A multi-dimensional perception of water, land and air internet of things termite precision prevention and control system and method

Abstract

The present application belongs to the technical field of termite control, and particularly relates to a multi-dimensional sensing water-land-air Internet of Things termite precision control system. The system comprises a movable device, a fixed device and a background software system. The method comprises: collecting termite damage information, meteorological elements and position information through multiple devices and uploading; a termite intelligent monitoring platform of the background software system identifies termites based on a convolutional neural network model, analyzes and predicts termite outbreaks in combination with a multivariate statistical regression model and a Bayesian neural network model, and performs early warning; devices are dispatched for killing according to the early warning result, and the identification and prediction models are dynamically optimized based on the killing data. The present application realizes intelligent collaborative control of the whole process of monitoring, early warning, killing and evaluation.

Description

Technical Field

[0001] This application relates to the field of termite control technology, and in particular to a multi-dimensional sensing IoT-based precision termite control system and method for land, sea and air. Background Technology

[0002] The rise of multi-device collaborative termite control technology is a product of the deep integration of traditional termite control methods with modern information technology, the Internet of Things, artificial intelligence and automated equipment.

[0003] Traditional methods heavily rely on manual inspections and experience-based judgment. Termite activity is often concealed (e.g., inside walls, underground), making early detection difficult; by the time obvious surface traces appear, the damage is already quite severe. Regular monitoring points are scattered, data feedback is untimely, and real-time early warning and trend analysis are impossible, resulting in reactive control. Spraying pesticides and setting up bait stations are often carried out independently, lacking information linkage between different devices (such as monitoring stations and extermination devices), failing to form a closed loop of "monitoring-location-extermination-verification." Data generated during the control process (such as location, quantity, and species) is not systematically recorded and analyzed, making it difficult to optimize control strategies and assess long-term effects.

[0004] Chinese Patent, Publication No. CN105230588B, Publication Date: July 20, 2018, discloses an Internet of Things-based remote automatic termite monitoring system, comprising: multiple termite monitoring devices placed at different depths underground, each termite monitoring device used to store bait to attract termites; multiple automatic monitoring data acquisition devices, each used to monitor environmental data and video data of termite activity within the shell, and to issue an alarm signal when the environmental data exceeds the standard or the termite activity is abnormal; a remote automatic monitoring subsystem receives environmental data and video data, forms continuous historical termite monitoring data, and analyzes the impact of environmental data on termite activity and analyzes the ecological characteristics of termites based on the historical termite monitoring data.

[0005] The shortcomings of the above technical solutions are: 1. They can only achieve monitoring and data collection, but cannot perform real-time intervention or extermination. For example, the system collects termite activity data through sensors and cameras, but can only issue alarm signals and cannot automatically perform extermination operations. This leads to low control efficiency, reliance on manual follow-up treatment, and difficulty in dealing with sudden termite infestations. 2. Although it can analyze the impact of environmental data on termite activity (such as temperature and feeding relationships), it lacks real-time evaluation and feedback on control effectiveness. The document mentions "forming continuous historical termite monitoring data," but does not address the evaluation of the effect after extermination, resulting in a lag in adjusting control strategies and an inability to achieve precise optimization. Summary of the Invention

[0006] The purpose of this application is to provide a multi-dimensional sensing IoT-based precision termite control system and method. By introducing multi-device collaboration and extermination devices across land, water, and air, it solves the technical problems of disconnect between monitoring and prevention, lack of real-time active extermination, and closed-loop effect evaluation in the existing technology.

[0007] To achieve the above objectives, this application adopts the following technical solution:

[0008] This application provides a multi-dimensional sensing IoT-based precision termite control system for land, water, and air. The system includes a mobile device, a fixed device, and a backend software system. The mobile and fixed devices collect termite damage information and upload it to the backend software system. The backend software system performs statistical analysis based on the termite damage information to obtain termite control measures and termite outbreak early warning information. At least one of the mobile devices is equipped with a termite extermination device, which is controlled based on the termite control measures. The termite extermination device performs termite extermination, and the mobile device collects termite extermination data and uploads it to the backend software system. The backend software system evaluates the control effect based on the termite extermination data to obtain a control effect evaluation result.

[0009] As a preferred technical solution, the mobile device includes at least one of the following: patrol drone, extermination drone, self-propelled robot, robot dog, unmanned boat, nest explorer, wheeled (tracked) drone;

[0010] The patrol drone is equipped with a camera, an infrared thermal imaging device, and a GPS locator. The patrol drone captures video information through the camera and the infrared thermal imaging device, obtains the location of termite damage through the GPS locator, and sends the video information and the location of termite damage to the backend software system in real time. Alternatively, the patrol drone analyzes and identifies termite damage information through its own processor and uploads the termite damage information identification results to the backend software system.

[0011] The termite extermination drone is equipped with a termite extermination device or a laser emitter or an ultrasonic emitter. The termite extermination device on the drone is a spraying device. The termite extermination device on the drone kills termites by spraying powder, bait or liquid, or by emitting laser or ultrasonic waves, and uploads the termite extermination data to the background software system.

[0012] The wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs are equipped with a second camera, a second infrared thermal imaging device, and a second GPS locator. The wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs capture video information via the second camera and the second infrared thermal imaging device. They also obtain the location of termite infestations via the second GPS locator. The wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs are equipped with termite extermination devices. The wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs are equipped with termite extermination devices. The termite extermination device includes at least one of the following: a powder sprayer, a liquid sprayer, a robotic arm capable of dispensing bait, a laser emitting device, and an ultrasonic emitting device. The wheeled or tracked unmanned vehicle, the self-propelled robot, and the robot dog re-inspect the termite infestation locations discovered by the patrol drone and exterminate the termites based on the termite extermination data. The wheeled or tracked unmanned vehicle, the self-propelled robot, and the robot dog also collect termite infestation information in confined spaces. The wheeled or tracked unmanned vehicle, the self-propelled robot, and the robot dog upload the termite infestation information and termite extermination data to the background software system.

[0013] The unmanned vessel is equipped with a termite-detecting device, which collects termite damage information. The unmanned vessel is also equipped with a termite extermination device, which is a termite attractant lamp. The unmanned vessel uploads the termite damage information and termite extermination data to the background software system.

[0014] The nest detector is used to detect underground termite nests, and it uploads termite damage information to the background software system.

[0015] As a preferred technical solution, the fixed equipment includes at least one of the following: a termite trapping lamp and a termite monitoring device;

[0016] The termite trapping lamp integrates a termite attracting lamp, a weather station, and a camera. The weather station transmits weather information to the background software system. When termites swarm, the termite attracting lamp attracts winged adult termites into the lamp. The camera identifies and counts termite species. The termite trapping lamp uploads termite damage information and termite extermination data to the background software system.

[0017] The termite monitoring device includes a shell, bait to attract termites, and camera three. The termite monitoring device is installed on the surface or underground. Camera three identifies termite species and obtains termite damage information. The termite monitoring device uploads termite damage information and termite extermination data to the background software system.

[0018] This application also provides a multi-dimensional sensing method for precise termite control via the Internet of Things (IoT) covering land, water, and air, the method comprising:

[0019] S101 collects termite damage information and meteorological elements through multiple devices on land, sea, and air, and marks termite-infested areas using GPS positioning.

