An evaluation method for the Giant Panda National Park to resist the invasion of alien species

Abstract

The present application discloses an evaluation method and device for the Giant Panda National Park to resist the invasion of alien species, relating to the technical field of ecological protection, aiming to solve the technical problem that the prior art lacks an evaluation method for the method of the Giant Panda National Park to resist the invasion of alien species. The method includes: after resisting the target alien species, obtaining and based on the real-time video data of the Giant Panda National Park, obtaining the attention weight vector of the target alien species; constructing context features based on the analysis of the influencing factors of the historical growth of the target alien species; constructing the growth state characteristics of the target alien species group, obtaining the historical alien species invasion information and the corresponding measure decisions; obtaining the historical information experience feature vector based on the historical alien species invasion information and the measure decisions; and evaluating the method of the Giant Panda National Park to resist the invasion of alien species based on the growth state characteristics of the target alien species group and the historical information experience feature vector.

Description

Technical Field

[0001] This application relates to the field of ecological protection technologies, and particularly to an evaluation method and device for a giant panda national park to resist alien species invasion. Background Art

[0002] In 2021, China officially established the Giant Panda National Park, realizing the transformation of giant panda protection from decentralized protection in multiple nature reserves to overall protection after the establishment of the national park. The Giant Panda National Park is located in the alpine canyon area where the Sichuan Basin transitions to the Qinghai-Tibet Plateau, with complex topography and rich wildlife resources, numerous rare and endemic species. There are 316 species of national key protected wild animals and plants living in the area, known as the "natural gene pool". The Giant Panda National Park spans three provinces of Sichuan, Shaanxi, and Gansu, with a total area of 22,000 square kilometers, of which 87.9% of the area is located in Sichuan Province. While protecting giant pandas, the Giant Panda National Park has an umbrella protection effect on other wild animals and plants with sympatric distribution, which is of great significance to regional biodiversity and ecosystem protection.

[0003] Biological invasion is the second largest cause of biodiversity loss, second only to habitat destruction, and has seriously threatened global ecological security and human health. China is one of the countries with the richest biodiversity in the world and also one of the countries most threatened by biological invasion. Alien target alien species such as water hyacinth (Eichhornia crassipes) and crofton weed (Eupatorium adenophorum) have swept across the southwestern region, causing huge losses to agriculture, forestry, animal husbandry in the invaded areas.

[0004] As the first national park established in the natural protected area system of Sichuan Province, the Giant Panda National Park is the cornerstone of biodiversity conservation in the giant panda habitat. Evaluating the effectiveness of the Giant Panda National Park in resisting alien species invasion, revealing the spatial distribution pattern of alien species inside and outside the national park, and clarifying the diffusion risk of alien species in the national park and its surrounding buffer zones are crucial for proposing biosecurity strategies in nature reserves with the national park as the main body. Summary of the Invention

[0005] This application provides an evaluation method, device, equipment, and medium for a giant panda national park to resist alien species invasion, aiming to solve the technical problem that the existing technology lacks an evaluation method for the method of a giant panda national park to resist alien species.

[0006] To solve the above technical problem, an embodiment of this application provides: an evaluation method for a giant panda national park to resist alien species invasion, including the following steps:

[0007] After adopting target resistance measures to resist target alien species, obtain real-time video data of the Giant Panda National Park; based on the real-time video data of the Giant Panda National Park, extract the real-time growth characteristics of the target alien species; based on the real-time growth characteristics of the target alien species, obtain the attention weight vector of the target alien species;

[0008] Based on the analysis of the influencing factors of the historical growth of the target alien species, construct context features; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data, and target resistance measures;

[0009] Based on the attention weight vector of the target alien species and the context features, construct a multi-modal growth feature vector of the target alien species; based on multiple multi-modal growth feature vectors of the target alien species, construct the group growth state characteristics of the target alien species;

[0010] Based on the semantic extraction model, obtain historical alien species invasion information and the corresponding decision-making on measures in the Giant Panda National Park; based on the historical alien species invasion information and the decision-making on measures in the Giant Panda National Park, obtain the historical information experience feature vector;

[0011] Based on the group growth state characteristics of the target alien species and the historical information experience feature vector, predict the growth characteristics of the target alien species at time t+1; based on the growth characteristics of the target alien species at time t+1, evaluate the method for the Giant Panda National Park to resist alien species invasion.

[0012] As some optional embodiments of the present application, the step of obtaining the attention weight vector of the target alien species based on the real-time growth characteristics of the target alien species includes:

[0013] Obtain the historical growth feature sequence of the target alien species;

[0014] Based on the historical growth feature sequence of the target alien species and the real-time growth characteristics of the target alien species, obtain the attention weight vector of the target alien species.

[0015] As some optional embodiments of the present application, the step of obtaining the attention weight vector of the target alien species based on the historical growth feature sequence of the target alien species and the real-time growth characteristics of the target alien species includes:

[0016] Extract the attention weights of different resistance measures and the current growth feature vector of the target alien species from the historical growth feature sequence of the target alien species;

[0017] Based on the current growth feature vector and attention weights of the target alien species, obtain the attention weight vector of the target alien species.

[0018] In some alternative embodiments of the present application, the step of constructing a multi-modal growth feature vector of the target alien species based on the attention weight vector of the target alien species and the context features includes:

[0019] Converting the attention weight vector of the target alien species and the context features into binary codes by using one-hot encoding and multi-hot encoding to obtain the growth feature vector coding information of the target alien species;

[0020] Inputting the growth feature vector coding information of the target alien species and the associated historical growth feature vectors into a trained deep interest network to construct a multi-modal growth feature vector of the target alien species.

[0021] In some alternative embodiments of the present application, the step of obtaining historical alien species invasion information and corresponding decision-making on measures in the Giant Panda National Park based on the semantic extraction model; and obtaining a historical information experience feature vector based on the historical alien species invasion information and the decision-making on measures in the Giant Panda National Park includes:

[0022] Obtaining a BERT semantic extraction model; wherein, the BERT semantic extraction model is trained based on historical data, and the historical data includes biological invasion time, biological invasion range, biological invasion degree change information, and types of measures taken in the Giant Panda National Park;

[0023] Based on the BERT semantic extraction model, obtaining the key information of historical biological invasions and a set of guiding measures in the Giant Panda National Park;

[0024] Encoding and mapping the key information of historical biological invasions and the set of guiding measures in the Giant Panda National Park into a high-dimensional information matrix, and extracting a preliminary biological invasion state description vector by embedding;

[0025] Inputting the preliminary biological invasion state description vector into a trained EventKG model to model the temporal relationship between the measures taken and the changes in the biological invasion state, and then representing the relationship between the measures taken and the changes in the growth state of the alien species population as a historical information experience vector in a continuous vector space.

