Cluster-based fusion-based data visualization analysis and decision-making platform for red imported fire ant monitoring

Abstract

This invention relates to the field of intelligent monitoring and management technology, specifically a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. The platform includes a data acquisition and integration module that collects information from heterogeneous data sources such as drones and sensors to establish a target object status database; a clustering fusion module that uses the OPTICS algorithm to perform clustering fusion, identifying the aggregation areas, diffusion paths, and population characteristics of target objects; an intelligent monitoring and prediction module that constructs identification and density estimation models and uses a CNN-LSTM dual-branch architecture for spatiotemporal prediction to assess future diffusion risks; a visualization and collaborative decision-making module that presents the analysis results intuitively through heatmaps, 3D geographic information systems, etc., and supports multi-scenario prevention and control strategy simulation based on agent modeling; and a dynamic resource optimization and scheduling module that uses an approximate dynamic programming algorithm to intelligently allocate and schedule resources such as prevention and control teams, drones, and pesticides to achieve precise and efficient prevention and control decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and management technology, specifically to a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. Background Technology

[0002] Red imported fire ants are recognized as a major invasive pest, and their rapid spread poses a serious threat to agricultural and forestry production, the ecological environment, and public safety.

[0003] Traditional management methods rely primarily on manual inspections and experience-based judgment. This approach suffers from limitations in data collection, including limited scope, low timeliness, and poor data quality. Consequently, subsequent strategy support lacks a comprehensive and reliable data foundation. While existing data processing methods can analyze monitoring data from single sources, they generally suffer from isolated data sources, difficulty in integrating heterogeneous data, and limited processing techniques. These technical bottlenecks restrict the ability to extract deeper information from multi-dimensional data, preventing the full utilization of monitoring data to support scientific forecasting, early warning, and resource allocation. Therefore, constructing a system that can systematically integrate multi-source heterogeneous data and utilize advanced data processing methods to provide a scientific basis for the monitoring, management, forecasting, early warning, and resource allocation of such dynamic targets has become a challenging problem in the field of data processing technology.

[0004] To address this, a data visualization analysis and decision-making platform for red imported fire ant monitoring based on cluster fusion is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. By constructing a comprehensive decision-making platform that integrates data collection, intelligent analysis, prediction and early warning, and resource optimization, the red imported fire ant control work is upgraded from the traditional experience-driven model to a data-driven modern management model.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A clustering-fusion-based data visualization, analysis, and decision-making platform for red imported fire ant monitoring includes: Data acquisition and integration module: used to collect target object monitoring data from multi-source heterogeneous data, support real-time monitoring and localized processing driven by edge AI, and form a target object status database based on prior knowledge base and quality assurance mechanism; Clustering and Fusion Module: The OPTICS clustering algorithm is applied to cluster and fuse the monitoring data of the target objects. Based on the target object status database, the clustering areas, diffusion paths and population characteristics of the target objects are identified to obtain in-depth information. Intelligent monitoring and prediction module: Combining the deep information, a target object identification and population density estimation model is constructed; a CNN-LSTM based dual-branch architecture is used to perform spatiotemporal prediction to obtain intelligent analysis results, including the diffusion path of the target object, hotspot areas and future outbreak risk level; Visualization and Collaborative Decision-Making Module: Presents the intelligent analysis results in a visual manner and provides multi-scenario prevention and control strategy simulation and evaluation functions based on agent modeling (ABM). Dynamic resource optimization and scheduling module: Based on an approximate dynamic programming algorithm, dynamically schedule and manage resources.

[0007] Preferably, the data acquisition and integration module includes: determining the deployment area and inspection route of UAVs and ground sensors based on the distribution and ecological habits of the target objects recorded in the prior knowledge base, and obtaining a monitoring network deployment scheme based on prior knowledge; the prior knowledge base includes comprehensive information on the distribution, ecological habits, and prevention and control methods of the target objects in existing scientific research; and collecting target object monitoring data from multi-source heterogeneous data, including but not limited to UAV imagery, ground sensor data, environmental factor data, and historical prevention and control data.

[0008] Preferably, the data acquisition and integration module further includes: using edge AI-driven methods, based on a prior knowledge base and a quality assurance mechanism, wherein the quality assurance mechanism is a unified data acquisition standard and an intelligent verification rule engine, and the intelligent verification rule engine includes dynamic threshold verification, spatiotemporal consistency verification, and multimodal data cross-verification; performing real-time monitoring and localization processing on the multi-source heterogeneous data, wherein the localization processing includes standardization processing, real-time verification and cleaning, to obtain a target object state database.

[0009] Preferably, the clustering fusion module includes: applying the OPTICS clustering algorithm to cluster and fuse the monitoring data of the target object; based on the aggregation point coordinates, population density, and environmental factor data in the target object status database, calculating the core distance and accessibility distance of the data points, processing the data points in sequence, and adding the reachable points to an ordered list in ascending order of accessibility distance, using the horizontal axis to represent the order of the data points in the list and the vertical axis to represent the accessibility distance of the data points, generating an accessibility map, and identifying the target object aggregation area, diffusion path, and population characteristics based on the density, scale, and interrelationships of the aggregation points to obtain deeper information.

[0010] Preferably, the intelligent monitoring and prediction module includes: combining the deep information to construct a target object recognition and population density estimation model. The model is based on a machine learning framework, which trains the machine learning structure by learning the visual and spatial features of the identified nests, automatically identifies aggregation points from new UAV imagery and ground sensor data, and directly outputs the geographic coordinates and morphological features of the newly discovered nests, and estimates the population density based on the geographic coordinates and morphological features.

[0011] Preferably, the intelligent monitoring and prediction module further includes: obtaining current target object distribution data from population density, integrating it with historical diffusion data and environmental factors from a prior knowledge base, including future weather forecasts and seasonal changes; extracting spatial feature vectors using a CNN branch to identify geographical areas suitable for the survival and diffusion of red imported fire ants, extracting temporal feature vectors using an LSTM branch to learn the changes in the target object diffusion pattern over time, and concatenating the spatial and temporal feature vectors to obtain a spatiotemporal pattern, which includes the diffusion speed and direction of the target object under temperature, rainfall, and topographic conditions, simulating and predicting the future migration of the target object, and obtaining prediction results, which include the diffusion path of the target object, hotspot areas, and future outbreak risk levels.

