direction and level. A variety of monitoring devices, such as infrared sensors and video surveillance equipment, are set up at the boundaries of the protected area to detect illegal intrusion behavior. For example, when the infrared sensor detects someone crossing the boundary of the protected area, or the video surveillance equipment identifies unauthorized personnel or vehicles entering the protected area, this is determined as illegal intrusion behavior. At the same time, the detection of high-risk animal behavior is also based on multi-modal ecological monitoring data streams. For example, when the animal activity trajectory data shows that aggressive wild animals are frequently approaching the boundary and their behavior is extremely agitated, this is regarded as high-risk animal behavior. Once these situations are detected, the intelligent acoustic-optic deterrence device will respond according to the direction and level of the threat. If the threat comes from the northeast direction of the protected area, the device will direct the directional acoustic interference and laser warning modes towards the northeast direction. The judgment of the threat level also determines the intensity of the response measures. If it is a low-level threat, it may only emit low-intensity acoustic interference and softer laser warnings briefly; while if it is a high-level threat, such as a large-scale illegal intrusion or an extremely dangerous animal approaching, then a high-intensity directional acoustic interference and strong laser warning mode will be activated. The directional acoustic interference uses sound waves with specific frequencies and intensities to drive away illegal intruders or deter high-risk animals. The frequency and intensity of these sound waves are carefully designed to achieve the deterrence purpose without causing unnecessary harm to the organisms and environment within the protected area. The laser warning mode warns illegal intruders or high-risk animals by emitting bright laser beams, making them stay away from the boundary of the protected area.

[0146] Record the spatio-temporal distribution characteristics of the deterrence device trigger events for optimizing the sensitivity parameters of the subsequent early warning algorithm. Each time the intelligent acoustic-optic deterrence device is triggered, the spatio-temporal distribution characteristics of the trigger event will be recorded in detail. For example, record that at a specific time (such as 10 am on May 10, 2023), a deterrence device trigger event occurred in the southwest area of the protected area boundary because illegal intrusion behavior was detected. Over time, a large number of such trigger event records have been accumulated. These records are of great significance for optimizing the sensitivity parameters of the subsequent early warning algorithm. If deterrence device trigger events frequently occur in a certain area, it may indicate that the threat situation in this area is more complex or the current early warning algorithm is not accurate enough in terms of sensitivity in this area. By analyzing these spatio-temporal distribution characteristics, the sensitivity parameters of the early warning algorithm can be adjusted. For example, if illegal intrusion events frequently occur at night in a certain area but the early warning algorithm fails to trigger the deterrence device in a timely manner, then it can be analyzed based on the records that the sensitivity of the early warning algorithm at night in this area needs to be improved so that the intelligent acoustic-optic deterrence device can respond to threats more timely and accurately.

[0147] In terms of the control steps of the intelligent acoustic and optical deterrence device, first, a species-adaptive deterrence strategy library is established based on the sound-sensitive frequency bands and light response characteristics of different animal species. There are multiple animal species in the giant panda nature reserve, and different animals have different sensitive frequency bands and response characteristics to sound and light. For example, giant pandas may be more sensitive to certain low-frequency sound waves, while some small mammals may be more sensitive to high-frequency sound waves. Regarding the light response characteristics, some nocturnal animals may have a strong avoidance reaction to sudden strong light, while diurnal animals may have different responses to specific colors of light. Through research and experiments on these different animal species, a species-adaptive deterrence strategy library is established. In this strategy library, appropriate acoustic and optical deterrence plans are formulated for each animal. When a specific animal is detected approaching the reserve boundary and the deterrence device needs to be activated, the operation can be carried out according to the plan in this strategy library. For example, when a giant panda is found approaching the boundary and behaving abnormally, according to the deterrence plan for giant pandas in the strategy library, select sound waves with appropriate frequencies and intensities and lasers with specific colors and intensities for warning, which can not only achieve the deterrence purpose but also avoid causing excessive fright or harm to the giant panda.