[0020] S102, uploads termite damage information, meteorological elements and termite-infested areas to the backend software system. The backend software system includes a termite infestation big data center and a termite intelligent monitoring platform. The termite infestation big data center uses a layered architecture to store data. The termite infestation big data center includes a data access layer, an entity class library and a business logic layer, and supports NoSQL features to store variable fields in JSON format.

[0021] S103, the termite intelligent monitoring platform uses a convolutional neural network model to identify collected termite damage information to obtain identification results. The convolutional neural network model includes convolutional layers and pooling layers. The convolutional layers calculate termite damage feature values ​​by sliding the convolutional kernel over the input termite damage information, and the pooling layers perform feature dimensionality reduction on the termite damage feature values ​​through average pooling or max pooling. The identification results are compared with a termite image database to determine the termite species and quantity.

[0022] S104, the termite intelligent monitoring platform combines meteorological elements, historical termite outbreak information, and termite-infested areas. It establishes a quantitative relationship between meteorological elements and termite damage characteristic values ​​through a multivariate statistical regression model, and uses a Bayesian neural network model to predict termite swarming paths, outbreak times, and scale. When the termite damage characteristic values ​​exceed the preset damage threshold, the termite intelligent monitoring platform issues an early warning and obtains the warning results.

[0023] S105, the background software system dispatches termite extermination devices carried by multiple equipment on land, sea and air according to the early warning results, so as to carry out termite extermination in the early warning termite-infested areas and obtain termite extermination data.

[0024] S106, the background software system optimizes the convolutional neural network model in step S103 and the Bayesian neural network model in step S104 based on the termite extermination data.

[0025] As a preferred technical solution, the multi-device (land, sea, and air) in step S101 includes at least one of the following: patrol drone, extermination drone, self-propelled robot, robot dog, unmanned boat, nest finder, wheeled or tracked unmanned vehicle;

[0026] Among them, patrol drones prioritize patrolling large areas without tree cover, wheeled or tracked unmanned vehicles conduct secondary inspections of relatively flat areas such as grasslands and dam tops, self-propelled robots and robot dogs conduct secondary inspections of forests, shrublands, and small indoor spaces, and unmanned boats automatically navigate to designated locations according to instructions from the backend system, then turn on termite attracting lights, while nest detectors simultaneously detect underground termite nest structures; the multiple devices on land, sea, and air achieve location synchronization through GIS data services, which support personnel tracking and geofencing functions, can define a dynamic patrol range with a radius of 1 kilometer centered on the monitoring point, and adjust the scheduling priority of multiple devices on land, sea, and air in real time according to the termite risk level.

[0027] As a preferred technical solution, the optimization of the convolutional neural network model in step S103 includes local reparameterization: for termite image recognition tasks, a mixed-scale Gaussian prior distribution is introduced into the weights of the convolutional layers, and the posterior distribution is approximated through variational inference; during the training process of the convolutional neural network model, local reparameterization is used to replace global parameter sampling, and the mean and variance of the output termite damage feature values ​​are directly calculated to reduce computational costs; at the same time, the pooling layer captures the local invariance of termite morphology through the receptive field mechanism, and combines image enhancement technology to preprocess blurred termite images, thereby improving the robustness of the convolutional neural network model in recognizing subtle features including termite swarming holes and mud lines.

[0028] As a preferred technical solution, the linkage prediction mechanism of the multivariate statistical regression model and the Bayesian neural network model in step S104 includes: the multivariate statistical regression model establishes a linear relationship between meteorological elements and termite damage characteristic values ​​through the least squares method, and outputs a preliminary prediction interval; the Bayesian neural network model refines the preliminary prediction interval, calculates the probability distribution of termite swarming paths through variational inference and Monte Carlo sampling; if the confidence interval output by the multivariate statistical regression model exceeds the preset confidence interval range, the high uncertainty mode of the Bayesian neural network model is triggered, the weights are regularized using a mixed-scale Gaussian prior, and the prediction results are dynamically adjusted in combination with the termite damage information of historical termite outbreaks; finally, a fan-shaped termite risk map is displayed through the visualization interface of the termite intelligent monitoring platform, with different colors marking the termite risk level.

[0029] As a preferred technical solution, the feedback control mechanism for the multi-device termite extermination operation involving land, water, and air in step S105 includes: after the extermination drone accurately applies pesticides based on the early warning results, wheeled or tracked unmanned vehicles, self-propelled robots, or robot dogs carrying high-definition cameras re-inspect the extermination area and collect actual termite extermination data; if residual termite activity is detected, the wheeled or tracked unmanned vehicles, self-propelled robots, or robot dogs use robotic arms to deliver bait or spray powder for secondary treatment; simultaneously, unmanned boats check the condition of the water-facing slope through monitoring equipment and provide real-time feedback on the termite infestation situation on the water-facing slope to the background software system; all termite extermination data is stored through a static file server and linked to a GIS geofence database to form a mapping relationship between termite extermination data and the spatial location of termites, which is used to optimize subsequent scheduling strategies.

[0030] As a preferred technical solution, the high-availability architecture of the termite infestation big data center includes: using sharding technology to distribute termite monitoring data for storage, supporting physical replication between master and slave databases and off-site disaster recovery; the data access layer maps database tables to objects through entity class libraries, and the business logic layer optimizes the caching of query results; for unstructured data, a lightweight distributed file system is used to achieve file synchronization and load balancing, while supporting NoSQL features to store variable fields in JSON format, ensuring the stability of the backend software system when data surges.

[0031] As a preferred technical solution, the model optimization in step S106 specifically includes an incremental learning mechanism: the background software system compares the early warning results with the actual termite damage information of the termite outbreak, and calculates the prediction error; if the prediction error exceeds the preset prediction error threshold, the incremental learning process is started, and the newly collected termite damage information and meteorological elements are used as incremental samples, and the weight parameters are locally updated through the reparameterization technology of the Bayesian neural network model; at the same time, the feature extraction layer of the convolutional neural network model adopts a dynamic convolution kernel adjustment strategy, and optimizes the convolution step size and filling rules according to the seasonal activity characteristics of termites, so that the convolutional neural network model can continuously adapt to the morphological changes of termites in different geographical environments.

[0032] Compared with the prior art, the beneficial effects of this application are as follows:

[0033] Mobile devices (such as drones or ground robots) and fixed devices (such as underground sensors) work together to cover multiple dimensions of space, including water, land, and air, overcoming the limitations of existing technologies that are limited to underground monitoring. For example, mobile devices can quickly respond to early warning information and accurately exterminate termite-infested areas, while fixed devices continuously monitor environmental changes. The backend software system integrates multi-source data (termite damage information and extermination data), and generates prevention and control measures and early warnings through statistical analysis, achieving integrated "monitoring-decision-action".

[0034] At least one mobile device is equipped with a termite extermination device (such as a chemical spraying or physical removal device). Based on a backend software system, the extermination measures are automatically executed, overcoming the limitation of existing technologies that can only issue alarms but cannot intervene. After extermination, the device collects and uploads extermination data (such as termite reduction or area coverage), and the backend software system evaluates the control effectiveness and generates assessment results. This closed-loop design allows the system to continuously optimize control strategies, for example, by adjusting extermination parameters based on historical data to enhance long-term control effectiveness.