[0012] Preferably, the visualization and collaborative decision-making module includes: presenting the intelligent analysis results in a visual manner, the visualization manner including but not limited to heat maps, geographic information maps, time series charts and immersive 3D geographic information systems; mapping the deep information and the prediction results to charts and three-dimensional geographic space through data rendering to generate a multi-dimensional view showing the distribution density, spatiotemporal change trend and environmental correlation of target objects.

[0013] Preferably, the visualization and collaborative decision-making module further includes: providing multi-scenario prevention and control strategy simulation and evaluation functions based on agent modeling; by integrating target population characteristics, environmental factors, historical prevention and control data, and resource consumption and environmental interaction data generated by the ABM model, simulating the behavior and interaction of each agent in the target population under multi-scenario prevention and control strategies, and performing quantitative statistics and analysis on the generated simulation data to obtain population dynamics, diffusion paths, quantity change data, and specific evaluation data on prevention and control costs and environmental impacts in multiple time dimensions.

[0014] Preferably, when simulating the behavior and interaction of each agent in the target population under the multi-scenario prevention strategy, the foraging radius and state transition probability matrix of each target agent for generating a queen nuptial ant are dynamically determined by the second derivative of the local reachability distance of the corresponding cluster area in the reachability map output by the clustering fusion module; wherein, the second derivative represents the ecological crowding degree of the cluster area; when the second derivative exceeds the preset crowding degree threshold, the probability of generating a queen nuptial ant in the state transition probability matrix increases exponentially with the increase of the ecological crowding degree, with the base of the natural constant, and the foraging radius expands anisotropically towards the unsaturated grid with lower reachability distance in spatial distribution.

[0015] Preferably, the dynamic resource optimization and scheduling module includes: dynamically allocating, planning pathways, and optimizing application strategies for control teams, drones, and pesticide control resources based on an approximate dynamic programming algorithm; using the specific evaluation data as input for dynamic programming, calculating resource allocation schemes in real time, and obtaining optimized schemes for resource allocation, pathway planning, and application strategies. Compared with the prior art, the beneficial effects of the present invention are: 1. This platform integrates heterogeneous data from multiple sources, including drone imagery and ground sensors, and combines OPTICS clustering and deep learning technologies to achieve real-time and comprehensive monitoring of red imported fire ant distribution. This transforms prevention and control efforts from traditional passive response to proactive early warning, enabling accurate prediction of red imported fire ant spread paths, hotspot areas, and future outbreak risks, providing a forward-looking basis for prevention and control decisions.

[0016] 2. This platform provides a variety of intuitive visualization tools, such as heat maps and 3D GIS, to clearly present complex analysis results. More importantly, through the multi-scenario simulation function of intelligent agent modeling, decision-makers can intuitively assess the long-term effects of different prevention and control strategies on population dynamics, prevention and control costs, and environmental impacts, thereby making more scientific and accurate decisions and promoting collaborative work among different departments.

[0017] 3. One of the core advantages of this platform is its dynamic resource optimization and scheduling module. Based on an approximate dynamic programming algorithm, it intelligently allocates, plans routes, and optimizes application strategies for limited resources such as prevention and control teams, drones, and pesticides according to real-time prediction results. This fundamentally solves the problems of uneven resource allocation and low utilization rate in traditional prevention and control, and maximizes the prevention and control effect while minimizing resource consumption. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of the clustering fusion-based red imported fire ant monitoring data visualization analysis and decision-making platform of the present invention. Figure 2 This is a flowchart of the steps of the clustering fusion-based data visualization analysis and decision-making platform for red imported fire ant monitoring according to the present invention. Figure 3 This is a flowchart of the OPTICS clustering fusion process proposed in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1 to 3 This invention provides a data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on clustering fusion. The technical solution is as follows: (Refer to...) Figure 1 This is a schematic diagram of the platform's structure, for reference. Figure 2 A flowchart of the platform's steps; A clustering-fusion-based data visualization, analysis, and decision-making platform for red imported fire ant monitoring includes: Data acquisition and integration module: used to collect target object monitoring data from multi-source heterogeneous data, support real-time monitoring and localized processing driven by edge AI, and form a target object status database based on prior knowledge base and quality assurance mechanism; Clustering and Fusion Module: The OPTICS clustering algorithm is applied to cluster and fuse the monitoring data of the target objects. Based on the target object status database, the clustering areas, diffusion paths and population characteristics of the target objects are identified to obtain in-depth information. Intelligent monitoring and prediction module: Combining the deep information, a target object identification and population density estimation model is constructed; a CNN-LSTM based dual-branch architecture is used to perform spatiotemporal prediction to obtain intelligent analysis results, including the diffusion path of the target object, hotspot areas and future outbreak risk level; Visualization and Collaborative Decision-Making Module: Presents the intelligent analysis results in a visual manner and provides multi-scenario prevention and control strategy simulation and evaluation functions based on agent modeling (ABM). Dynamic resource optimization and scheduling module: Based on an approximate dynamic programming algorithm, dynamically schedule and manage resources. Example

[0021] This embodiment provides a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. It mainly introduces the data acquisition and integration module. Starting from the input end, i.e., data acquisition, it gathers heterogeneous information from multiple sources within an orchard, including drone data, ground sensor data, environmental data, and historical data. With the orchard center as (0,0), the north direction is defined as the positive Y-axis direction, and the east direction as the positive X-axis direction. Edge AI and real-time, intelligent localized preliminary processing are employed at the data source. Specifically, a lightweight convolutional neural network based on the MobileNetV2 architecture is used. Standardization, verification, and cleaning are performed based on a "prior knowledge base" and a "quality assurance mechanism." Drone images are standardized to a uniform size of 1280x720 pixels. The system performs real-time verification of ground sensor data based on thresholds such as temperature (5℃-50℃) and humidity (10%-90%) defined in the prior knowledge base, and uses median filtering to clean outliers, generating high-quality, standardized red imported fire ant distribution data. On May 20, 2023, at 09:30, the platform's monitoring network was triggered. When a drone (ID: UAV-Orchard-01) equipped with an edge AI module was patrolling over the M area of ​​the orchard, its onboard AI algorithm, based on the pre-set red imported fire ant mound image features in the prior knowledge base, such as irregularly raised fresh soil mounds, identified a highly suspected target in real time. The system immediately captured a suspected ant mound image with a confidence level of 92% at the coordinate position (30,-78).