[0148] Combined with real-time meteorological conditions and terrain shielding coefficients, dynamically adjust the acoustic wave propagation angle and light intensity attenuation compensation parameters. Meteorological conditions and terrain factors have an important impact on the effect of the acoustic and optical deterrence device. For example, in windy weather conditions, the propagation direction and distance of sound waves will be affected. If the wind direction is from the inside of the reserve to the boundary, the sound waves may be scattered and unable to effectively reach the target area. At this time, it is necessary to dynamically adjust the acoustic wave propagation angle according to the wind direction and speed to ensure that the sound waves can accurately reach the direction where the threat source is located. Similarly, the terrain shielding coefficient will also affect the effect of acoustic and optical deterrence. If there are terrain obstacles such as mountains or tall trees at the reserve boundary, it will block the propagation of sound waves and light. By measuring and analyzing the terrain, calculate the terrain shielding coefficient, and adjust the light intensity attenuation compensation parameters according to this coefficient. If in an area with more shielding objects, it is necessary to increase the light intensity of the laser to ensure that the warning effect can penetrate the shielding objects and be received by the target.

[0149] When the deterrence trigger frequency in the same area exceeds the set frequency threshold, control the intelligent audible and visual deterrence device to switch to a continuous active monitoring mode and increase the image sampling rate. In some areas of the protected area, frequent threat events may occur, resulting in an overly high deterrence trigger frequency in the same area. For example, at the boundary of the protected area near areas with frequent human activities, there may often be cases of people accidentally entering or small animals frequently moving, triggering the deterrence device. When this deterrence trigger frequency exceeds the set frequency threshold, in order to better understand the situation in this area and conduct effective management, control the intelligent audible and visual deterrence device to switch to a continuous active monitoring mode. In this mode, the device no longer just waits passively for triggering, but continuously monitors this area. At the same time, increase the image sampling rate and collect image information of this area more frequently through video surveillance equipment. This helps to observe the activities in this area in more detail, determine whether there is a continuous threat source or whether it is necessary to further adjust the deterrence strategy. For example, if it is found that the frequent triggering of the device is due to the nest of a small animal being located in this area, then targeted measures can be taken based on the more detailed image information, such as adjusting the trigger sensitivity of the deterrence device or changing the deterrence method without affecting the survival of the small animal.

[0150] Figure 2 The hardware structure diagram of the monitoring and early warning system 100 for a wildlife natural protected area provided by an embodiment of the present invention to implement the above-mentioned monitoring and early warning method for a wildlife natural protected area is shown, as Figure 2 shown, the monitoring and early warning system 100 for a wildlife natural protected area may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0151] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 can store the data and / or instructions used by the monitoring and early warning system 100 for a wildlife natural protected area to execute or use to complete the exemplary methods described in the present invention.

[0152] In the specific implementation process, one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the monitoring and early warning method for a wildlife natural protected area in the above method embodiment. The processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiver actions of the communication unit 140.

[0153] For the specific implementation process of the processor 110, reference can be made to the various method embodiments executed by the above-mentioned monitoring and early warning system 100 for wildlife nature reserves. Their implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0154] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned monitoring and early warning method for wildlife nature reserves is implemented.

[0155] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing or description thereof.

Claims

1. A monitoring and early warning method for wildlife nature reserves, characterized in that, The method comprises: Collecting multimodal ecological monitoring data streams within the target wildlife nature reserve, wherein the multimodal ecological monitoring data streams include animal activity trajectory data, habitat environment parameter sequences, and bio-voiceprint signal sets; Dynamically matching the animal activity trajectory data with a baseline threat pattern in a predefined ecological threat indicator library to generate a first ecological association network feature, wherein the first ecological association network feature characterizes the spatiotemporal coupling relationship between the individual behavior of the target animal and the potential threat scenario; Extracting the migration path fragments in the animal activity trajectory data, combining the change gradient of the habitat environmental parameter sequence, and deriving the second ecological association network characteristics through the threat pattern evolution model, the second ecological association network characteristics reflect the synergistic effect strength of abnormal fluctuations in the migration path and sudden changes in environmental parameters; Performing cross-modal feature fusion on the first ecological association network feature and the second ecological association network feature to generate a multi-level threat coupling vector; Based on the frequency domain characteristic distribution of the biometric voiceprint signal set, a voiceprint anomaly detection map is constructed, and the voiceprint anomaly detection map is spatially superimposed and analyzed with the multi-level threat coupling vector. After a comprehensive threat situation matrix is generated, a dynamic threshold calibration is performed on the comprehensive threat situation matrix using an adaptive weight allocation algorithm to output a real-time threat level spectrum; According to the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library, the ecological protection intervention instruction set of the corresponding level is activated.