[0035] This application achieves full-process automation through Internet of Things (IoT) technology, eliminating the need for human intervention from data collection to extermination execution, thus significantly improving prevention and control efficiency and response speed. Attached Figure Description

[0036] Figure 1 This is a general framework diagram of the intelligent termite monitoring system (i.e., the multi-device collaborative termite control system involving land, water, and air as described in this application);

[0037] Figure 2 A topology diagram of the intelligent termite monitoring system (i.e., the multi-device collaborative termite control system involving land, sea, and air as described in this application);

[0038] Figure 3 This is a diagram illustrating database fusion.

[0039] Figure 4 This is a schematic diagram illustrating the principle of a convolutional neural network model.

[0040] Figure 5 Convolutional fill map for the image;

[0041] Figure 6 A pixel value map of the image within the receptive field;

[0042] Figure 7 The graphs show average pooling and max pooling.

[0043] Figure 8 This is a magnified view of the sensing field of the pooling layer.

[0044] Figure 9 This is a schematic diagram of an intelligent monitoring and analysis system.

[0045] Figure 10 A graph showing the quantitative relationship between meteorological elements and termite damage parameters;

[0046] Figure 11 A schematic diagram of the structure of a Bayesian neural network prediction model for termite outbreaks;

[0047] Figure 12 A schematic diagram showing the location of the termite invasion;

[0048] Figure 13 A graph showing model training and model scheduling;

[0049] Figure 14 A schematic diagram demonstrating the functionality;

[0050] Figure 15 For standardized cross-sectional diagrams of dikes;

[0051] Figure 16 This is a schematic diagram of the dam. Detailed Implementation

[0052] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0053] This application provides a multi-dimensional sensing IoT-based precision termite control system for land, water, and air, which includes mobile devices, fixed devices, and a backend software system.

[0054] Mobile and fixed devices collect termite damage information and upload it to the backend software system. The backend software system performs statistical analysis based on the termite damage information to obtain termite control measures and early warning information for termite outbreaks.

[0055] At least one mobile device is equipped with a termite extermination device, and at least one mobile device controls the termite extermination device to exterminate termites based on termite control measures. The mobile device collects termite extermination data and uploads it to the background software system.

[0056] The backend software system evaluates the prevention and control effectiveness based on termite extermination data and obtains the prevention and control effectiveness evaluation results.

[0057] Furthermore, mobile devices include at least one of the following: patrol drones, extermination drones, self-propelled robots, robot dogs, unmanned boats, nest explorers, and wheeled or tracked unmanned vehicles.

[0058] The patrol drone is equipped with a camera, an infrared thermal imaging device, and a GPS locator. The patrol drone captures video information through the camera and the infrared thermal imaging device, and obtains the location of termite damage through the GPS locator. The patrol drone sends the video information and the location of termite damage to the backend software system in real time, or the patrol drone analyzes and identifies termite damage information through its onboard processor and uploads the termite damage information identification results to the backend software system.

[0059] The termite extermination drone is equipped with a termite extermination device, which may be a spraying device, a laser emitter, or an ultrasonic emitter. The drone uses powder or liquid to kill termites in the affected areas, or uses a laser emitter or ultrasonic emitter to kill individual termites, and uploads the termite extermination data to the backend software system.

[0060] Wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs are equipped with two cameras, two infrared thermal imaging devices, and two GPS locators. The cameras and thermal imaging devices capture video information, while the GPS locators pinpoint the location of termite infestations. These vehicles are also equipped with termite extermination devices. The termite control device includes at least one of the following: a powder sprayer, a liquid sprayer, a robotic arm or laser emitter capable of dispensing bait, an ultrasonic emitter, a wheeled or tracked unmanned vehicle, a self-propelled robot, or a robot dog. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog also re-inspects the termite infestation locations discovered by the patrol drone and carries out termite extermination based on the termite extermination data. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog also collects termite infestation information in confined spaces. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog uploads the termite infestation information and termite extermination data to the backend software system.

[0061] The unmanned surface vessel (USV) is equipped with termite-detecting equipment to collect information on termite damage. It also carries a termite-killing device, which is a termite-attracting lamp. The USV uploads the termite damage information and termite-killing data to the backend software system.

[0062] The nest detector is used to detect underground termite nests and uploads information about termite damage to the backend software system.

[0063] Furthermore, the fixed equipment includes at least one of the following: termite trapping lamps and termite monitoring devices.

[0064] The termite trapping lamp integrates a termite attractant lamp, a weather station, and a camera. The weather station transmits meteorological information to the backend software system. When termites swarm, the termite attractant lamp lures winged adult termites into the lamp. The camera identifies and counts termite species. The termite trapping lamp uploads information on termite damage and termite extermination data to the backend software system.

[0065] The termite monitoring device includes a shell, bait to attract termites, and three cameras. The termite monitoring device is installed on the surface or underground. The three cameras identify termite species and obtain information on termite damage. The termite monitoring device uploads the termite damage information and termite extermination data to the background software system.

[0066] This application also provides a multi-dimensional sensing method for precise termite control via the Internet of Things (IoT) covering land, water, and air, the method comprising:

[0067] S101 collects termite damage information and meteorological elements through multiple devices on land, sea, and air, and marks termite-infested areas using GPS positioning.

[0068] S102 uploads termite damage information, meteorological factors, and termite-affected areas to the backend software system, which includes a termite infestation big data center and a termite intelligent monitoring platform. The termite infestation big data center uses a layered architecture to store data, including a data access layer, an entity class library, and a business logic layer, and supports NoSQL features to store variable fields in JSON format.

[0069] S103, the intelligent termite monitoring platform uses a convolutional neural network (CNN) model to identify collected termite damage information and obtain identification results. The CNN model includes convolutional layers and pooling layers. The convolutional layers calculate termite damage feature values ​​by sliding convolutional kernels across the input termite damage information, while the pooling layers perform feature reduction on the termite damage feature values ​​using average pooling or max pooling. The identification results are compared with a termite image database to determine the termite species and quantity.

[0070] S104, the intelligent termite monitoring platform combines meteorological factors, historical termite outbreak information, and termite-affected areas. It establishes a quantitative relationship between meteorological factors and termite damage characteristic values ​​through a multivariate statistical regression model, and uses a Bayesian neural network model to predict termite swarming paths, outbreak times, and scale. When the termite damage characteristic values ​​exceed a preset damage threshold, the intelligent termite monitoring platform issues an early warning and provides the warning results.

[0071] S105, the background software system dispatches termite extermination devices carried by multiple equipment on land, sea and air according to the early warning results, so as to carry out termite extermination in the early warning termite-infested areas and obtain termite extermination data.

[0072] S106, the background software system optimizes the convolutional neural network model in step S103 and the Bayesian neural network model in step S104 based on the termite extermination data.

[0073] Furthermore, the multi-device system in step S101 includes at least one of the following: patrol drone, extermination drone, self-propelled robot, robot dog, unmanned boat, nest explorer, wheeled or tracked unmanned vehicle.

[0074] Among these measures, patrol drones prioritize surveying large areas without tree cover, while wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs conduct secondary surveys of forests, shrublands, and confined indoor spaces. Unmanned boats automatically navigate to designated locations based on instructions from the backend system, activate termite-attracting lights, and simultaneously detect underground termite nest structures using nest detectors. The multiple devices across land, sea, and air are synchronized via GIS data services. These services support features such as tagging and geofencing, enabling the delineation of a dynamic patrol area with a 1-kilometer radius centered on a monitoring point, and real-time adjustment of the scheduling priority for these devices based on the termite risk level.