[0022] The prior knowledge base, also known as the scientific package, is compiled from existing scientific research literature, historical monitoring records, and prevention and control archives. It is structured and stored in three categories: distribution, ecological habits, and control methods, using a triplet structure of "parameter name—value range—source." Distribution includes high-incidence zone classifications (e.g., zone B is a high-incidence zone) and the basis for monitoring deployment. Ecological habits include ant mound image characteristics (irregularly raised fresh mounds), temperature thresholds of 5–50℃ and humidity thresholds of 10%–90%, a biovibration index threshold of 0.7, high activity combination criteria, characteristics of highly suitable areas such as gently sloping loam near water sources, hydrotropism, peak nuptial flight after high temperatures and rain, and patterns such as accelerated swarming due to overcrowding. Control methods include historical diffusion data (velocity, direction, hydrotropism) retrieved from OPTICS, as well as pesticide types and baseline dosages (high-efficiency cypermethrin, 0.1% indoxacarb). This content can be accessed as needed by various modules, including monitoring deployment, image recognition, data verification, spatiotemporal prediction, ABM behavior rules, and resource scheduling.

[0023] Almost simultaneously, a ground sensor (ID: GroundSensor-B07) deployed near the location (at a distance of <2 meters) reported key data: soil temperature 28.5°C, soil moisture 65.2%, and an abnormal biological vibration index of 0.87 (threshold 0.7). This data had already undergone local calibration according to preset rules, such as cleaning within a temperature range of 5-50°C. The edge processing unit determined that the current environment—air temperature 27.8°C, 5mm of precipitation within 24 hours, and sensor readings—highly matched the conditions for high red imported fire ant activity defined in the scientific package.

[0024] The platform's quality assurance mechanism (which includes a unified data collection standard and an intelligent verification rule engine: the former specifies a unified format for all source data (e.g., images are unified to 1280×720 pixels); the latter instantiates prior knowledge base thresholds into three types of rules: dynamic threshold verification (tolerance ±15%), spatiotemporal consistency verification (matching degree ≥99%), and multimodal cross-verification, and executes them sequentially, combined with median filtering to clean outliers. Only after all rules pass are stored records generated.) immediately activates the intelligent verification rule engine. For example, dynamic threshold verification is for environmental data fluctuation tolerance ±15%, spatiotemporal consistency verification is for trajectory and operation timestamp matching degree ≥99%, and multimodal data cross-verification is for image recognition of pests and mutual verification of sensor data. It is found that the alarms from drones and ground sensors are highly consistent in time and space, and match the record of area B as a high-incidence area in historical prevention and control data. Therefore, the system automatically fuses this multi-source evidence and confirms that this is a highly credible red imported fire ant activity event. A structured, high-quality data record is created and stored in the target object status database.

[0025] This module scientifically plans the monitoring network based on prior knowledge, optimizing the quality and efficiency of data collection from the source, avoiding the waste of resources caused by blind deployment, and using edge AI for real-time localized processing and data cleaning, which improves the timeliness of data processing, reduces the computing and network burden on the central platform, and ensures the quality and accuracy of the data entering the database. Example

[0026] This embodiment provides a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. It mainly introduces the clustering fusion module, which is the core part of this platform. This module uses the OPTICS clustering algorithm to cluster and fuse the monitoring data of the target object. Based on the target object status database, it extracts in-depth information by identifying the aggregation area, diffusion path and population characteristics of red imported fire ants.

[0027] After a month of continuous monitoring, 36 similar valid nest records were accumulated in the database. At this point, the platform's deep mining module was automatically activated, and the OPTICS clustering algorithm was applied to analyze these 36 data points. These data points showed obvious spatial heterogeneity: among them, the coordinates of 22 nests were highly concentrated in orchard M area, and the estimated population density was generally high (average about 200,000); while another group of 4 nests were linearly arranged, extending from B area to F area (near the water source), and their population density was low (average about 40,000); F area itself formed another relatively sparse cluster, containing 10 nests.

[0028] Based on the location data of clustered points in the target object state database, the OPTICS clustering algorithm is applied to process the above-mentioned nest data points: with a neighborhood radius of 50 meters and a minimum number of neighbors of the core point of 4 as parameters, the core distance (the minimum radius required for its neighborhood to contain exactly the minimum number of neighbors) and the accessibility distance (the density reachability radius of the point relative to the core point, taking the larger of the core distance and the actual distance) of each point are first calculated. The points are then processed in ascending order of accessibility distance and added to an ordered list, while noise points with undefined accessibility distances are removed. When calculating the accessibility distance, terrain elevation is superimposed as an additional dimension to distinguish points at the same distance located at different elevations, thereby identifying the core clustering area, diffusion corridor, and potential settlement points of the target object. This significantly improves the identification accuracy of the core clustering area and diffusion corridor, and the fusion of multi-source environmental variables has stronger scalability in ecological protection and prevention scenarios.

[0029] Reference Figure 3 The diagram below shows the OPTICS clustering fusion process proposed in this invention. The OPTICS algorithm, with a set neighborhood radius of 50 meters and a minimum number of neighbors for core points of 4, calculates the core distance and reachability distance of all points and generates a reachability map. The horizontal axis of this map represents the order in which the algorithm processes the points, and the vertical axis represents the reachability distance of each point. After generating the reachability map, the system automatically identifies peak areas in the map where the reachability distance suddenly increases using a peak-valley detection algorithm. If the peak area is located between two obvious valleys (core clustering areas), then the area is determined to be a potential diffusion path. Simultaneously, combined with time-series data from the database, if the average discovery time of nests in area B is later than that in area A, then the final determination of the diffusion path is from area A to area B. Analyzing this map, the system reveals the following deeper information: The clustering areas were identified by two distinct valleys in the image. The first valley was wide and deep, with a consistently low accessibility distance of 5 to 20 meters, which precisely corresponds to the 10 nests in orchard area M. The system identified it as the core clustering area A. The second valley was relatively shallow, with an accessibility distance between 20 and 35 meters, corresponding to the nests in orchard area F. It was identified as the emerging clustering area B. The depth of the valleys directly reflects the density of the nests.