[0026] In some alternative embodiments of the present application, the step of predicting the growth characteristics of the target alien species at the t+1 moment based on the growth state characteristics of the target alien species population and the historical information experience feature vector includes:

[0027] Input the growth state characteristics of the target alien species population into the trained stacked transformers network model to extract the historical experience information extraction network, the context information Ct extraction network, and the real-time state feature extraction network;

[0028] Perform decoding processing on the historical experience information extraction network, the context information Ct extraction network, and the real-time state feature extraction network respectively to obtain decoding results; input the decoding results into the prediction module respectively to predict the growth characteristics of the alien species population at time t + 1.

[0029] As some alternative embodiments of the present application, the step of inputting the decoding results into the prediction module respectively to predict the growth characteristics of the alien species population at time t + 1 includes:

[0030] Based on the decoding results, obtain multiple candidate state allocation regions, and obtain multiple regions; where each region represents a growth state of the alien species.

[0031] Determine the region where the growth state characteristics of the target alien species population are located, calculate the growth states and confidence levels generated under different optional measures, define an overall loss function based on the historical information experience characteristics; and train the parameters in the loss function to obtain a prediction network model.

[0032] Input the current growth state characteristics of the target alien species population into the prediction network model to obtain the growth state types of the alien species population at time t + 1 under different measures; where the prediction network model uses stacked transformers as the prediction network architecture.

[0033] On the other hand, the embodiments of the present application also provide: An evaluation device for a national park for giant pandas to resist alien species invasion, including:

[0034] A first acquisition module, configured to acquire real-time video data of the national park for giant pandas after using a target resistance measure to resist a target alien species; extract real-time growth characteristics of the target alien species based on the real-time video data of the national park for giant pandas; and obtain an attention weight vector of the target alien species based on the real-time growth characteristics of the target alien species.

[0035] A second acquisition module, configured to construct context features based on an analysis of the influencing factors of the historical growth of the target alien species; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data, and target resistance measures.

[0036] Construct a target alien species population growth state module, which is used to construct a target alien species multimodal growth feature vector based on the attention weight vector of the target alien species and the context features; and construct a target alien species population growth state feature based on multiple target alien species multimodal growth feature vectors.

[0037] A semantic extraction module, which is used to obtain historical alien species invasion information and corresponding Giant Panda National Park measure decisions based on a semantic extraction model; and obtain a historical information experience feature vector based on the historical alien species invasion information and the Giant Panda National Park measure decisions.

[0038] An evaluation module, which is used to predict the growth features of the target alien species at time t+1 based on the target alien species population growth state feature and the historical information experience feature vector; and evaluate the method for the Giant Panda National Park to resist alien species invasion based on the growth features of the target alien species at time t+1.

[0039] On the other hand, an embodiment of the present application also provides: a computer device, which includes a memory and a processor. A computer program is stored in the memory, and the processor executes the computer program to implement the above-mentioned evaluation method for the Giant Panda National Park to resist alien species invasion.

[0040] On the other hand, an embodiment of the present application also provides: a computer-readable storage medium, on which a computer program is stored, and the processor executes the computer program to implement the above-mentioned evaluation method for the Giant Panda National Park to resist alien species invasion.

[0041] Compared with the prior art, the technical solution of the present application solves the deficiencies in the prior art in the evaluation method for the Giant Panda National Park to resist alien species invasion by introducing multiple data sources and analysis methods, which are specifically reflected in the following aspects:

[0042] 1. Fusion of multimodal data

[0043] Current alien species invasion evaluation methods usually rely on a single data source, such as ecological surveys or expert evaluations, and lack real-time and dynamic information updates. This technical solution realizes the all-round monitoring of the state of alien species by collecting real-time video data and combining various information on the growth characteristics of target alien species and the ecological environment (such as climate conditions and biodiversity data). The integration of this multimodal data improves the accuracy and timeliness of the evaluation results.

[0044] 2. Application of attention weight vector

[0045] By extracting the attention weight vector of the target alien species, this solution can identify the key factors affecting the growth of alien species. Compared with traditional methods that only rely on statistical models for data analysis, it is more flexible and dynamic, and can reflect the adaptability and changes of alien species under specific environmental conditions in real time. This method enhances the pertinence and scientific nature of the evaluation.

[0046] 3. Effective utilization of historical information

[0047] This solution obtains historical alien species invasion information and corresponding management measure decisions through a semantic extraction model, and combines historical data with the current evaluation. The construction of the historical information experience feature vector can provide an important reference basis for the current evaluation, thereby improving the referability and effectiveness of the evaluation. By analyzing historical data, the solution can better identify successful management measures and existing problems, providing a scientific basis for future decisions.

[0048] 4. Enhancement of prediction ability

[0049] By constructing the growth state characteristics of the target alien species group and the historical information experience feature vector, the technical solution can predict the growth characteristics of alien species at time t + 1. This prediction ability enables the evaluation not only to stay at the analysis of the current situation, but also to look ahead to future change trends. On the one hand, it provides an early warning mechanism for managers, and on the other hand, it provides a scientific basis for policy formulation.

[0050] 5. Objectivity and evaluation standardization

[0051] By constructing a multi-modal growth feature vector and combining context features, this solution makes the evaluation method more objective and reduces the interference of human factors on the evaluation results. At the same time, the standardized method in the evaluation process enables data at different times and different locations to be compared horizontally, providing a quantitative reference standard for long-term ecological management.

[0052] In summary, the technical solution of this application greatly improves the evaluation ability of the method for the Giant Panda National Park to resist alien species invasion by integrating advanced technologies such as real-time monitoring, multi-modal data analysis, historical data utilization, and prediction modeling. It is beneficial to improving the effectiveness of evaluating the resistance of the Giant Panda National Park to biological invasion; predicting the suitable habitat distribution of invasive species and demarcating key prevention and control areas; providing a basis for formulating the most practical management measures for preventing and controlling alien invasive species in the Giant Panda National Park. This not only solves the problems of lack of systematicness and objectivity in the existing technology, but also provides a scientific basis and practical reference for the formulation and adjustment of relevant management measures. Brief description of the drawings

[0053] To more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts do not necessarily draw to actual scale.

[0054] Figure 1 Distribution map of the survey routes for alien invasions in the forest, grassland, and wetland ecosystems of Sichuan Province involved in the present application;

[0055] Figure 2 Relationship diagram between the key alien invasive species to be focused on for prevention and control and the Giant Panda National Park involved in the present application;

[0056] Figure 3 Heat map of the spatial distribution of forestry alien invasive species in 21 cities and prefectures of Sichuan Province involved in the present application;

[0057] Figure 4 Schematic flow diagram of the evaluation method for the Giant Panda National Park to resist alien species invasions involved in the present application;

[0058] Figure 5 Schematic structural diagram of the local attention module involved in the present application;

[0059] Figure 6 Schematic diagram of the multi-modal feature fusion and representation of target alien species under alien species involved in the present application;

[0060] Figure 7 Logic diagram for obtaining and describing the historical information experience feature vector H involved in the present application;

[0061] Figure 8 Schematic structural diagram of the alien species growth state prediction network involved in the present application;

[0062] Figure 9 Schematic diagram of the training strategy based on region division and decision-making involved in the present application;

[0063] Figure 10 Schematic structural diagram of the evaluation device for the Giant Panda National Park to resist alien species invasions involved in the present application.