[0075] Based on the above technical solutions, a GIS-based dynamic task allocation strategy is proposed. By linking spatial data (such as geofencing) with real-time risk levels, precise scheduling of multiple devices can be achieved. This integrates the functions of scattered devices into an organic collaborative logic, improving patrol efficiency.

[0076] Furthermore, the optimization of the convolutional neural network model in step S103 includes local reparameterization: for the termite image recognition task, a mixed-scale Gaussian prior distribution is introduced into the weights of the convolutional layers, and the posterior distribution is approximated through variational inference. During the training of the convolutional neural network model, local reparameterization is used to replace global parameter sampling, and the mean and variance of the output termite damage feature values ​​are directly calculated, reducing computational costs. At the same time, the pooling layer captures the local invariance of termite morphology through the receptive field mechanism, and combines image enhancement technology to preprocess blurred termite images, improving the robustness of the convolutional neural network model in recognizing subtle features, including termite swarming holes and mud lines.

[0077] Based on the above technical solutions, we focus on the fusion and optimization of Bayesian neural networks and image enhancement, and integrate local reparameterization and image enhancement techniques into a dedicated training process for termite identification. Through uncertainty modeling and image enhancement, we improve the adaptability of the model in complex environments.

[0078] Furthermore, the linkage prediction mechanism between the multivariate statistical regression model and the Bayesian neural network model in step S104 includes: the multivariate statistical regression model establishes a linear relationship between meteorological elements and termite damage characteristic values ​​using the least squares method, outputting a preliminary prediction interval. The Bayesian neural network model refines this preliminary prediction interval, calculating the probability distribution of termite swarming paths through variational inference and Monte Carlo sampling. If the confidence interval output by the multivariate statistical regression model exceeds the preset confidence interval range, the high uncertainty mode of the Bayesian neural network model is triggered. A mixed-scale Gaussian prior is used to regularize the weights, and the prediction results are dynamically adjusted based on historical termite outbreak damage information. Finally, a fan-shaped termite risk map is displayed through the visualization interface of the intelligent termite monitoring platform, with different colors used to indicate the termite risk level.

[0079] Based on the above technical solution, the advantages of statistical models and neural networks are combined through a multi-model linkage uncertainty management mechanism, and the prediction mode is adaptively switched through a threshold triggering mechanism.

[0080] Furthermore, the feedback control mechanism for the multi-device termite extermination operation involving land, sea, and air in step S105 includes: after the extermination drone accurately applies pesticides based on the early warning results, wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs equipped with high-definition cameras re-inspect the extermination area and collect actual termite extermination data. If residual termite activity is detected, the wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs use robotic arms to deliver bait or spray powder for secondary treatment. Simultaneously, unmanned boats check the condition of the water-facing slope through monitoring equipment and provide real-time feedback on the termite infestation status of the water-facing slope to the backend software system. All termite extermination data is stored on a static file server and linked to a GIS geofence database, forming a mapping relationship between termite extermination data and termite spatial location, which is used to optimize subsequent scheduling strategies.

[0081] Based on the above technical solution, through the closed-loop control logic of extermination-review-feedback, the functions of independent equipment (such as robots and robot dogs for "review" and unmanned boats for "monitoring") are integrated into a continuously optimized dynamic system, thereby enhancing the sustainability of prevention and control effects.

[0082] Furthermore, the high-availability architecture of the termite monitoring data center includes: using sharding technology to distribute termite monitoring data for storage, supporting master-slave physical replication and off-site disaster recovery. The data access layer maps database tables to objects through an entity class library, and the business logic layer optimizes query results through caching. For unstructured data, a lightweight distributed file system is used to achieve file synchronization and load balancing, while supporting NoSQL features to store variable fields in JSON format, ensuring the stability of the backend software system when data surges.

[0083] Based on the above technical solutions, a high-availability data architecture design for termite monitoring scenarios is proposed, which combines database technologies (sharding, NoSQL) with actual business needs to solve the performance bottleneck problem when multiple devices upload data concurrently.

[0084] Furthermore, the model optimization in step S106 specifically includes an incremental learning mechanism: the background software system compares the early warning results with the actual termite damage information of the termite outbreak, and calculates the prediction error. If the prediction error exceeds a preset prediction error threshold, the incremental learning process is initiated, using newly collected termite damage information and meteorological elements as incremental samples, and locally updating the weight parameters through the reparameterization technique of the Bayesian neural network model. Simultaneously, the feature extraction layer of the convolutional neural network model adopts a dynamic convolution kernel adjustment strategy, optimizing the convolution stride and filling rules according to the seasonal activity characteristics of termites, enabling the convolutional neural network model to continuously adapt to termite morphological changes in different geographical environments.

[0085] Based on the above technical solutions, this application is aimed at a long-term deployment incremental learning framework that combines "prevention and control effect evaluation" with algorithm-level reparameterization technology to solve the problem of insufficient adaptability of traditional models in the field environment and demonstrate the sustainable evolution capability of the method.

[0086] It should be noted that this method consists of mobile patrol drones, extermination drones, wheeled or tracked unmanned vehicles, self-propelled robots, robot dogs, unmanned boats, nest finders, and fixed termite trapping lights, termite monitoring devices, plus a background software system.

[0087] It should be noted that this method is mainly used for water conservancy projects such as reservoir dams and river embankments, but it can also be used in industries such as landscaping, building construction, cultural relics and ancient buildings, energy and power, farmland crops, and forestry and grassland.

[0088] It should be noted that this method can be used in conjunction with all devices, several devices, or a single device.

[0089] It should be noted that drones, wheeled or tracked unmanned vehicles, self-propelled robots, and robot dogs are mainly used to inspect for termite damage information in the following areas: 1. Mud lines on the soil; 2. Swarming holes; 3. Termite feeding traces; 4. Termite nest indicators (such as termite mushrooms); 5. Termite nests on the ground (such as Yunnan subterranean termites and giant termites); 6. Live termites crawling on the ground (such as giant termites).

[0090] I. Main Functions of Each Equipment

[0091] (a) Mobile devices

[0092] 1. Patrol drones: rapid patrol and location tracking.

[0093] The patrol drone is equipped with a high-definition camera, infrared thermal imaging equipment, GPS locator, etc., and sends the captured video and other information to the back-end software system in real time, or analyzes and identifies termite activity traces through its own processor and uploads the identification results to the back-end software system.

[0094] (1) The patrol drone can be equipped with a high-definition camera and infrared thermal imaging equipment to conduct rapid and efficient patrols of large areas, and send the captured video and other information to the background software system in real time, or analyze and identify termite activity traces through its own processor and upload the identification results to the background software system.

[0095] (2) Through image recognition technology, signs of termite activity can be quickly detected, such as termite feeding traces, mud lines, swarming holes, termite nest indicators (termite fungus, etc.), termite ground nests (Yunnan subterranean termite, ground mound termite, etc.), and live termites on the ground (ground mound termite, etc.).

[0096] (3) Using GPS positioning technology, the termite-infested areas are accurately marked and the information is transmitted to the background software system in real time to provide accurate location information for subsequent prevention and control work.

[0097] 2. Killing Drones: Precision Application and Elimination

[0098] The extermination drone is equipped with professional spraying equipment or laser and ultrasonic emitters. It uses powder or liquid to kill termites in the affected area, or uses laser and ultrasonic emitters to kill individual termites, and uploads real-time data to the backend software system.