[0030] After identifying the aggregation areas, the system further identifies the red imported fire ant's dispersal path and population characteristics, determines the red imported fire ant aggregation density, and searches for the red imported fire ant's migration trajectory based on surrounding conditions to block the dispersal path. For example, the identified dispersal path and population characteristics are between two valleys, and the graph shows a series of high peaks, with the accessibility distance increasing to 30 to 50 meters. This area corresponds to the four linearly arranged nests, indicating that although they do not constitute a high-density cluster, they are still spatially interconnected and serve as a bridge connecting the two aggregation areas. Therefore, the system identifies a dispersal path from area A to area B. Combining the attribute data in the database, the system finds that the nests on this path and in area B not only have low population density, indicating that they are newly formed ant nests, but their locations are also significantly closer to water sources, with an average distance of less than 40 meters from the water source, while the nests in area A are far from the water source, with an average distance of more than 140 meters.

[0031] Ultimately, the module outputs not a static distribution map, but a dynamic population situation analysis report: "A mature, high-density core population (area A) located in zone M has been identified. This population is spreading towards the water source (area F) along a clear path and has formed a secondary aggregation (area B) in area F. It is recommended to immediately implement blocking measures on the diffusion path and the emerging aggregation area B, and to suppress the core aggregation area A to prevent it from continuing to be a source of diffusion." Thus, the platform has completed the entire process of transforming raw signals into strategic decision-making intelligence, demonstrating its core value in the precise and efficient control of red imported fire ants.

[0032] By using the OPTICS clustering algorithm to deeply mine data, complex clustering regions and diffusion paths can be identified. Its beneficial effect lies in its ability to extract deep information and patterns from massive amounts of data that are difficult to detect with the naked eye, providing more insightful basis for management decisions. Example

[0033] This embodiment provides a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. It mainly introduces the intelligent monitoring and prediction module, which is mainly used for continuous monitoring and prediction. This module combines the above-mentioned deep information to construct a target object identification and population density estimation model. It adopts a CNN-LSTM-based dual-branch architecture for spatiotemporal prediction. Its prediction content includes the spread path of red imported fire ants, hotspot areas, and future outbreak risk level.

[0034] The platform utilizes validated data accumulated in the "Target Object Status Database" to train a target object recognition and population density estimation model. This model adopts a cascaded structure of "detection network + density regression": the front-end detection network reuses the edge-end MobileNetV2 as the backbone feature extractor, followed by a detection head, outputting the bounding box of suspected ant mounds in UAV images and the pixel coordinates of the image center. Then, based on the UAV's pose during inspection and the coordinate system "with the orchard center as (0,0), due north as the positive Y-axis, and due east as the positive X-axis," the pixel coordinates are converted into the geographic coordinates of the nest. The back-end density regression network uses the detected ant mound morphological features (ant mound diameter, soil color, vegetation coverage) as the input vector and the number of worker ants in a single nest confirmed by manual verification as the regression label for supervised training, outputting an estimated value of the number of worker ants in the nest. Among them, the geographic coordinates are used to locate the nest and write it into the distribution map, and the morphological features are used to regress the population density, with the two respectively undertaking the functions of localization and quantification. To standardize the units of measurement, the ratio of the number of worker ants per nest to the area controlled by that nest is called the areal density (unit: ants / km²). For example, for the ant mound in the RIFA_EVENT_20230520_001 event, the model regression yielded approximately 200,000 worker ants per nest; for newly discovered initial nests at the end of the diffusion path, the model output a diameter of approximately 25 cm and a corresponding areal density of approximately 35,000 ants / km².

[0035] During a routine inspection, the model was put into practical use when the drone flew over the end of the previously identified "diffusion path." The drone captured a new suspicious mound of soil, and the model immediately analyzed it, outputting the following judgment: A suspected new red imported fire ant nest has been found, with a diameter of about 25 centimeters, a new mound of soil, a confidence level of 95%, and an estimated population density of 35,000 ants / square kilometer, which is consistent with the characteristics of an early nest. It is recommended to check and exterminate the ants within 48 hours to prevent further spread. This process is fully automated, achieving second-level discovery and quantitative assessment of new ant nests. The results are updated in real time on the platform's dynamic distribution map, providing the latest battlefield situation for subsequent spatiotemporal predictions.

[0036] Furthermore, the platform's spatiotemporal prediction model based on the CNN-LSTM architecture was launched, aiming to predict the risk of red imported fire ants spreading over the next 14 days. This model integrates four key types of data: The current distribution status includes a complete distribution map containing 36 nests, recently updated by the identification model; historical diffusion patterns are derived from the conclusions of OPTICS analysis, such as the diffusion rate from "core cluster A" to "emerging cluster B" being approximately 5 meters per day, with a clear tendency towards water; the static spatial factor CNN input is a high-resolution geographic information layer, including the orchard's digital elevation model (DEM), soil type maps such as loam and sandy soil distribution, and a raster map of "distance from water source"; the dynamic temporal factor LSTM input is the daily meteorological records of the past 90 days, including temperature, humidity, rainfall, and the weather forecast for the next 14 days, such as the meteorological interface predicting that there will be three consecutive days of temperatures reaching 33°C with afternoon thunderstorms in the coming week.

[0037] When the model starts working, the CNN branch acts like a geographer, analyzing the spatial factor layer and quickly identifying “highly suitable areas for red imported fire ants”; for example, it highlights areas with gentle slopes, plenty of sunlight, proximity to water sources, and loam soil, generating a spatial feature vector map of “habitat suitability”.

[0038] Meanwhile, the LSTM branch of the Long Short-Term Memory network acts like an experienced ecologist. Analyzing time-series data and learning from historical data, it has mastered the pattern that "high temperatures" and "after rain" are the peak periods for red imported fire ants' nuptial flights and nest building. Combining this with future weather forecasts, the LSTM predicts that under the upcoming hot and humid weather, the intensity of red imported fire ants' dispersal activity will increase by about 40%, and the success rate of nest building will increase significantly; it encodes this judgment into a temporal feature vector.

[0039] Furthermore, dynamic adversarial training is introduced to enhance model robustness. The generated adversarial samples are injected into the CNN branch as an additional spatial layer, alternating with real samples in the main model training at a set ratio. The output burst probability and ecological carrying capacity interval serve as confidence labels for the risk level map. A lightweight adversarial generative network is embedded in the CNN-LSTM dual-branch architecture. The generator adopts an improved U-shaped network architecture, including symmetrical downsampling and upsampling paths. The downsampling process progressively compresses the image size and extracts deep features, while the upsampling process fuses features at different scales through skip connections. Controllable random noise is injected into the output to enhance sample diversity. The discriminator reuses the first three convolutional layers of the main model's CNN branch and adds two fully connected layers for judging real and fake samples. An adversarial loss function with gradient penalty is used to improve training stability. Extreme environmental disturbances, including the impact of climate change and human interference on the spread of target objects, are simulated. Adversarial training samples are generated in real time to reduce the model's path error in complex scenarios and output risk confidence intervals, including burst probability and ecological carrying capacity, providing quantitative support for ABM strategy simulation.