[0064] The realization of the purpose of the present application, functional features, and advantages will be further described with reference to the embodiments and the drawings. Specific Embodiments

[0065] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0066] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative positional relationship and movement conditions between components in a specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0067] In the present application, unless otherwise clearly defined and limited, terms such as "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components or the interaction relationship between two components, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0068] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the meaning of "and / or" appearing throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or a solution that satisfies both A and B at the same time. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present application.

[0069] The computer device of the hardware operating environment involved in the solution of the present application may include: a processor, such as a central processing unit (CPU), a communication bus, a user interface, a network interface, and a memory. Among them, the communication bus is used to realize the connection and communication between these components. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory may be a high-speed random access memory (Random Access Memory, RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. The memory may also be a storage device independent of the aforementioned processor. The memory as a storage medium may include an operating system, a network communication module, a user interface module, and an electronic program, and may also include a data storage module. The network interface is mainly used for data communication with the network server; the user interface is mainly used for data interaction with the user; the processor and memory in the computer device of this embodiment can be set in the computer device, and the computer device calls the evaluation device for the Giant Panda National Park to resist the invasion of alien species stored in the memory through the processor, and executes the evaluation method for the Giant Panda National Park to resist the invasion of alien species provided in this embodiment.

[0070] It should be noted that the above-mentioned computer device may be an external hardware device that can run independently, or it may be a hardware device that comes with the internal part of the evaluation device for resisting the invasion of alien species in the above-mentioned Giant Panda National Park.

[0071] It should be noted that the data source of the technical solution described in this application is the census of alien invasive species in forest, grassland and wetland ecosystems in Sichuan Province carried out from 2021 to 2023. Figure 1 The following is a distribution map of alien invasion census routes in forest, grassland and wetland ecosystems in Sichuan Province drawn based on survey data, as well as Figure 2 The relationship between the key invasive alien species and the Giant Panda National Park shown in the figure, and Figure 3The spatial distribution heat map of forestry alien invasive species in 21 cities and prefectures of Sichuan Province as shown. The investigation scope comprehensively covers all 183 counties (cities, districts) within the administrative region of Sichuan Province, and protected areas at all levels with clear boundaries are used as independent investigation units. According to the types and distribution points of invasive species, the spatial distribution pattern of key invasive species in and around the Giant Panda National Park is clarified, the number, occurrence area, growth characteristics, etc. of invasive species in the Giant Panda National Park are calculated, and the effectiveness of the national park in resisting alien species invasion is evaluated.

[0072] Referring to Figure 4 , based on the foregoing hardware environment, this embodiment further provides an evaluation method for the Giant Panda National Park to resist alien species invasion, including the following steps:

[0073] Step S10: After using the target resistance measures to resist the target alien species, obtain the real-time video data of the Giant Panda National Park; based on the real-time video data of the Giant Panda National Park, extract the real-time growth characteristics of the target alien species; based on the real-time growth characteristics of the target alien species, obtain the attention weight vector of the target alien species.

[0074] It should be noted that the real-time video data of the Giant Panda National Park mainly comes from the network cameras in the Giant Panda National Park. These cameras are installed in key areas of the Giant Panda National Park, such as the entrance area and transit channels, which are the most likely paths for alien species to enter. By monitoring these key entrances, the initial appearance of alien species can be captured in a timely manner, and measures can be taken promptly to prevent their further spread. Or water source areas. Water sources are hotspots of biodiversity, attracting various organisms to come and drink water and inhabit. Monitoring these areas helps to observe the interactions between alien species and native species and evaluate whether alien species pose a threat to the ecological balance of water source areas. Or important ecological areas, such as forests, grasslands, etc. with special ecological functions. These areas are often important support points for biodiversity. Installing cameras in these areas can help researchers monitor the potential impacts of alien species on the local ecosystem. Or human activity-intensive areas. Alien species may be introduced through human activities (such as tourism, transportation, etc.). Monitoring these areas can help identify the role of human factors in the spread of alien species. Or known alien species activity areas. For areas where alien species have already appeared, it is necessary to continuously monitor their dynamics in order to evaluate the effectiveness of the measures taken and adjust the strategies in a timely manner.

[0075] By installing network cameras in these key areas, the monitoring coverage can be maximized, direct evidence of alien species invasion can be obtained in a timely manner, and protection measures can be effectively evaluated and adjusted. Such an installation strategy can provide a comprehensive and real-time ecological monitoring system for the Giant Panda National Park, which helps to maintain its biodiversity and ecological balance.

[0076] For example, in the San Bernardino Mountains of California, a camera system was installed to monitor the invasion of non-native species. It successfully recorded the activities of feral pigs and domestic cats, and this data is crucial for taking timely management measures. Another example is in the Serengeti National Park in Africa. By installing cameras around waterholes and rivers, researchers were able to observe the impact of non-native species invasion on the activities of native crocodiles and hippos. Also, in Tasmania, Australia, researchers deployed cameras in the native forest area and recorded the impact of non-native foxes on the populations of native wombats and other small mammals.

[0077] In practical situations, extracting the real-time growth characteristics of non-native species can help researchers and managers accurately understand the health status, reproductive situation, and population dynamics of these species. This information is crucial for assessing the specific impact of non-native species on native ecosystems. And by monitoring the growth and spread patterns of non-native species in real time, managers can timely detect potential ecological risks and hotspots, and thus implement early intervention measures, such as adjusting ecological isolation areas or initiating emergency population control plans. Also, detailed growth data supports scientific research, such as research on ecological adaptability, competition relationships, and predation pressure, which are all key scientific issues for understanding and dealing with non-native species invasion.

[0078] When using machine learning and deep learning models to analyze video data, the attention mechanism can help the model more accurately identify and track target non-native species, especially in complex natural environments. The attention weight vector can indicate which parts of the image the model should focus on, thus improving the recognition accuracy.

[0079] The attention model can be used to analyze the growth patterns of non-native species, such as their activity frequencies and habits at specific times and locations. This helps to study their ecological habits and environmental adaptability, providing data support for formulating management strategies. By knowing which areas and time periods non-native species are most active (based on the attention weight distribution), managers can more effectively allocate monitoring resources and control measures, such as increasing the patrol frequency during peak activity periods or optimizing the configuration of monitoring equipment.