[0099] (1) The extermination drone can carry professional ant extermination agents to accurately apply the pesticide to the affected area and transmit the information to the back-end software system.

[0100] (2) During the termite swarming period, winged adult termites swarm in large numbers. The killing drone adjusts its flight altitude and speed to spray or laser-kill the winged adult termites in the air, killing a large number of winged adult termites. The information is then transmitted to the background software system.

[0101] (3) For areas that require a drug barrier (such as a nuclear power plant pit), large-scale rapid spraying of pesticides can be carried out using extermination drones. The information is then transmitted to the backend software system.

[0102] 3. Wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs for precise inspection and pesticide application.

[0103] Wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs are equipped with professional high-definition cameras and infrared thermal imaging equipment, GPS locators, spraying equipment, and laser emission equipment, as well as robotic arms capable of delivering bait. During inspection and processing, real-time information is fed back to the backend software system.

[0104] (1) Re-inspect the termite damage points discovered by the patrol drone, and treat them by spraying powder or placing bait as needed. Transmit the information to the back-end software system.

[0105] (2) Conduct a detailed inspection of termite damage in forests, shrublands, and small spaces such as indoors. If termites are found, treat them by spraying powder or placing bait as appropriate. Transmit the information to the backend software system.

[0106] 4. Unmanned Surface Vessels: Highly Efficient Monitoring, Lure, and Green Extermination

[0107] The unmanned surface vessel (USV) is equipped with a professional termite attractant light and high-definition termite imaging equipment. The power of the attractant light can be adjusted according to the actual usage scenario. The USV is used for reservoir dams or river embankments.

[0108] (1) During the swarming period of winged adult termites, the unmanned vessel automatically navigates to the designated location according to the instructions of the background system, turns on the termite attractant light, lures the winged adult termites to the water surface to drown, and transmits the information to the background software system.

[0109] (2) The unmanned boat equipped with the camera mainly patrols the water-facing slope of the dam (dike) to check for signs of termite activity and whether the debris carries live termites, and transmits the information to the background software system.

[0110] 5. Nest finder

[0111] The termite nest detector is used to detect underground termite nests in dams (levees). During the detection, it uploads information about the termite nest (nest volume, distance from the ground, termite trails, etc.) and coordinates to the back-end system platform.

[0112] (ii) Fixed equipment

[0113] 1. Termite trapping lamp

[0114] The termite trapping lamp integrates core components such as an attraction lamp, a weather station, a high-definition high-speed camera, a processing chip, a solar panel, and a battery. When winged adult termites swarm, weather information is transmitted to the backend system. During the swarming, winged adult termites are attracted to the lamp, and the camera captures images to identify and count the termite species. The swarming video and identification results are transmitted to the backend system in real time.

[0115] 2. Termite monitoring device

[0116] The termite monitoring device consists of bait and a casing. Inside the casing are core components such as a high-definition infrared camera, a processing chip, and a battery. Installed on the surface or underground, the device monitors termite activity in the protected area in real time. It captures and identifies termite species and numbers using the camera and uploads the information to a backend system in real time.

[0117] (III) Software System

[0118] The software system, also known as the backend software system, consists of a termite infestation big data center and a termite intelligent monitoring platform. The termite monitoring platform comprises a large intelligent termite monitoring screen, computer software, and a mobile app. It has real-time data upload, statistics, calculation, storage, and analysis functions, allowing users to view monitoring data in real time, evaluate the effectiveness of prevention and control measures, and provide prevention and control recommendations and early warnings of termite outbreaks based on historical data and front-end information.

[0119] like Figure 1 The diagram shown is the overall framework of the intelligent termite monitoring system (i.e., the multi-device collaborative termite control system for land, sea and air as described in this application).

[0120] like Figure 2The diagram shown is a topology diagram of the intelligent termite monitoring system (i.e., the multi-device collaborative termite control system for land, sea and air as described in this application).

[0121] (iv) System Algorithm

[0122] 1. Big Data Center

[0123] (1) Physical data storage

[0124] The entity data storage adopts a layered design. This system adopts the MVC development pattern, uses a code-first approach, and is based on the layered design.

[0125] ① Data Access Layer (DAL)

[0126] The system calls the database through the DAL layer instead of directly using Action or Service to operate the database. The DAL is the main database control system, which implements operations such as adding, deleting, modifying, and querying data, and feeds the operation results back to the Business Logic Layer (BLL).

[0127] ② Entity class library

[0128] Each table in the database is associated with an entity class, Model. A Model class encapsulates the operational details of a table, and its name is the same as the table's name. Entity classes are the software models of entities; one entity class describes the structure of one table.

[0129] Entity libraries are mapping objects to database tables. In the actual development of information system software, object instances are created to represent relational database tables in an object-oriented manner. This assists in the control and execution of various system functions during software development. GET and SET methods are used to map all fields in the database tables to system objects, establishing entity libraries to enable parameter transmission between different structural layers and improve code readability. Essentially, entity libraries primarily serve the presentation layer, business logic layer, and data access layer, facilitating data parameter transmission between these three layers and enhancing the simplicity of data representation.

[0130] ③ Business logic layer DLL (datalogiclayer)

[0131] Complex business logic that cannot be implemented in the Entity model can be encapsulated in a Business Logic Layer (BLL). The BLL's function is to perform logical judgments and execute operations on specific problems. After receiving user commands from the presentation layer (UI), it connects to the Data Access Layer (DAL). In the three-tier architecture, the DAL is located between the presentation layer and the data layer, acting as a bridge between them. It enables data connection and command transmission between the three layers, performs logical processing on the received data, and implements functions such as data modification, retrieval, and deletion. The processing results are then fed back to the presentation layer (UI) to realize the software functionality.

[0132] (2) NoSQL

[0133] To ensure the system is scalable and can integrate other functionalities in the future, the design of entities must support NoSQL features, and variable fields must be stored using JSON. Therefore, the database design must support non-relational features, i.e., support the JSON data type, including creating indexes for JSON data types.

[0134] (3) Full-text search

[0135] In response to the platform's requirement to establish a big data center, the database design should support full-text search. Full-text search needs to evolve from the initial string matching and simple Boolean logic search techniques to a composite technology capable of comprehensively managing massive amounts of unstructured data such as text, audio, images, and live video.

[0136] (4) High Availability Solution

[0137] As business scales and data volume increases, the pressure on the database also grows. The database layer may become a critical point and performance bottleneck in the entire system, making high availability of the data layer a crucial issue to address. Ensuring high performance and stability at the data layer requires the database to support sharding, master-slave full-database physical replication, logical replication of partial data tables, delayed backup, and off-site disaster recovery.

[0138] (5) Accessing static file servers and database tables

[0139] To address the unformatted, large-scale data storage issues mentioned in the system, a lightweight distributed file system needs to be designed and developed for file management. Functions include file storage, file synchronization, and file access (file upload and download), resolving issues related to large-capacity storage and load balancing. Mechanisms such as redundancy backup, load balancing, and linear scaling should be fully considered, with a focus on high availability and high performance. The system should support large capacity and a volume-based (or grouped) organization of storage nodes (servers).

[0140] Since the data generated by the terminal may be text data, the database design needs to support external extensions so that the database can access external data sources via SQL.