[0040] Finally, the model concatenates and fuses the spatial feature vectors extracted by the CNN and the temporal feature vectors extracted by the LSTM, forming a comprehensive model with both spatiotemporal dimensions. The CNN branch takes high-resolution spatial layers such as the digital elevation model, soil type map, and water source distance raster map as input, and extracts spatial feature vectors of habitat suitability through multiple convolutions and pooling. The LSTM branch takes a time series consisting of daily meteorological records (temperature, humidity, and rainfall) from the past 90 days and weather forecasts for the next 14 days as input, and extracts temporal feature vectors of diffusion activity intensity on a daily basis. The two feature vectors are concatenated end-to-end to form a comprehensive spatiotemporal feature vector, which is then mapped to the diffusion velocity (unit: meters / day) and diffusion direction (azimuth) of the target object through a regression output layer. During training, the model uses "historical spatial layer and meteorological sequence at a certain moment" as input and "subsequent measured nest displacement velocity and direction" as labels, employing a regression loss for supervised learning. In this example, under the high temperature and humidity forecast, the temporal feature vector increases the predicted intensity of diffusion activity by approximately 40%. After concatenation with the spatial feature vector, regression yields a prediction that emerging population B will spread northeastward along the irrigation canal at a rate of approximately 7-8 meters per day. This model precisely describes, under specific future weather conditions, the most likely direction and speed of diffusion for a red imported fire ant population located in a particular geographical location.

[0041] After model calculation, the final output is an intuitive risk prediction report for the next 14 days, the core content of which includes: Predicted spread path: "The ant colony in the emerging cluster B is expected to continue to spread northeastward along the irrigation canal. Affected by future rainfall, the spread rate is expected to increase to 7-8 meters per day." Predicted new hotspot area: "A depression in orchard G area will become a new high-risk cluster point in the next 7-10 days because it meets the conditions of low-lying water accumulation, suitable soil and being located at the forefront of the predicted path." Future Outbreak Risk Level Map: A dynamically updated electronic map that clearly marks the risk level of each area within the orchard for the next two weeks using color blocks (e.g., green - low risk, yellow - medium risk, red - high risk). The report explicitly states that the risk level of the low-lying area in Zone G will be raised from "low" to "high (level 4 / 5)" in 5 days.

[0042] This forward-looking report was directly sent to pest control managers, enabling them to strategically shift limited pest control resources from passively treating existing ant nests to preventative pesticide application in high-risk areas and blocking key spread pathways, thereby achieving the most effective control at the lowest cost and truly "preventing problems before they occur."

[0043] By automatically identifying target object clusters and estimating population density through machine learning models, the distribution of target objects can be quickly and accurately grasped, greatly improving monitoring efficiency and accuracy and reducing the cost of manual identification. The use of a CNN-LSTM dual-branch architecture to predict future trends in time and space can provide forward-looking early warning information. Based on the early warning information, future hotspot areas and risk levels can be predicted, supporting decision-makers in formulating proactive preventive strategies. Example

[0044] This embodiment provides a data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion. It mainly introduces the visualization and collaborative decision-making module. The platform is also equipped with a visualization and collaborative decision-making module, which presents intelligent analysis results in a visual manner and provides multi-scenario prevention and control strategy simulation and evaluation functions based on agent modeling.

[0045] When decision-makers log into the system, they are first greeted by an immersive 3D Geographic Information System (GIS) that integrates multidimensional information, dynamically displaying all analysis and prediction results, and visualizing distribution and density: on the 3D map, the 36 aggregation points recorded in the database are no longer isolated points, but rather represented by their estimated population densities. For example, the 200,000 worker ants in RIFA_EVENT_20230520_001 are rendered as pulsed halos of different sizes and brightness; the core aggregation area A is thus presented as a dazzling red spot, while the halo of the emerging aggregation area B is relatively dim, intuitively demonstrating the maturity and potential for harm of the population. Users can click on any halo to immediately access its detailed information, including the discovery time, evidence chain, and historical control records.

[0046] Visualization of Diffusion and Risk: The predictions from the CNN-LSTM model are dynamically overlaid on a map. A semi-transparent, arrow-shaped flowing light band extends from area A to area B and continues towards the predicted depression in area G, clearly indicating the predicted diffusion path. The speed of the light band's flow (e.g., animation frame rate) is proportional to the predicted diffusion speed (7-8 meters per day). Simultaneously, the ground is rendered with risk level maps of different colors. The color of the depression in area G gradually changes from green to a striking red over time, visually indicating that the area will upgrade to a level 4 / 5 high-risk zone in the next 5 days.

[0047] Spatiotemporal correlation analysis: The system also provides linked charts. Users can review the distribution evolution of red imported fire ants over the past month using a timeline slider. As the slider moves, nests appear on the 3D map over time, while the adjacent time series chart synchronously displays the relationship between the number of newly added nests each day and the daily temperature and rainfall. It clearly reveals that after the three consecutive days of 33°C high temperature and thunderstorms in mid-June, the number of newly added nests showed a significant peak, thus confirming the driving role of high temperature and humidity on dispersal at the data level.

[0048] Furthermore, to select the optimal prevention and control strategy, decision-makers activated multi-scenario simulation capabilities. The system constructed a digital twin orchard in a microscopic world based on existing data. In this virtual world, each cluster is an "intelligent agent," whose behavior is determined by its attributes (population density, maturity) and environment (temperature, humidity, distance from water source). For example, a mature nest (such as one in area A) has a 0.8 probability of sending out a foraging team daily, and a 0.3 probability of producing a nuptial queen ant on a hot day after rain.

[0049] Specifically, policymakers set out two prevention and control plans to be evaluated: Option 1: Conventional comprehensive eradication Strategy Description: All 36 discovered ant nests were treated simultaneously with 0.1% indoxacarb bait. Resource consumption: Budgeted cost of 8000 yuan, requiring 4 person-days of work. Simulation process: The system initiates a simulation, with virtual control team members applying the "drug effect" to all ant nests. The "worker ant population" variable within the treated ant nests begins to decrease at a rate of 5% per day. However, due to the 3-5 day lag effect of the drug, newly formed ant nests located along the diffusion path still have a 0.1% probability of successfully differentiating into new queens and establishing tiny nests before being completely eradicated.