2. The monitoring and early warning method for wildlife nature reserves according to claim 1, wherein Before the step of dynamically matching the animal activity trajectory data with the baseline threat pattern in the predefined ecological threat indicator library, the method further includes: Performing noise filtering and trajectory interpolation processing on the animal activity trajectory data to generate a standardized activity trajectory sequence; Performing time alignment and outlier correction on the habitat environmental parameter sequence to generate a continuous environmental parameter surface; The step of dynamically matching the animal activity trajectory data with the baseline threat pattern in the predefined ecological threat indicator library to generate the first ecological association network feature includes: Performing spatiotemporal grid mapping on the standardized activity trajectory sequence and the continuous environmental parameter surface, and extracting the behavior density distribution and environmental parameter gradient within the grid unit; The behavior density distribution and the environment parameter gradient are compared with the spatiotemporal constraints of the baseline threat pattern through a threat pattern matching engine to generate a threat matching confidence of each grid unit; A three-dimensional threat heat distribution map is constructed based on the threat matching confidence of all grid cells as the first ecological association network feature.

3. The monitoring and early warning method for wildlife nature reserves according to claim 2, characterized in that, The step of extracting the migration path segments in the animal activity trajectory data comprises: Identifying path turning points and dwelling periods in the standardized activity trajectory sequence, and segmenting migration path sub-segments with continuous motion vectors; Perform curvature analysis and speed variation detection on each migration path sub-segment, and mark abnormal path morphological features; The step of generating the second ecological association network feature includes: Perform causal correlation analysis on the abnormal path morphological characteristics and the environmental parameter mutation events in the corresponding time window, and calculate the path-environment coupling coefficient; Construct a dynamic threat propagation chain based on the path-environment coupling coefficients of all migration path sub-segments as the second ecological association network feature; And, the step of performing cross-modal feature fusion on the first ecological association network feature and the second ecological association network feature to generate a multi-level threat coupling vector includes: Perform principal component dimensionality reduction processing on the three-dimensional threat heat distribution map to extract the core threat space vector; Convert the dynamic threat propagation chain into a spatio-temporal propagation probability matrix and perform a tensor product operation with the core threat space vector to generate the multi-level threat coupling vector.

4. The monitoring and early warning method for wildlife nature reserves according to claim 1, wherein, The step of constructing a voiceprint anomaly detection map based on the frequency domain feature distribution of the bio-voiceprint signal set includes: Perform frame windowing processing on the bio-voiceprint signal set and extract the Mel frequency cepstral coefficients and harmonic energy ratios of each frame of the signal; Perform joint discrimination on the Mel frequency cepstral coefficients and harmonic energy ratios through an abnormal voiceprint classifier and output a voiceprint anomaly probability distribution map; Based on a sound source localization algorithm, perform spatial clustering on abnormal voiceprint events to generate a voiceprint anomaly detection map with azimuth markings; And, the step of performing spatial overlay analysis on the voiceprint anomaly detection map and the multi-level threat coupling vector to generate a comprehensive threat situation matrix includes: Register the azimuth markings of the voiceprint anomaly detection map with the spatial coordinates of the multi-level threat coupling vector; Calculate the overlapping area ratio and spatial correlation index between the voiceprint anomaly area and the threat coupling area to generate the comprehensive threat situation matrix.

5. The monitoring and early warning method for wildlife nature reserves according to claim 1, wherein The step of using an adaptive weight allocation algorithm to perform dynamic threshold calibration on the comprehensive threat situation matrix and output a real-time threat level spectrum includes: According to the threat level distribution law in the historical threat event database, establish a baseline threshold curve for each threat dimension; Adopt a sliding time window algorithm to calculate the cumulative intensity values of each threat dimension in the current comprehensive threat situation matrix in real time; Map the cumulative intensity values to the baseline threshold curve through a non-linear interpolation method and output a dynamically adjusted real-time threat level spectrum; And, the step of activating the corresponding level of ecological protection intervention instruction set according to the mapping relationship between the real-time threat level spectrum and the preset emergency response strategy library includes: Analyze the weight distribution of each threat dimension in the real-time threat level spectrum and match the multi-level response protocols in the preset emergency response strategy library; Generate a drone patrol path plan, an ecological corridor blockade instruction, and an artificial inspection priority queue according to the spatial coverage range and duration of the threat level; Synchronize the instruction set to the protected area management terminal and mobile law enforcement equipment through a low-latency communication network.