[0141] (6) GIS data services

[0142] The spatial data involved in the system has a complex structure, large volume, and query patterns that differ significantly from ordinary data, making it difficult for general DBMSs to meet the requirements. Therefore, the system design needs to provide a GIS data server for storing geometric data such as points, lines, polygons, and composite geometric objects, as well as for performing calculations on these geometric objects. Minimum system support:

[0143] ① Circle people

[0144] A common requirement in LBS services is to find all objects within a certain distance of a given central point that meet certain criteria. For example, finding all monitoring points within a 1-kilometer radius of a user and sorting them by distance.

[0145] ②Geofencing

[0146] Determine which geofences a point falls within. For example, given the location coordinates of a monitoring point and a user, the goal is to determine the area where the monitoring point and user are located from the coordinates; another example is identifying key detection areas for a certain type of ant colony or prohibited areas of a dam.

[0147] The system plans to use PostgreSQL as the database support environment, which supports data types, data access scheduling, and integration with existing data systems, etc. Figure 3 As shown.

[0148] 2. Termite Identification and Statistical Model

[0149] Establish a database of insect characteristic parameters and an image database of insect species; enhance the AI-based targeted identification technology of termites in termite monitoring devices by studying image-based targeted termite samples, analyzing whether they are autocorrelation-based, and optimizing the design of identification parameters; based on this, determine the types of insects to be observed, select appropriate observation times and locations, and use sensors to monitor upper-air meteorological environmental data; wirelessly transmit video images and environmental data to the database in real time, and further adjust the observation position and viewing angle on the software side; use image processing technology to extract target morphological feature parameters from the images, compare them with the insect image database, and identify the possible insect types.

[0150] Based on a constructed termite image database, intelligent algorithms from image recognition are used to identify termites. Convolutional Neural Networks (CNNs) are a class of feedforward neural networks that include convolutional computations and have a deep structure; they are one of the representative algorithms of deep learning. CNNs possess representation learning capabilities and can classify input information in a translation-invariant manner according to their hierarchical structure; therefore, they are also called "Shift-Invariant Artificial Neural Networks (SIANNs)." CNNs, with their unique structure of local weight sharing, have unique advantages in speech recognition and image processing.

[0151] CNNs (Convolutional Neural Networks) use artificial neurons that can respond to surrounding units within a certain coverage area, making them excellent for large-scale image processing. They consist of convolutional layers and pooling layers.

[0152] like Figure 4 The diagram shown is a schematic representation of the principle of a convolutional neural network model.

[0153] A convolutional layer contains multiple convolutional kernels, each element of which corresponds to a weight coefficient and a bias. The area covered by the convolutional kernel is called the "receptive field." Convolution is a method of integral transformation in mathematical analysis. Taking two dimensions as an example, given an image: and a filter: Then the cross-correlation is:

[0154] ;

[0155] Convolution is:

[0156] ;

[0157] In the formulas for cross-correlation and convolution operations:

[0158] X: Input image matrix, with size X ;

[0159] W: Filter (convolution kernel) matrix, with size [value missing]. ;

[0160] ij: Outputs the coordinates of elements in the feature map;

[0161] uv: Relative coordinate position within the convolution kernel.

[0162] Assume the height and width of the convolution kernel are k. h and k w Then it will be called k h × kw Convolution. In convolutional neural networks, a convolution operator, in addition to the convolution process described above, also includes adding a bias term. After one convolution, the image size decreases. The size of the convolution output feature map is calculated as follows:

[0163] ;

[0164] in:

[0165] : Output the height and width of the feature map;

[0166] : Input the height and width of the image;

[0167] The height and width of the convolution kernel.

[0168] When the kernel size is greater than 1, the output feature map size will be smaller than the input image size. If multiple convolutions are performed, the output image size will continue to decrease. To prevent the image size from shrinking after convolution, padding is usually applied around the image, such as... Figure 5 As shown.

[0169] The padded image, after passing through a process of size k, h × k w After the convolution kernel operation, the output image size is:

[0170] ;

[0171] in:

[0172] : Output the height and width of the feature map;

[0173] : Input the height and width of the image;

[0174] The height and width of the convolution kernel;

[0175] The number of layers to fill the top and bottom sides of the input image with 0s (h1, h2 correspond to the top and bottom respectively);

[0176] The number of layers to fill the left and right sides of the input image with 0 (w1 and w2 correspond to the left and right sides respectively).

[0177] During convolution calculations, equal padding is typically applied to both sides of the height or width. Then the transformed dimensions become:

[0178] ;

[0179] in:

[0180] : Output the height and width of the feature map;

[0181] : Input the height and width of the image;

[0182] The height and width of the convolution kernel;

[0183] The number of layers of 0s that are equally padded on the top and bottom of the input image;

[0184] The number of layers to fill the left and right sides of the input image with an equal amount of zeros.

[0185] Convolutional kernel sizes are typically odd numbers such as 1, 3, 5, and 7. If the padding size used is... If the image size remains unchanged after convolution, then the image size will remain unchanged.

[0186] The stride is the number of pixels the convolution kernel slides in each iteration. The image below shows a convolution process with a stride of 2, where the kernel moves 2 pixels at a time across the image. When the stride in the width and height directions are respectively... and At that time, the output feature map size is:

[0187] ;

[0188] in:

[0189] : Output the height and width of the feature map;

[0190] : Input the height and width of the image;

[0191] The height and width of the convolution kernel;

[0192] The number of layers of 0s that are equally padded on the top and bottom of the input image;

[0193] The number of layers to fill the left and right sides of the input image with an equal amount of zeros.

[0194] : The stride in the vertical direction, that is, the step size of each slide of the convolution kernel in the vertical direction;

[0195] : The stride in the horizontal direction, that is, the step size of the convolution kernel in the horizontal direction each time it slides.

[0196] The value of each point in the output feature map is derived from the value of a point of size 1 in the input image. The region is obtained by multiplying each element of the convolution kernel by the elements of the region and then adding them together. Therefore, the input image... A change in the value of each element within a region will affect the pixel value of the output point. This region is called the receptive field of the corresponding point on the output feature map. Changes in the value of each element within the receptive field will affect the value of the output point.

[0197] like Figure 6 The image shown is a pixel value map of the image within the receptive field.

[0198] Pooling layers, also known as convergence layers or subsampling layers, primarily perform feature selection, reducing the number of features and thus the number of parameters. Pooling is equivalent to dimensionality reduction in space, acting on each input feature and reducing its size. Pooling layers contain predefined pooling functions that replace the result of a single point in the feature map with the statistics of the feature maps of its neighboring regions. Using the overall statistical features of the network's outputs at a given location instead of the network's output at that location has the advantage that most of the outputs remain unchanged after a small shift in the input data. There are generally two types of pooling: average pooling and max pooling.

[0199] like Figure 7 The figure shows the average pooling and max pooling graphs.

[0200] Similar to convolution kernels, pooling windows (using...) When a pooled window slides across an image, the step size of each movement is called the stride. When the width and height movements are different, they are represented by... and This indicates that the image to be pooled can also be padded, similar to the padding method used in convolution. For example, padding can be applied before the first row. Line, fill in after the last line Row. Fill in before the first column. Columns, fill after the last column. If the column is used, then the size of the output feature map of the pooling layer is:

[0201] .

[0202] Pooling layers can not only effectively reduce the number of neurons, but also make the network invariant to small local morphological changes and have a larger receptive field.

[0203] like Figure 8 The image shown is a magnified view of the sensing field of the pooling layer.