[0050] The simulation results after 30 days showed that all 36 existing ant nests were eliminated, but the simulation also showed that 5 tiny new ant nests were generated on the edge of areas G and F. The total population was reduced to about 2%, the control effect was rated as good, but there was a risk of recurrence, the total cost was 8,000 yuan, and the environmental impact rating was moderate (for comprehensive application of pesticides).

[0051] Option 2: Precise blocking and suppression Strategy Description: Suppressive treatment will be applied only to "Core Cluster Area A" using a more efficient contact-killing agent. Preventative baiting will also be conducted along the predicted "diffusion path" and in the "high-risk area G." Resource consumption: Budgeted cost of 5500 yuan, requiring 2 people / day of work. Simulation process: The number of worker ants in the nesting intelligence of Core Area A decreases sharply by 20% per day, and the probability of producing a mating queen ant rapidly drops to 0. A "drug barrier" is formed on the virtual ground along the diffusion path and in Area G, reducing the survival rate of any queen ant intelligence attempting to cross or establish a nest there to 0.05.

[0052] The simulation results after 30 days showed that core area A was quickly cleared, and the nests in emerging area B naturally declined due to the loss of "source" replenishment. Most importantly, the simulation showed that no new nests were successfully established, and the total population dropped to below 0.1%. The control effect was rated as excellent, the risk of recurrence was extremely low, the total cost was 5,500 yuan, and the environmental impact score was low (precision application).

[0053] By comparing the simulation data of the two schemes, decision-makers clearly saw that although Scheme 2 initially treated fewer nests, it achieved a more thorough and lasting control effect with lower cost and less environmental impact, thanks to a deep understanding of diffusion dynamics and precise intervention. Ultimately, the system recommended and adopted Scheme 2, and automatically generated detailed work instructions including specific application locations, pesticide types, and schedules.

[0054] This module presents complex analysis results intuitively, enabling decision-makers to quickly understand and grasp the current situation and future trends. It achieves efficient human-machine collaboration and can evaluate the effects, costs, and environmental impacts of different intervention strategies in a virtual environment, effectively reducing trial-and-error costs and helping to select the optimal strategy.

[0055] Furthermore, the behavioral rules of the aforementioned agents are not statically preset, but are deeply coupled with the reachability map output by the clustering fusion module to achieve dynamic adaptation. Specifically, the foraging radius of each target agent and the state transition probability matrix for generating a queen ant are dynamically determined by the second derivative of the local reachability distance of the corresponding cluster area in the reachability map output by the clustering fusion module. This second derivative characterizes the curvature of the reachability curve in the cluster area, i.e., the spatial density of nest clusters, and is used to quantify the ecological crowding of the cluster area. Taking the core cluster area A as an example, it appears as a wide and deep trough segment on the reachability map curve, with a consistently low reachability distance. The second derivative of the local reachability distance is large, and after normalization, the corresponding ecological crowding is as high as 0.85, exceeding the preset crowding threshold of 0.6. Specifically, the crowding threshold is the threshold for determining whether a gathering area has entered a state of saturation and accelerated swarming. Its dimension is consistent with the normalized ecological crowding (the second derivative of the reachability distance is normalized to [0,1]). Based on the ecological law in the prior knowledge base that "saturation of the nest means accelerated mating flight and swarming", it is calibrated to 0.6. When the ecological crowding of a certain area exceeds this value, it triggers an exponential increase in the probability of mating flight and an anisotropic expansion of the foraging radius.

[0056] At this point, the system determines that resource competition in area A is intensifying and the capacity within the nest is approaching saturation. Therefore, it dynamically adjusts its state transition probability matrix—the system performs a discrete second-order difference on the reachability distance of each point along the point sequence direction on the reachability map, obtaining the local second-order derivative, and then applies it to the global... Figure 2 The maximum value of the first derivative is linearly normalized to the interval [0,1], and is taken as the ecological crowding degree of the cluster. The state transition probability matrix is ​​arranged with the agent's three states, "routine activity - foraging - producing a queen ant for nuptial flight," as rows and columns, and the matrix elements are the transition probabilities between adjacent time steps. When the ecological crowding degree exceeds the threshold of 0.6, the transition probability corresponding to "producing a queen ant for nuptial flight" increases monotonically with the ecological crowding degree from the base probability of 0.3 according to an exponential function with the natural constant e as the base. Taking the core cluster A as an example, its normalized ecological crowding degree is 0.85. After the above exponential amplification, the transition probability jumps from 0.3 to about 0.51, while the remaining elements of the matrix are normalized row by row to keep the sum of the probabilities of each row at 1. The foraging radius stretches linearly in proportion to the degree of unsaturation in the low-accessibility direction: in the unsaturated direction of the southwest side of area A, which is adjacent to the water source and has an accessibility distance of only 5 to 20 meters, the foraging radius stretches from the base value of 20 meters to about 35 meters, while the saturated high-accessibility direction maintains the base value, thus forming a flat and elongated foraging area that extends directionally toward area F. This realistically reproduces the ecological law of red imported fire ant populations accelerating their swarming and outward dispersal under overcrowding stress at the simulation level.

[0057] Meanwhile, the foraging radius of the intelligent agent is no longer an isotropic fixed circle, but expands anisotropically towards the unsaturated grid with lower accessibility distances. After the system rasterizes the accessibility map, it identifies a low-value unsaturated area in the southwest direction of area A, adjacent to a water source, with an accessibility distance of only 5 to 20 meters. This area is determined to be a potential expansion direction where food and habitat resources have not yet been fully occupied. Therefore, the foraging radius of mature nesting agents in area A is stretched from the base value of 20 meters to about 35 meters in this direction, while it remains contracted in the direction of high accessibility distances that are already approaching saturation, ultimately forming a flat and elongated foraging area extending directionally towards the water source area F. This anisotropic foraging behavior causes the virtual population to spontaneously migrate towards area B in the digital twin orchard along the water-oriented diffusion corridor that was actually monitored, forming a closed-loop mechanism that confirms the "diffusion path from area A to area B" identified by OPTICS clustering in Example 2. It is precisely based on these dynamic behavioral rules driven by the second derivative of the reachability map that the multi-scenario simulations of the following two prevention and control schemes can accurately depict the real impact of intervention measures on population swarming and diffusion dynamics. Example

[0058] The Dynamic Resource Optimization and Scheduling module serves as the platform's command and execution center. Its core task is to transform the strategic insights output by upper-level modules, such as risk level maps and optimal prevention and control solutions, into executable and optimized tactical action commands. It utilizes an approximate dynamic programming algorithm to solve in real-time how to allocate prevention and control resources most efficiently and at the lowest cost in a changing battlefield environment to achieve the best prevention and control results.