6. The monitoring and early warning method for wildlife nature reserves according to any one of claims 1-5, characterized in that, The method further includes: Establish a digital twin model of the ecosystem of the target wildlife nature reserve, and receive and map the multi-modal ecological monitoring data stream in real time; Embed a threat evolution simulator in the digital twin model and predict the threat diffusion path within a future time window based on the current comprehensive threat situation matrix; Optimize the execution timing and resource allocation plan of the ecological protection intervention instruction set according to the prediction results; Among them, the update step of the digital twin model includes: Collect the ecological status feedback data after the implementation of the actual intervention measures, and calculate the threat suppression efficiency coefficient; Input the threat suppression efficiency coefficient into the threat evolution simulator for backpropagation training to update the prediction parameters of the threat diffusion path; When the threat suppression efficiency coefficient is lower than the preset threshold, trigger the secondary response mechanism and regenerate the cross-regional collaborative intervention instruction set.

7. The monitoring and early warning method for wildlife nature reserves according to claim 1, characterized in that, The method further includes: Construct a human activity monitoring network around the target wildlife nature reserve, and collect traffic flow data, night light intensity distribution, and illegal intrusion alarm signals; Cross-validate the human activity monitoring data with the comprehensive threat situation matrix to identify compound ecological threats induced by human factors; Automatically generate law enforcement evidence collection clue packages and public warning information push strategies according to the compound threat characteristics; Among them, the step of cross-validating the human activity monitoring data with the comprehensive threat situation matrix to identify compound ecological threats induced by human factors includes: Analyze the spatio-temporal coupling degree between the spatio-temporal trajectories of human activities and animal abnormal behavior events, and calculate the human interference contribution factor; Combine the persistent change trend of habitat environment parameters to evaluate the long-term cumulative effect of human activities on ecological threats; When the human interference contribution factor exceeds the critical value, activate the directional tracking and image acquisition functions of the high-precision video monitoring device.

8. The monitoring and early warning method for wildlife nature reserves according to claim 1, characterized in that The method further includes: Deploy edge computing devices at key ecological nodes in the target wildlife nature reserve to perform real-time feature extraction on the multi-modal ecological monitoring data streams collected locally; Encrypt and aggregate the feature extraction results of each node through the federated learning framework to update the global threat detection model parameters; Dynamically distribute the updated model parameters to each edge computing device to achieve the co-evolution of distributed threat perception capabilities; Among them, the update steps of the federated learning framework include: Design a differential privacy protection strategy to add geographical mask noise to sensitive animal location information; Adopt the model parameter differential aggregation technology to eliminate the influence of the data distribution skew of edge nodes on the global model; When a node device is detected to be abnormally offline, start the model parameter rollback mechanism and reallocate the computing tasks to adjacent nodes.

9. The monitoring and early warning method for wildlife nature reserves according to claim 1, wherein, The method further includes: Integrate the historical ecological restoration records and species reproduction data of the target wildlife nature reserve to construct an ecological system resilience assessment matrix; Perform multi-dimensional correlation analysis on the real-time threat level spectrum and the ecological system resilience assessment matrix to predict the ecological restoration potential under different intervention strategies; Dynamically adjust the implementation intensity and action cycle of the ecological protection intervention instruction set according to the prediction results; Among them, the step of performing multi-dimensional correlation analysis on the real-time threat level spectrum and the ecological system resilience assessment matrix to predict the ecological restoration potential under different intervention strategies includes: Establish a threat intensity-restoration efficiency response surface model to quantify the restoration rate of different intervention measures on the key indicators of the ecological system; Simulate the ecological resilience transition path under the superposition of continuous multi-round intervention measures to identify the optimal intervention strategy combination; When it is predicted that the repair rate is lower than the degradation rate, a request for expert consultation is triggered and a control plan for protected area zoning is initiated.

10. A monitoring and early warning system for wildlife nature reserves, characterized in that, The monitoring and early warning system for wildlife nature reserves includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the monitoring and early warning method for wildlife nature reserves according to any one of claims 1-9 above.

Images