[0204] The infrared imaging camera can collect images of the internal environment of the device, which can intuitively display the number of termites. Through the data transmission module, it can realize remote real-time monitoring of termites and environmental information at various points in the reservoir, and monitor and warn of termites at each point.

[0205] 3. Termite activity time prediction model

[0206] By combining collected ground and air termite monitoring data and meteorological data, and based on the occurrence and spread patterns of termite damage, as well as data on termite outbreak time, types of termites during peak outbreak periods, and maximum number of termites at a single site, a termite prediction model based on machine learning algorithms is constructed. When the termite damage epidemic threshold is reached, an early warning is issued, and the termite swarming paths are analyzed, and the early warning predicts the termite outbreak time and the maximum number of termites at a single site.

[0207] Intelligent monitoring and analysis system such as Figure 9 As shown.

[0208] (1) Multivariate statistical regression model

[0209] A quantitative relationship between meteorological factors and termite damage parameters was established using a multiple regression model.

[0210] like Figure 10 The figure shown is a quantitative relationship diagram between meteorological elements and termite damage parameters.

[0211] The relationship between the two is as follows:

[0212] Y = XB + E;

[0213] in:

[0214] Y is a The column vector represents the observed values ​​of the dependent variable;

[0215] X is a The matrix represents the observation matrix of the independent variables, where the first column is usually 1 (intercept term);

[0216] B is a The column vector represents the regression coefficients (including the intercept term).

[0217] E is a The column vector represents the random error term.

[0218] In summary, the relationship between the two can be expressed in matrix form as follows:

[0219] ;

[0220] in:

[0221] Represents the observed values ​​of a specific dependent variable;

[0222] Represents the observed value of the independent variable;

[0223] Represents the regression coefficient;

[0224] This represents random error.

[0225] Model assumptions:

[0226] A) The expected value of the residual term is O: Or rather ;

[0227] B) The explanatory variables in each row have the same covariance structure:

[0228] ;

[0229] C) There is no correlation between the rows of explanatory variables:

[0230] ;

[0231] Estimate parameter matrix B using the least squares method:

[0232] ;

[0233] Hypothesis testing was performed on the model parameters, including Wilk's Lambda Test, Roy's Test, Pillai's Test, and Lawley-Hotelling Test.

[0234] Given a new observation Substitute into the parameter matrix Predict its corresponding output Focus on interval predictions, including:

[0235] Expected value Confidence interval:

[0236] ;

[0237] random variable Prediction interval:

[0238] .

[0239] (2) Bayesian neural network model

[0240] The timing and outbreak patterns of winged reproductive termites may be somewhat uncertain, but Bayesian neural network algorithms can predict them relatively well. By constructing a Bayesian neural network model between meteorological parameters and termite outbreak parameters (as shown in the figure below), and through model training, the characteristics of termite outbreaks under given time and meteorological conditions can be gradually predicted.

[0241] like Figure 11 The diagram shown is a schematic of the Bayesian neural network prediction model for termite outbreaks.

[0242] Bayesian neural networks regularize by introducing uncertainty into the weights of the neural network, essentially ensemble an infinite number of neural networks based on a certain weight distribution for prediction. According to Bayesian theory, if the prior distribution has the following form:

[0243] ;

[0244] The posterior distribution can then be expressed as:

[0245] ;

[0246] in:

[0247] Represents the observation dataset;

[0248] Indicates weight;

[0249] This represents the posterior distribution of the weights, i.e., the conditional distribution of the weights given the data.

[0250] This represents the likelihood function. That is, given a set of weights... In the case of observation The possibility;

[0251] This represents the prior distribution of the weights;

[0252] Represents marginal likelihood, i.e., data The absolute probability under the model.

[0253] Using variational methods, one can use a set of parameters... control Distribution To approximate the true posterior :

[0254] ;

[0255] In objective function form, it would be:

[0256] ;

[0257] in:

[0258] Represents variational parameters;

[0259] Describes the variational posterior distribution, given by the parameters Control, used to approximate the true posterior ;

[0260] The KL divergence represents the difference between the variational distribution and the prior distribution, and is used to measure the difference between the two distributions.

[0261] for In the form of, a mixed-scale Gaussian prior is given:

[0262] ;

[0263] in:

[0264] : Represents the weight parameter vector of the model, representing the set of all parameters that need to be learned;

[0265] : Represents the j-th individual weight parameter;

[0266] : Represents the mixing coefficient, with a value range of [0,1], which controls the weight ratio of the two Gaussian components;

[0267] The weight of the second Gaussian component ensures a total probability of 1.

[0268] To represent the variance of two Gaussian components, it is usually set to... Achieving "hybrid scale";

[0269] This indicates that the mean is 0 and the variance is 0. The Gaussian distribution.

[0270] Then, the Monte Carlo approximation of the above objective function is obtained by using the reparameterization operation:

[0271] ;

[0272] in:

[0273] The objective function that needs to be optimized;

[0274] The entire observation (training) dataset;

[0275] Variational parameters. They define the approximate posterior distribution. The shape;

[0276] The number of Monte Carlo samplings. That is, the number of samples drawn from the variational distribution to approximate the expected value;

[0277] : The i-th variational distribution The model weight vector obtained by sampling in the middle;

[0278] Variational posterior distribution. This is a distribution consisting of parameters. Defined posterior distributions for approximating the true but computationally difficult ones. A simple distribution;

[0279] Prior distribution of weights;

[0280] Likelihood function. Represents the likelihood function given weights. In the case of observed data The probability of.

[0281] Bayesian mini-batch gradient descent:

[0282] ;

[0283] get

[0284] ;

[0285] in:

[0286] i: The i-th mini-batch of data;

[0287] KL divergence. It measures the difference between the variational posterior distribution q and the prior distribution P, acting as a regularization term to prevent the model from overfitting the training data.

[0288] Regarding distribution Expectations;

[0289] M: The total number of mini-batches in the dataset, i.e., M = mini-batch size / total data volume;

[0290] : The weight coefficient corresponding to the i-th mini-batch.

[0291] The above describes the global uncertainty introduced into the weights of a neural network. Global uncertainty means that all parameters must be sampled globally during inference computation, which is costly. Therefore, local reparameterization is proposed:

[0292] Assuming all parameters follow an independent Gaussian distribution, the results of matrix multiplication will also follow an independent Gaussian distribution. In other words, for... If there is:

[0293] ;

[0294] Then for Y, we would have:

[0295] ;

[0296] in:

[0297] Weight The variational posterior distribution;

[0298] Gaussian distribution (normal distribution);

[0299] Weight The mean of the variational posterior distribution;

[0300] Weight The variance of the variational posterior distribution;

[0301] Given input Under the condition of output The probability distribution;

[0302] Output The mean of the distribution. Its calculation formula is:

[0303] ;

[0304] Output The variance of the distribution. Its calculation formula is:

[0305] .

[0306] Therefore, it is unnecessary to sample the parameter W every time. We can directly calculate the mean and variance of the result Y, sample it, and then backpropagate it to W. In this way, the sampling performed in each calculation is a local sampling of the corresponding data points, which is called local reparameterization.

[0307] (3) Study on the swarming model of termite activity during the outbreak period

[0308] 1): AI Image Enhancement + Intelligent Recognition

[0309] ① The Retinex-Net image enhancement technique using a convolutional neural network is used for image preprocessing. Based on this, deep learning algorithms are used to train and recognize the images.