[0059] Input: Strategic directives and real-time situation from the decision-making level Upon startup, the module first receives "Option Two: Precise Blocking and Suppression" as the overall objective from the decision support module. Simultaneously, it loads the latest situational data. Target list: The target of suppression is 10 mature nests in the core cluster area A, such as RIFA_EVENT_20230520_001, which requires the use of contact-type pesticides (such as high-efficiency cypermethrin).

[0060] The target for blocking is the predicted spread path and the high-risk area in zone G, with a total length of approximately 300 meters. Preventive baiting (such as 0.1% indoxacarb) is required.

[0061] Eliminate 10 initial nests in emerging cluster area B using standard bait.

[0062] Available resources: Manpower: 2 prevention and control teams, 2 people in each team. Team 1 carries two types of pesticides, while team 2 carries only bait.

[0063] Equipment: One pesticide spraying drone (UAV-Spray-01), with a flight time of 25 minutes and a payload of 10 liters, specifically designed for precise application to barrier strips.

[0064] Chemicals: The warehouse currently has 5 liters of high-efficiency cypermethrin and 20 kg of 0.1% indoxacarb bait.

[0065] Dynamic environment: Real-time GIS map, including travel time between points, terrain obstacles, etc. Weather forecast shows thunderstorms tomorrow afternoon, making the morning the best window for operations.

[0066] Core Engine: Real-time optimization based on approximate dynamic programming The module's ADP algorithm treats the entire prevention and control task as a multi-stage decision-making process. The "state" is the set of currently unfinished tasks and the remaining resources; the "decision" is to assign the next task to each team and drone at each time step, such as every 15 minutes; and the "objective" is to minimize the total operation time and cost.

[0067] This approximate dynamic programming uses a deep Q-network as a value function approximator: each prevention and control team and drone is defined as an agent, whose state is encoded as a vector of current position, remaining drug dosage, remaining endurance, and set of unfinished tasks; the action is a discrete choice of "going to and processing a target"; the reward is defined as the reduction in the expected number of future new nests by task completion minus the weighted sum of operation time and drug cost, where the expected number of future new nests is calculated from the prevention and control cost and environmental impact assessment data of each scenario output by the ABM module, and thus the specific assessment data is incorporated into the reward function as the input of dynamic programming. The Q-network takes the above state vector as input and the long-term value of each candidate action as output, uses experience replay to store historical transfer samples, target network separation to estimate stable Q values, and is pre-trained with a large amount of historical prevention and control data and simulation data, and selects the action with the highest value in any state. Since treating the source nests (core area A) can significantly reduce the expected number of future new nests, its long-term value is the highest, so it is prioritized for team 1 carrying highly effective agents; drones are the most efficient at treating linear blocking zones, so the B-Path task is assigned to drones. This model has been pre-trained with a large amount of historical prevention and control data and simulation data; it not only knows that prioritizing the treatment of "source" nests (such as core area A) can effectively reduce the number of future new nests, thus having a higher long-term value; but also knows that drones are far more efficient than manpower in treating linear areas (blocking zones).

[0068] Real-time decision-making and path planning: The algorithm begins to plan actions for the upcoming workday.

[0069] Task Priority and Allocation: Based on the value function, the algorithm determines that suppressing core area A and blocking the spread path have the highest priority. It assigns P-Task to the more experienced team 1 because they carry the necessary high-efficiency reagents. Simultaneously, it assigns B-Task to UAV-Spray-01, as this is the most efficient method of execution. The task of clearing secondary cluster B is assigned to team 2.

[0070] Path and sequence optimization: For team 1, the algorithm calculated the shortest path to visit 10 nests in area A, with a total distance of 1.2 kilometers and an estimated time of 2.5 hours. For the drone, the algorithm planned an "S"-shaped flight path covering a 300-meter barrier to ensure uniform drug coverage, with an estimated flight time of 18 minutes; the path for team 2 was also optimized similarly.

[0071] Resource and Time Constraints: The algorithm checks resource consumption. Team 1 needs 4 liters of high-efficiency cypermethrin (sufficient stock), and the drone and Team 2 together need 15 kg of bait (sufficient stock). All tasks are expected to be completed within 3 hours, fully meeting the operational window before the afternoon thunderstorm. Furthermore, the module issues specific, quantifiable action instructions to the mobile terminals of frontline personnel and the drone ground station: Team 1's task is to suppress core area A, time: 08:30-11:00, path: already loaded into the handheld GIS device, please operate according to the navigation sequence; Instruction: For targets P01-P10 in the list, use high-efficiency cypermethrin for root irrigation treatment, 0.4 liters per nest.

[0072] The operator of UAV-Spray-01 is tasked with blocking the spread path. Time: 08:45-09:15. Flight path: B-Path-01. Flight path has been uploaded. Please confirm and execute. Instructions: Use 0.1% indoxacarb bait and spread it evenly at a standard rate of 20 grams per square meter.

[0073] During task execution, this module continuously receives real-time status feedback from frontline personnel and equipment, such as task completion and pesticide remaining levels. In case of unexpected situations, such as equipment malfunction or the discovery of unmarked large ant nests, the ADP algorithm will immediately recalculate, adjusting subsequent task allocation and paths within seconds to dynamically optimize the plan and ensure the entire prevention and control operation remains on the optimal track. In this way, the platform accurately translates high-level strategies into efficient, low-cost, and flexible ground operations, forming a complete intelligent closed loop of "perception-analysis-decision-execution".