[0080] Therefore, it can be seen that by using technical means to extract the real-time growth characteristics of non-native species and applying the attention weight vector, the monitoring and response capabilities of ecological reserves to non-native species invasion can be greatly enhanced. This not only helps to protect native biodiversity but also improves the scientific nature and effectiveness of management measures.

[0081] In an actual situation, the growth characteristics of the target alien species can, to a certain extent, quantify the current reproduction degree of the target alien species. Therefore, this application intends to achieve a unified characterization of the growth of the target alien species by integrating influencing factors based on the real-time growth characteristics of the target alien species. Specifically, as shown in step S20.

[0082] Specifically, based on the real-time growth data of the target alien species, by extracting the attention weight vectors of the growth in different current periods from its historical growth sequence, the current growth vector of the target alien species is calculated adaptively; effectively removing redundant information in growth and accurately reflecting the current reproduction state of the target alien species.

[0083] Specifically, the step of obtaining the attention weight vector of the target alien species based on the real-time growth characteristics of the target alien species includes: obtaining the historical growth characteristic sequence of the target alien species; obtaining the attention weight vector of the target alien species based on the historical growth characteristic sequence of the target alien species and the real-time growth characteristics of the target alien species.

[0084] Among them, the step of obtaining the attention weight vector of the target alien species based on the historical growth characteristic sequence of the target alien species and the real-time growth characteristics of the target alien species includes: extracting the attention weights of different control measures and the current growth characteristic vector of the target alien species from the historical growth characteristic sequence of the target alien species; obtaining the attention weight vector of the target alien species based on the current growth characteristic vector and the attention weights of the target alien species.

[0085] In an example, assume that the target alien species under study is the American mink, which has been found to pose a threat to the native bird population in a certain area. The research team hopes to use the latest technology to monitor the growth and spread of this mink in order to develop effective control measures. Then the above steps may include:

[0086] 1) Obtaining the historical growth characteristic sequence of the target alien species:

[0087] First, collect the growth data of American mink in the past few years, including body weight, body length, reproduction rate, food source, etc. Using this data to construct a time series model, which helps to understand how American mink adapt to the new environment and infer their possible growth trends. Based on the historical growth characteristic sequence, we can predict how the population size and distribution of American mink may change in a certain period in the future.

[0088] 2) Obtaining the attention weight vector of the target alien species:

[0089] By setting up wildlife monitoring cameras, collect the activity and growth data of American minks in real time. Combine the historical growth data and real-time observation data, and use a deep learning model to analyze the importance of the current data and generate an attention weight vector. Identify the locations and times when American minks are most active under specific environmental conditions, which are the key areas emphasized by the attention weight vector.

[0090] 3) Extract the attention weights of different control measures and the current growth characteristics:

[0091] Analyze the impact of different control measures implemented in the past (such as trapping, fencing, introducing ecological competitors, etc.) on the growth characteristics of American minks. Based on the effects of these measures, adjust the attention weights of the model to highlight the growth characteristics related to effective control measures. If it is found that under certain specific measures, the growth rate of American minks slows down or the population decreases, then give higher weights to these characteristics in the model. This helps to more accurately predict and implement the most effective control strategies in the future.

[0092] Through the above analysis and application, it can be seen that the method of combining historical and real-time growth characteristics and optimizing the management measures of alien species through the attention mechanism can not only improve the accuracy of monitoring and response, but also make ecological management more scientific and systematic.

[0093] Step S20: Based on the analysis of the influencing factors of the historical growth of the target alien species, construct context features; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data, and target control measures.

[0094] In the actual technical solution, the characteristics of the target alien species among alien species are of multiple types and structures. To achieve the fusion representation of the multi-modal characteristics of the target alien species. Therefore, first use the method of unified feature encoding to convert the multi-modal feature data into binary codes that can be uniformly processed; then construct a deep interest network based on the attention module to realize the association between the historical growth feature vector and the real-time growth feature vector of the target alien species.

[0095] Next, a detailed description of the step of converting multi-modal feature data into binary codes that can be uniformly processed by using the method of unified feature encoding is given with an example:

[0096] Suppose the target alien species under study is the invasive plant Eichhornia crassipes, which is rich in multimodal features. The growth characteristics of Eichhornia crassipes include visual data (such as images), biological data (such as DNA sequences), and environmental data (such as water quality conditions), all of which are multi-type and multi-structured data. Eichhornia crassipes is a fast-growing aquatic plant that can quickly cover the entire water surface, obstruct water flow, affect water quality, and pose a great threat to the local ecosystem. Effective management and control of its spread require a detailed analysis of its growth characteristics.

[0097] Step 211: Collection of feature data

[0098] Visual data: Use drones and underwater cameras to collect image and video data of Eichhornia crassipes, and record its growth density, distribution range, etc.

[0099] Biological data: Obtain the DNA sequences of Eichhornia crassipes through sample collection and analyze its genetic variation.

[0100] Environmental data: Measure environmental parameters such as water temperature, pH value, dissolved oxygen, etc. in the growth area.

[0101] Step 212: Feature preprocessing

[0102] Image preprocessing: Crop, scale, and standardize the images to make them suitable for subsequent calculations and analyses.

[0103] DNA data processing: Convert the DNA sequence data into numerical features, such as through a certain coding strategy (such as k-mer counting).

[0104] Environmental data standardization: Normalize the environmental parameters to eliminate the influence of dimensions.

[0105] Step 213: Multimodal data encoding

[0106] Feature vectorization: Convert all preprocessed feature data into vector form. For example, features of images are extracted through a convolutional neural network (CNN), DNA data are converted through specific bioinformatics tools, and environmental data are directly used as numerical inputs.

[0107] Binary encoding: Use an autoencoder or other suitable deep learning models to further compress and convert the feature vectors into binary encoding. The key to this step is to reduce the data dimension and unify the data formats from different sources for unified processing in the deep network.

[0108] Step 214: Comprehensive feature expression

[0109] Unified processing: The feature data encoded in binary will be input into a deep interest network based on an attention module, which can identify and strengthen the associations between key features, such as the interaction between historical growth patterns and real-time environmental conditions.

[0110] Through the above steps, the multi-modal data of water hyacinth can be effectively converted into a unified internal representation, and further analyzed and predicted using deep learning models. This method not only improves the efficiency of data processing, but also enhances the model's understanding and prediction ability of the growth dynamics of alien species.

[0111] In some other embodiments, the step of constructing the multi-modal growth feature vector of the target alien species based on the attention weight vector and the context features of the target alien species may further include:

[0112] Using one-hot encoding and multi-hot encoding methods to convert the attention weight vector and the context features of the target alien species into binary encoding to obtain the growth feature vector encoding information of the target alien species; inputting the growth feature vector encoding information of the target alien species and the associated historical growth feature vectors into a trained deep interest network to construct the multi-modal growth feature vector of the target alien species.