[0074] Based on an approximate dynamic programming algorithm, this module takes the output of the intelligent monitoring and prediction module as input, and calculates and optimizes the dynamic allocation, path planning and application strategies of prevention and control teams, drones and pesticide resources in real time. This enables dynamic, precise and efficient scheduling of prevention and control resources, such as personnel, equipment and pesticides, which significantly improves the intervention effect and saves prevention and control costs.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on clustering and fusion, characterized in that: include: Data acquisition and integration module: used to collect target object monitoring data from multi-source heterogeneous data, support real-time monitoring and localized processing driven by edge AI, and form a target object status database based on prior knowledge base and quality assurance mechanism; Clustering and Fusion Module: The OPTICS clustering algorithm is applied to cluster and fuse the monitoring data of the target objects. Based on the target object status database, the clustering areas, diffusion paths and population characteristics of the target objects are identified to obtain in-depth information. Intelligent monitoring and prediction module: Combining the deep information, a target object identification and population density estimation model is constructed; a CNN-LSTM based dual-branch architecture is used to perform spatiotemporal prediction to obtain intelligent analysis results, including the diffusion path of the target object, hotspot areas and future outbreak risk level; Visualization and Collaborative Decision-Making Module: Presents the intelligent analysis results in a visual manner and provides multi-scenario prevention and control strategy simulation and evaluation functions based on agent modeling (ABM). Dynamic resource optimization and scheduling module: Based on an approximate dynamic programming algorithm, dynamically schedule and manage resources.

2. The data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on cluster fusion as described in claim 1, characterized in that, The data acquisition and integration module includes: Based on the distribution and ecological habits of the target objects recorded in the prior knowledge base, the deployment areas and inspection routes of UAVs and ground sensors are determined, resulting in a monitoring network deployment scheme based on prior knowledge. The prior knowledge base includes comprehensive information on the distribution, ecological habits, and prevention and control methods of the target objects from existing scientific research. Monitoring data of the target objects are collected from multi-source heterogeneous data, including but not limited to UAV imagery, ground sensor data, environmental factor data, and historical prevention and control data.

3. The data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on cluster fusion as described in claim 1, characterized in that, The data acquisition and integration module also includes: Driven by edge AI, and based on a prior knowledge base and a quality assurance mechanism, the quality assurance mechanism consists of a unified data collection standard and an intelligent verification rule engine. The intelligent verification rule engine includes dynamic threshold verification, spatiotemporal consistency verification, and multimodal data cross-verification. The multi-source heterogeneous data is monitored in real time and processed locally. The local processing includes standardization, real-time verification, and cleaning to obtain a target object state database.

4. The data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on cluster fusion as described in claim 1, characterized in that, The clustering fusion module includes: The OPTICS clustering algorithm is applied to cluster and fuse the monitoring data of the target object. Based on the aggregation point coordinates, population density, and environmental factor data in the target object status database, the core distance and accessibility distance of the data points are calculated, and the data points are processed in sequence. The accessible points are added to an ordered list in ascending order of accessibility distance. The horizontal axis represents the order of the data points in the list, and the vertical axis represents the accessibility distance of the data points to generate an accessibility map. Based on the density, scale, and interrelationship of the aggregation points, the target object aggregation area, diffusion path, and population characteristics are identified to obtain deeper information.

5. The data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion as described in claim 1, characterized in that, The intelligent monitoring and prediction module includes: combining the deep information to construct a target object recognition and population density estimation model. The model is based on a machine learning framework. By learning the visual and spatial features of the identified nests, the machine learning structure is trained to automatically identify aggregation points from new UAV imagery and ground sensor data. The model directly outputs the geographic coordinates and morphological features of the newly discovered nests and estimates the population density based on the geographic coordinates and morphological features.

6. The data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on cluster fusion as described in claim 1, characterized in that, The intelligent monitoring and prediction module also includes: The distribution data of the current target object is obtained from the population density and integrated with historical diffusion data and environmental factors from the prior knowledge base, including future weather forecasts and seasonal changes. Spatial feature vectors are extracted using a CNN branch to identify geographical areas suitable for the survival and diffusion of red imported fire ants, and temporal feature vectors are extracted using an LSTM branch to learn the changes in the diffusion pattern of the target object over time. The spatial and temporal feature vectors are concatenated to obtain a spatiotemporal pattern, which includes the diffusion speed and direction of the target object under temperature, rainfall, and topographic conditions. The future migration of the target object is simulated and predicted to obtain prediction results, which include the diffusion path of the target object, hotspot areas, and the future outbreak risk level.

7. The data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion as described in claim 1, characterized in that, The visualization and collaborative decision-making module includes: This is used to present the intelligent analysis results in a visual manner, including but not limited to heat maps, geographic information maps, time series charts, and immersive 3D geographic information systems; through data rendering, the deep information and the prediction results are mapped onto charts and three-dimensional geographic space to generate a multi-dimensional view showing the distribution density, spatiotemporal change trends, and environmental correlations of target objects.

8. The data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion as described in claim 1, characterized in that, The visualization and collaborative decision-making module also includes: It provides simulation and evaluation functions for multi-scenario prevention and control strategies based on agent modeling; by integrating target population characteristics, environmental factors, historical prevention and control data, and resource consumption and environmental interaction data generated by the ABM model, it simulates the behavior and interaction of each agent in the target population under multi-scenario prevention and control strategies, and performs quantitative statistics and analysis on the generated simulation data to obtain population dynamics, diffusion paths, population change data, and specific assessment data on prevention and control costs and environmental impacts in multiple time dimensions.

9. The data visualization analysis and decision-making platform for red imported fire ant monitoring based on clustering fusion as described in claim 8, characterized in that, When simulating the behavior and interaction of each agent in the target population under the multi-scenario prevention strategy, the foraging radius and state transition probability matrix of each target agent for generating a queen nuptial ant are dynamically determined by the second derivative of the local reachability distance of the corresponding cluster area in the reachability map output by the clustering fusion module; wherein, the second derivative represents the ecological crowding degree of the cluster area; when the second derivative exceeds the preset crowding degree threshold, the probability of generating a queen nuptial ant in the state transition probability matrix increases exponentially with the increase of the ecological crowding degree to the base of the natural constant, and the foraging radius expands anisotropically toward the unsaturated grid with lower reachability distance in spatial distribution.

10. The data visualization, analysis, and decision-making platform for red imported fire ant monitoring based on clustering fusion as described in claim 1, characterized in that, The dynamic resource optimization and scheduling module includes: Based on an approximate dynamic programming algorithm, dynamic allocation, path planning, and pesticide application strategy optimization are performed on prevention and control teams, drones, and pesticide application resources. The specific evaluation data is used as input for dynamic programming to calculate resource allocation schemes in real time, resulting in optimized schemes for resource allocation, path planning, and pesticide application strategies.

Images