[0113] Taking Conyza canadensis (a common invasive plant) as an example to show how to construct its multi-modal growth feature vector through the attention weight vector and context features, the specific implementation steps are as follows:

[0114] Step 221: Collect feature data

[0115] Attention weight vector: This may include the weight evaluation of environmental factors that have the greatest impact on Conyza canadensis, such as sunlight, temperature, moisture, etc. These weights can be obtained through experimental data or on-site monitoring.

[0116] Context features: For example, soil pH value, surrounding vegetation type, land use situation, etc.

[0117] Step 222: Feature preprocessing and encoding

[0118] One-hot encoding: Suitable for categorical features, with each category in a separate position. For example, the land use situation (farmland, city, forest land) will be converted into a three-element vector, where only one element is 1 and the rest are 0.

[0119] Multi-hot encoding: Suitable for features that can belong to multiple categories. Suppose Conyza canadensis grows in multiple different climate conditions, such as the dry season and the rainy season, and these conditions will be encoded in the same vector.

[0120] Step 223: Convert to binary encoding

[0121] Construction of feature vectors: Combine the data encoded by one - hot and multi - hot to form complete growth feature vector encoding information. This information will reflect the adaptability and growth status of Conyza canadensis under different environmental conditions.

[0122] Step 224: Input into the deep interest network

[0123] Application of feature vectors: Input the encoded current feature vector and the historical growth feature vector into the deep interest network together. This network may be based on deep learning technology and is specifically trained to predict changes in plant growth patterns and potential distributions.

[0124] Deep learning model: The model will analyze the input feature vectors, identify the key influencing factors for the growth of Conyza canadensis, and predict its future growth trend and potential invasion risk.

[0125] Through the above steps, researchers or managers can obtain a comprehensive view of how Conyza canadensis interacts with its environment and accordingly formulate more effective management strategies to control or utilize this alien plant. This multimodal analysis method helps to more accurately predict and address plant invasion problems, especially in the current context of rapid environmental change.

[0126] According to the technical solutions described above, a table can be designed to display the feature encodings of the target alien species (such as Conyza canadensis). This table will contain multiple columns, respectively showing different features and their encoding methods. The following is an example table showing how to convert the attention weight vector and context features into one - hot and multi - hot encoding forms.

[0127]

[0128] Among them, the step of constructing a deep interest network based on the attention module to realize the association between the historical growth feature vector and the real - time growth feature vector of the target alien species includes:

[0129] Step S221, as Figure 5As shown, in this step, a local attention module is adopted (the local attention module refers to the following process, which calculates the growth representation of the current target alien species adaptively through the outer product of the historical growth feature vector of the target alien species and the current growth feature vector of the target alien species), that is: select any one growth feature vector in the current growth of the target alien species, perform outer product processing with the historical feature growth weight vector of the target alien species respectively, then connect the two vectors themselves, and finally input them into two MLP layers to obtain N real-valued weight values. The calculation process is shown in the following formula:

[0130]

[0131] Among them, (e1, e2,... e H ) is a list of embedding vectors of the growth of the target alien species with a length of H, and e j is the j-th growth of the target alien species; v u (A) is the weighted vector of the historical growth feature vector and the real-time growth feature vector of the target alien species finally obtained; v A is the historical growth feature vector of the target alien species; w j is the weight value output by the feed-forward network a(e j , v A ).

[0132] Step S222: The local attention layer does not perform normalization processing on the real-valued weight w j . Therefore, to a certain extent, Sum(w i ) can be equivalent to the attention weight vector of the target alien species, and this value can effectively quantify the current reproduction state of the target alien species. The higher the Sum value, the greater the probability that the target alien species reproduces too fast.

[0133] Step S30: Based on the attention weight vector of the target alien species and the context features, construct a multi-modal growth feature vector of the target alien species; based on multiple multi-modal growth feature vectors of the target alien species, construct the growth state features of the target alien species group.

[0134] The multi-modal feature fusion representation of the target alien species under the alien species described in this application is as Figure 6As shown, it can be seen that in this application, first, a real-time growth sequence of a target alien species is extracted based on the monitoring videos of the Giant Panda National Park, and the historical growth sequence of the target alien species is obtained by association; the real-time growth sequence and the historical growth sequence of the target alien species are feature-encoded and input into an embedding layer, so that the embedding layer outputs a historical growth feature vector and a real-time growth feature vector of the target alien species; then, the historical growth feature vector and the real-time growth feature vector of the target alien species are input into an attention module to obtain the current reproduction degree of the target alien species; at the same time, in order to improve the prediction accuracy, in this application, the context features are also feature-encoded and then input into the embedding layer and the connection layer; then, the context features and the attention weight vector of the target alien species are input into the connection layer at the same time to obtain the multi-modal growth features of the target alien species.

[0135] After obtaining the multi-modal growth features of each target alien species through the above steps, group growth state features of the target alien species are further constructed based on multiple ones, so as to facilitate the subsequent evaluation of the growth state of alien species, that is:

[0136] Step S40: Obtain historical alien species invasion information and corresponding decision-making on measures in the Giant Panda National Park based on a semantic extraction model; obtain a historical information experience feature vector based on the historical alien species invasion information and the decision-making on measures in the Giant Panda National Park.

[0137] In actual situations, there is an obvious correlation between the evolution laws among various states of alien species in the Giant Panda National Park and the measures taken by the Giant Panda National Park, and relevant experience information can be effectively obtained from historical biological invasions. Therefore, the steps of obtaining historical biological invasion characteristics and measures that can be taken by the Giant Panda National Park through a semantic extraction model, and obtaining group growth state characteristics of the target alien species based on the historical biological invasion characteristics and the decision-making on measures in the Giant Panda National Park include: obtaining a BERT semantic extraction model; wherein, the BERT semantic extraction model is trained based on historical data, and the historical data includes biological invasion time, biological invasion range, information on changes in the degree of biological invasion, and types of measures taken by the Giant Panda National Park; based on the BERT semantic extraction model, obtain the key information of historical biological invasions and the set of guiding measures in the Giant Panda National Park; encode and map the key information of historical biological invasions and the set of guiding measures in the Giant Panda National Park into a high-dimensional information matrix, and extract a preliminary description vector of the biological invasion state in an embedded manner; input the preliminary description vector of the biological invasion state into the trained EventKG model to model the temporal relationship between the measures taken and the changes in the biological invasion state, and then represent the relationship between the measures taken and the changes in the group growth state of alien species as a historical information experience vector in a continuous vector space.

[0138] That is, asFigure 7 As shown, obtain the semantic extraction model obtained by training the BERT semantic extraction model based on the historical data in the Giant Panda National Park database; thus, perform semantic extraction based on the BERT semantic extraction model, encode the information obtained from the semantic extraction, and input it into the embedding layer to output historical biological invasion features; and based on the historical biological invasion features, extract the event correlation relationship to classify the status of alien species and output the historical information experience vector.

[0139] Among them, the extraction of the event correlation relationship means that based on the historical biological invasion features, judge the relationship between the growth status of alien species and the resistance measures in the alien species growth status and resistance measure structure. For example, when alien species A grows to status i, if measure k is taken, how will the growth status change; if the growth status changes to status j, what resistance measures should be taken, such as taking measure I, etc. It should be noted that the Figure 7 is mainly for expressing the measure structure. The English part has no actual meaning and is only a naming reference for storing data such as growth status or resistance measures. Therefore, it will not be elaborated here.

[0140] Specifically, it is achieved through the following steps:

[0141] Step S41: Use the semantic extraction model based on the BERT model to obtain key information in historical information, such as the time of biological invasion, the location of occurrence, the types of measures taken by the Giant Panda National Park, the changes in the growth status of alien species, etc.;

[0142] Step S42: Extract the set E of resistance measures of alien species from the relevant materials for resisting alien biological invasions in the Giant Panda National Park to form the resistance measure set feature vector E=(e1,…,e n );

[0143] Step S43: Encode and map the obtained corresponding historical biological invasion multi-dimensional information to a high-dimensional information matrix, and extract the preliminary description vector of the growth status of alien species by embedding;

[0144] Step S44: Use the EventKG model to represent the relationship between the growth status and the taken resistance measures, and realize the temporal relationship modeling between the taken resistance measures and the changes in the growth status of alien species. Among them, eventKG-r represents an invasion event, and rdf:type indicates the type of this invasion event, such as event or relation, and the remaining attributes are other relevant descriptions such as the start and end times, locations, and effects corresponding to this invasion event;

[0145] Step S45: Classify each historical alien species through the EventKG clustering cluster according to the different degrees of the development status of alien species;

[0146] Step S46: Represent the relationship between the measures taken and the changes in the growth state of the alien species as a historical information experience vector in a continuous vector space. This vector contains the temporal and change relationships between the development of the growth state of the alien species and the measures taken by the Giant Panda National Park, as Figure 8 shown. That is:

[0147] To effectively utilize the experience information contained in the historical biological invasions of the Giant Panda National Park and the context information provided by the set of dynamic graph state features, a stacked transformers network as Figure 8 shown is adopted to achieve the feature extraction and fusion of multiple data sources. It consists of multiple transformers, namely the historical experience information extraction network, the context information Ct extraction network, and the real-time state feature extraction network. Each module unit uses the result from the previous transformer as the input to its decoder. The decoder part constructs a measure feature vector selector for the Giant Panda National Park. For different candidate measures, it aggregates multiple context information channels from the encoder, makes independent predictions. The measures are orthogonal to each other, and independent decoding is performed separately for different measures in parallel and input into the prediction module. Finally, the growth state X t+1 of the alien species under different measures taken by the Giant Panda National Park is predicted, and an evaluation is made on whether the measures taken to resist are effective. That is, Step S50:

[0148] Step S50: Based on the growth state characteristics of the target alien species group and the historical information experience feature vector, predict the growth characteristics of the target alien species at time t + 1; based on the growth characteristics of the target alien species at time t + 1, evaluate the method for the Giant Panda National Park to resist alien species invasion.

[0149] Specifically, the step of predicting the growth characteristics of the target alien species at time t + 1 based on the growth state characteristics of the target alien species group and the historical information experience feature vector includes: inputting the growth state characteristics of the target alien species group into the trained stacked transformers network model to extract the historical experience information extraction network, the context information Ct extraction network, and the real-time state feature extraction network; performing decoding processing on the historical experience information extraction network, the context information Ct extraction network, and the real-time state feature extraction network respectively to obtain decoding results; inputting the decoding results into the prediction module respectively to predict the growth characteristics of the alien species group at time t + 1.

[0150] Among them, the step of respectively inputting the decoding result into the prediction module to predict the growth characteristics of the alien species population at time t+1 includes: based on the decoding result, obtaining a plurality of candidate state allocation regions, obtaining a plurality of regions; wherein each region represents a state of the alien species; determining the region where the growth state characteristics of the target alien species population are located, calculating the growth state and confidence generated under different optional measures, defining an overall loss function based on historical information experience characteristics; and training the parameters in the loss function to obtain a prediction network model; inputting the current growth state characteristics of the target alien species population into the prediction network model to obtain the growth state types of the alien species population at time t+1 under different measures; wherein the prediction network model uses stacked transformers as the prediction network architecture.

[0151] That is to say, as Figure 9 shown, the present application predicts the predicted state and predicted probability of alien species in the Giant Panda National Park through the following steps:

[0152] (1) Allocate a region for each candidate state, and each region represents a growth state R of the alien species. In the training stage, by selecting the region (historical experience characteristics) where the Ground Truth information is located, calculate the trajectories and confidence generated under different optional measures, and calculate the loss function between the trajectory and the Groundtruth.

[0153] (2) Use the historical information experience feature vector H as training data to train the groundtruth. For the real-time growth state feature x t of the given historical dynamic graph, select a countermeasure e n from the set of candidate countermeasure feature vectors E=(e1,...,e i ), i=1~n. According to the corresponding countermeasure, select the corresponding real-time growth state feature of the dynamic graph from the historical information experience feature vector H as x gt . There is a loss function L i between the predicted state x gt (t+1) and x reg :

[0154]

[0155] Among them, x i (t+1) represents the predicted growth state of the alien species at time t+1 generated under the selection of the i-th measure.

[0156] (3) Define the relationship between the growth of invasive species in different growth states and the countermeasures taken in the Giant Panda National Park as a probability distribution. Therefore, the probability prediction of the growth state change of invasive species under different countermeasures in the Giant Panda National Park is equivalent to the difference evaluation between the probability distribution P (prediction) and the probability distribution G (ground truth).

[0157] (4) For the corresponding countermeasures E that can be taken in a certain area region =(e1,…,e k ), apply the MLP network m(·) to generate the corresponding confidence level, which represents the probability of taking measures to evolve the growth state of invasive species to this area:

[0158]

[0159] m(·) represents the MLP network, E region =(e1,…,e k ) represents the countermeasures that can be taken, k represents the number of countermeasures that can be taken, and R represents the state area.

[0160] (5) Suppose there is an optional set of countermeasures E in a certain area region =(e1,…,e k ), and its probability distribution is:

[0161]

[0162] m(·) represents the MLP network; E region =(e1,…,e k ) represents the countermeasures that can be taken; τ(·) represents the probability distribution.

[0163] (6) For the Giant Panda National Park under different countermeasures, the error of the probability prediction of the invasive species state can be defined as:

[0164]

[0165] Among them, D(x,x gt ) represents the distance between the predicted state and the actual state in the mapping space, D(x,x gt ) = ||x - x gt ||2; D(x i , x gt ) refers to the distance between the predicted growth state generated under the selection of the i-th measure and the actual growth state in the mapping space.

[0166] The probability distribution loss function L i between the predicted growth state x gt and x confAs shown below:

[0167]

[0168] Dkl represents the Kullback-Leibler divergence, which is a calculation method for measuring the difference between two probability distributions.

[0169] (7) In summary, during the training process, the overall training loss function based on historical information and empirical features can be defined as:

[0170]

[0171] where σ1 and σ2 are loss weights determined through learning and training; σ i represents (σ1, σ2).

[0172] (8) Based on the trained model, using stacked transformers as the prediction network architecture, output the growth state types and their probabilities of alien species in the next time period under different measures.

[0173] In some embodiments, evaluating the method for the Giant Panda National Park to resist alien species invasion based on the growth characteristics of the target alien species at time t + 1 means evaluating the effectiveness of the method for the Giant Panda National Park to resist alien species invasion based on the growth characteristics of the target alien species at time t + 1, and can also predict the strategies that need to be adjusted in the future. That is, according to the model output, evaluate the potential threats of alien species to giant pandas and their ecosystems, and propose suggestions for improving the resistance measures, such as adjusting physical barriers, chemical treatments, or biological control strategies. Evaluate the impact of the spread of alien species on local biodiversity, especially on the food chain and habitat of giant pandas. Analyze the environmental changes that alien species may cause, such as soil nutrient imbalance, water source pollution, etc. Quantify the effectiveness of current measures in slowing down the spread of alien species, determine which measures are the most effective and which need to be improved. Evaluate the cost-benefit of different measures and optimize resource allocation.

[0174] Next, taking water hyacinth as an example of an alien species, illustrate in detail how to use the method described in this application to evaluate its invasion in the Giant Panda National Park and evaluate the effectiveness of current resistance strategies:

[0175] 1. Data collection and real-time monitoring:

[0176] Use surveillance cameras installed in the national park to capture real-time images of the water surface, and conduct regular aerial photography by drones to obtain the coverage area and growth rate of water hyacinths.

[0177] Collect data related to water quality and climate change, as these factors may affect the growth rate and distribution of water hyacinths.

[0178] 2. Feature extraction and analysis:

[0179] Extract growth characteristics of water hyacinth, such as coverage area, density, growth rate, etc., from real-time images.

[0180] Construct context features by combining water quality parameters (such as pH value, dissolved oxygen, etc.) and climate data (such as temperature, precipitation).

[0181] 3. Model establishment and prediction:

[0182] Train a prediction model to predict the distribution and density of water hyacinth at the future time point t+1 based on current and historical growth data.

[0183] Use historical invasion information to evaluate the effectiveness of past measures, such as the efficiency of mechanical removal or chemical treatment.

[0184] 4. Real-time simulation and prediction evaluation:

[0185] According to the data predicted by the model, evaluate the specific impacts that water hyacinth may have on the water ecosystem if no further measures are taken, especially on the photosynthesis and oxygen supply of organisms in the water area.

[0186] Evaluate the potential impacts of the predicted spread of water hyacinth on fish and other aquatic organisms in the Giant Panda National Park, which are part of the giant panda's food chain.

[0187] 5. Adjust and optimize control strategies:

[0188] Based on the model output, if the prediction results show that the current measures are insufficient to control the rapid spread of water hyacinth, propose to increase the frequency of mechanical cleaning or use biological control methods, such as introducing natural enemies.

[0189] Plan to implement new management strategies, such as using specific chemical treatment methods in specific seasons or areas where water hyacinth grows rapidly.

[0190] That is, the model can help predict the impacts of the increase in water hyacinth coverage on water oxygen exchange and light, and evaluate the long-term impacts on aquatic plants and animals. And by comparing the model prediction results and actual observation data at different time points, the effectiveness of current control measures can be quantified, determining which methods are the most effective and whether the strategy needs to be adjusted. Combining the prediction model and real-time monitoring data, managers can implement an early warning system to respond in a timely manner to the abnormal spread of water hyacinth, effectively protecting the water quality and ecological security of the national park.

[0191] It can be seen that such systematic evaluation and prediction methods can help managers more accurately understand and address the challenges of alien species invasion, protecting the ecological environment of the Giant Panda National Park from the negative impacts of alien species.

[0192] That is, through video analysis, this application defines and predicts the growth status of alien species at each stage by combining the multimodal data of the Giant Panda National Park and historical biological invasion experience. Specifically, it involves taking the division of growth status regions and decision-making as the core, and based on multimodal data and historical biological invasion experience characteristics, finally realizing the prediction of the status and probability of alien species in the Giant Panda National Park, and evaluating the effectiveness of the adopted resistance measures.

[0193] That is, this application proposes a method for defining the status of alien species and future prediction that is based on multimodal data, relies on historical information and experience characteristics, and adapts to the scenario of the Giant Panda National Park. Specifically, it includes: for the multimodal data of the Giant Panda National Park, uniformly expressing the growth characteristics of target alien species, and studying the characterization of the individual growth characteristics of multimodal target alien species based on a deep interest network; for historical alien species in the Giant Panda National Park, extracting historical information and experience characteristics, and studying the definition of alien species status; based on multimodal data characteristics and historical information and experience characteristics, studying the prediction of the development status and probability of alien species.

[0194] On the other hand, as Figure 10 shown, to solve the above technical problems, the embodiments of this application also provide: an evaluation device for resisting alien species invasion in the Giant Panda National Park, including:

[0195] A first acquisition module, configured to, after using a target resistance measure to resist a target alien species, acquire real-time video data of the Giant Panda National Park; based on the real-time video data of the Giant Panda National Park, extract the real-time growth characteristics of the target alien species; based on the real-time growth characteristics of the target alien species, obtain the attention weight vector of the target alien species;

[0196] A second acquisition module, configured to construct context features based on an analysis of the influencing factors of the historical growth of the target alien species; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data, and target resistance measures;

[0197] A module for constructing the group growth status of the target alien species, configured to construct a multimodal growth feature vector of the target alien species based on the attention weight vector of the target alien species and the context features; based on multiple multimodal growth feature vectors of the target alien species, construct the group growth status characteristics of the target alien species;

[0198] A semantic extraction module, configured to obtain historical alien species invasion information and corresponding decision-making on measures in the Giant Panda National Park based on a semantic extraction model; based on the historical alien species invasion information and the decision-making on measures in the Giant Panda National Park, obtain a historical information and experience feature vector;

[0199] An evaluation module, configured to predict the growth characteristics of the target invasive alien species at time t+1 based on the growth state characteristics of the target invasive alien species group and the historical information empirical feature vector; and evaluate the method for the Giant Panda National Park to resist invasive alien species based on the growth characteristics of the target invasive alien species at time t+1.

[0200] Based on the same inventive concept as the foregoing embodiments, this embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the processor executes the computer program, the above method is implemented.

[0201] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or may be various devices including one or any combination of the foregoing memories. The computer may be various computing devices including intelligent terminals and servers.

[0202] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A method for evaluating the Giant Panda National Park's resistance to alien species invasion, characterized in that: The following steps are involved: After adopting the targeted defense measures to defend against the targeted alien species, obtaining real-time video data of the Giant Panda National Park; extracting the real-time growth characteristics of the targeted alien species based on the real-time video data of the Giant Panda National Park; Based on the real-time growth characteristics of the target alien species, obtaining an attention weight vector of the target alien species; Based on the analysis of factors influencing the historical growth of the target alien species, context features are constructed; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data and target resistance measures; The attention weight vector of the target alien species and the context feature are converted into binary code by one-hot encoding and multi-hot encoding to obtain the growth feature vector encoding information of the target alien species; the growth feature vector encoding information of the target alien species and the associated historical growth feature vector are input into the trained deep interest network to construct a multimodal growth feature vector of the target alien species; based on the multimodal growth feature vectors of multiple target alien species, the growth state characteristics of the target alien species population are constructed; Obtain a BERT semantic extraction model; wherein the BERT semantic extraction model is obtained based on historical data training, and the historical data includes the time of biological invasion, the scope of biological invasion, the change information of the degree of biological invasion, and the type of measures taken by the Giant Panda National Park; based on the BERT semantic extraction model, obtain key information of historical biological invasions and a set of guiding measures for the Giant Panda National Park; encode and map the key information of historical biological invasions and the set of guiding measures for the Giant Panda National Park to a high-dimensional information matrix, and extract a preliminary biological invasion status description vector by embedding; input the preliminary biological invasion status description vector into the trained EventKG model to achieve temporal relationship modeling between measures taken and changes in the biological invasion status, and then express the relationship between measures taken and changes in the growth status of alien species populations as a historical information experience vector in a continuous vector space; The growth status characteristics of the target alien species group are input into the trained stacked transformers network model to extract the historical experience information extraction network, the context information Ct extraction network and the real-time status feature extraction network; the historical experience information extraction network, the context information Ct extraction network and the real-time status feature extraction network are respectively decoded to obtain decoding results; the decoding results are respectively input into the prediction module to predict the growth characteristics of the alien species group at time t+1; based on the growth characteristics of the target alien species at time t+1, the method of resisting the invasion of alien species in the Giant Panda National Park is evaluated.

2. The method for evaluating the Giant Panda National Park's resistance to alien species invasion according to claim 1, characterized in that: The step of obtaining the attention weight vector of the target alien species based on the real-time growth characteristics of the target alien species comprises: Obtain historical growth characteristic sequences of target alien species; Based on the historical growth feature sequence of the target alien species and the real-time growth feature of the target alien species, an attention weight vector of the target alien species is obtained.

3. The method for evaluating the Giant Panda National Park's resistance to alien species invasion according to claim 2, characterized in that: The step of obtaining the attention weight vector of the target alien species based on the historical growth feature sequence of the target alien species and the real-time growth feature of the target alien species comprises: Extracting the attention weights of different defense measures and the current growth feature vector of the target alien species from the historical growth feature sequence of the target alien species; Based on the current growth feature vector and attention weight of the target alien species, an attention weight vector of the target alien species is obtained.

4. The method for evaluating the Giant Panda National Park's resistance to alien species invasion according to claim 3, characterized in that: The step of inputting the decoding results into the prediction module to predict the growth characteristics of the alien species population at time t+1 includes: Based on the decoding result, a plurality of candidate state allocation regions are obtained, and a plurality of regions are obtained; wherein each region represents a growth state of the alien species; Determine the area where the growth status characteristics of the target alien species population are located, calculate the growth status and confidence generated under different optional measures, define an overall loss function based on the empirical characteristics of historical information; and train the parameters in the loss function to obtain a prediction network model; The growth status characteristics of the current target alien species population are input into the prediction network model to obtain the growth status type of the alien species population at time t+1 under different measures; wherein the prediction network model uses stacked transformers as the prediction network architecture.

5. An evaluation device for the Giant Panda National Park to resist the invasion of alien species, characterized in that: include: The first acquisition module is used to acquire real-time video data of the Giant Panda National Park after adopting targeted defense measures to defend against targeted alien species; Based on the real-time video data of the Giant Panda National Park, extracting the real-time growth characteristics of the target alien species; Based on the real-time growth characteristics of the target alien species, obtaining an attention weight vector of the target alien species; The second acquisition module is used to construct context features based on the analysis of factors affecting the historical growth of the target alien species; the context features include target alien species information, ecological environment characteristics, climate conditions, biodiversity data and target resistance measures; Construct a target alien species population growth status module, which is used to convert the attention weight vector of the target alien species and the context feature into binary code using one-hot encoding and multi-hot encoding to obtain the growth feature vector encoding information of the target alien species; input the growth feature vector encoding information of the target alien species and the associated historical growth feature vector into the trained deep interest network to construct a multimodal growth feature vector of the target alien species; based on multiple multimodal growth feature vectors of target alien species, construct the growth status characteristics of the target alien species population; A semantic extraction module is used to obtain a BERT semantic extraction model; wherein the BERT semantic extraction model is obtained based on historical data training, and the historical data includes the time of biological invasion, the scope of biological invasion, the change information of the degree of biological invasion and the type of measures taken by the Giant Panda National Park; based on the BERT semantic extraction model, key information of historical biological invasions and a set of guiding measures of the Giant Panda National Park are obtained; the key information of historical biological invasions and the set of guiding measures of the Giant Panda National Park are encoded and mapped to a high-dimensional information matrix, and a preliminary biological invasion state description vector is extracted by embedding; the preliminary biological invasion state description vector is input into the trained EventKG model to realize the temporal relationship modeling between the measures taken and the changes in the biological invasion state, and then the relationship between the measures taken and the changes in the growth state of the alien species population is represented as a historical information experience vector in a continuous vector space; The evaluation module is used to input the growth status characteristics of the target alien species group into the trained stacked transformers network model to extract the historical experience information extraction network, the context information Ct extraction network and the real-time status feature extraction network; respectively decode the historical experience information extraction network, the context information Ct extraction network and the real-time status feature extraction network to obtain decoding results; respectively input the decoding results into the prediction module to predict the growth characteristics of the alien species group at time t+1; based on the growth characteristics of the target alien species at time t+1, the method of resisting the invasion of alien species in the Giant Panda National Park is evaluated.

6. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the evaluation method for the Giant Panda National Park to resist the invasion of alien species according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the processor executes the computer program to implement the evaluation method for the Giant Panda National Park to resist the invasion of alien species according to any one of claims 1-4.

Images