[0310] ② Image preprocessing is performed using the AlexNet image enhancement technique based on convolutional neural networks, specifically implemented through the tf.image or tf.keras.imageGenerator toolkit. Based on this, image recognition is trained.

[0311] 2): Raw data / fuzzy data + intelligent recognition

[0312] The original termite images were blurred using Tenengrad and Laplacian gradient algorithms to obtain termite images of varying resolutions, which were then used as a database. Deep learning algorithms were then combined to perform termite identification.

[0313] 3): Dynamic swarming ant images + intelligent recognition

[0314] The images of dynamically swarming flying ants are directly used as part of the database for training and prediction.

[0315] 4): High-definition camera images + intelligent recognition

[0316] High-resolution camera images are used as prediction samples, and predictions are made using deep learning algorithms.

[0317] The above four methods are used for multi-source data fusion.

[0318] 5): Main Principles

[0319] A) First, based on the termite monitoring device and termite trapping lamp, count the number of termites at each monitoring point and the azimuth of the monitoring point.

[0320] B) The meteorological information collected by the weather station was analyzed in combination with historical termite outbreak times and meteorological information, and the meteorological information and outbreak times were consistent.

[0321] C) Use self-organizing neural networks and other methods to cluster the termite population at each location and define risk levels, ensuring that the number of risk levels does not exceed 5.

[0322] D) Then, draw a vector pointing from the center of the downstream slope of the dam to the perpendicular bisector of the line connecting each monitoring point.

[0323] E) Draw a circle with a certain radius outward from the center of the downstream slope of the dam.

[0324] F) Assign the grade of each monitoring point to the fan-shaped area enclosed by the adjacent perpendicular line and the arc.

[0325] G) Fill the above sector area with the corresponding color according to the risk level to finally determine the possible location of termite intrusion.

[0326] like Figure 12 The image shown is a schematic diagram illustrating the location of the termite invasion.

[0327] Model training and model scheduling diagram as follows Figure 13 As shown.

[0328] II. Function Demonstration

[0329] (a) Schematic diagram

[0330] like Figure 14 The diagram shown is a functional demonstration of this application.

[0331] (II) Standard Levee

[0332] like Figure 15 The diagram shown is a standardized cross-sectional view of a levee.

[0333] The drones inspected the entire prevention and control area, focusing on areas without tree cover such as the top of the dike, the water-facing slope, and the back slope, and transmitted the inspection results to the back-end system.

[0334] Wheeled or tracked unmanned vehicles, self-propelled robots, and robot dogs are used to conduct key re-inspections of areas where drones have detected ant infestations. They also conduct comprehensive patrols of the dike's wave-breaking forests, suitable forests, and protective forests, and transmit the patrol results to the back-end system. Ant infestations are then dealt with according to instructions from the back-end system.

[0335] The unmanned boat focuses on inspecting the water-facing slope of the dike and the debris, and transmits the information to the back-end system. When the winged adult termites swarm, the unmanned boat travels to the designated location according to the instructions of the back-end system, turns on the attraction light, and lures the winged adult termites to the water surface to drown.

[0336] Once a nest finds an ant nest during a patrol, the information is transmitted to the backend system. Wheeled or tracked unmanned vehicles, self-propelled robots, and robot dogs then respond to the system's instructions and go to the site to deal with the ant infestation.

[0337] During the swarming period of winged adult termites, extermination drones carrying pesticides are used to spray and kill the winged adult termites in the air.

[0338] Termite trapping and monitoring devices can be installed based on the actual site conditions to monitor termites and transmit the information to the back-end system in real time. Equipment for dike safety monitoring and ecological monitoring can also be connected to the back-end system for timely handling when termite-related information is detected.

[0339] (III) Dam

[0340] like Figure 16 The image shown is a schematic diagram of the dam.

[0341] The drones inspected the entire prevention and control area, focusing on unobstructed areas such as the dam crest, the upstream slope, and the downstream slope, and transmitted the inspection results to the back-end system.

[0342] Wheeled or tracked unmanned vehicles, self-propelled robots, and robot dogs should be used to re-examine areas where drones have detected ant infestations. A comprehensive inspection of the forests and green belts on both sides of the dam should be conducted. In areas where subterranean termites and red-tailed termites are distributed, indoor areas prone to ant infestations should also be inspected. The inspection results should be transmitted to the back-end system, and the ant infestation should be handled according to the instructions of the back-end system.

[0343] Once a nest finds an ant nest during a patrol, the information is transmitted to the backend system. Wheeled or tracked unmanned vehicles, self-propelled robots, and robot dogs can then go to the site to deal with the ant infestation according to the system's instructions.

[0344] The unmanned boat focuses on inspecting the upstream slope of the dam and the debris, and transmits the information to the back-end system. When the winged adult termites swarm, the unmanned boat travels to the designated location according to the instructions of the back-end system, turns on the attraction light, and lures the winged adult termites to the water surface to drown.

[0345] Termite trapping lights are installed at a certain distance from the dam. During the swarming season of winged adult termites, cameras automatically monitor for swarming and transmit the information to the backend system. When swarming of winged adult termites is detected, the trapping lights and fans or high-voltage grids, high-speed cameras, and other components are automatically activated, and the species and number of swarming termites identified are uploaded to the backend system.

[0346] Termite monitoring devices are installed on the upstream and downstream slopes of the dam and in the surrounding core area to monitor underground termites in real time. When termites are detected entering the station, the species and number of invading termites are transmitted to the back-end system.

[0347] Based on the actual situation on site, relevant information from safety monitoring, ecological monitoring, and other equipment can be connected to the back-end system, and termite-related information can be processed in a timely manner.

[0348] The backend system stores the received information in the termite infestation big data center and analyzes and processes the relevant data. When termites are detected, it promptly issues an alarm and provides disposal suggestions based on the actual situation. The system combines historical and real-time data for comparative analysis to predict termite activity (outbreak) trends within a certain period and provides corresponding treatment measures.

[0349] It should be noted that the terms "first," "second," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. "A plurality" or "several" indicates at least two. Unless otherwise stated, terms such as "front," "back," "left," "right," "lower," and / or "upper" are for illustrative purposes only and are not limited to a location or spatial orientation. Terms such as "comprising" or "including" indicate that the elements or objects preceding "comprising" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0350] The singular forms “a,” “the,” and “the” used in this application specification and appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0351] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A multi-dimensional sensing, land, and air IoT-based precision termite control system, characterized in that, The system includes a mobile device, a fixed device, and a backend software system; the mobile device and the fixed device collect termite damage information and upload it to the backend software system; the backend software system performs statistical analysis based on the termite damage information to obtain termite control measures and termite outbreak early warning information. At least one of the mobile devices is equipped with a termite extermination device, and at least one of the mobile devices controls the termite extermination device to exterminate termites based on the termite prevention and control measures. The mobile device collects termite extermination data and uploads it to the background software system. The background software system evaluates the prevention and control effect based on the termite extermination data and obtains the prevention and control effect evaluation result.

2. The multi-dimensional sensing land, water, and air IoT precision termite control system according to claim 1, characterized in that, The mobile device includes at least one of the following: patrol drone, extermination drone, self-propelled robot, robot dog, unmanned boat, nest explorer, wheeled or tracked unmanned vehicle; The patrol drone is equipped with a camera, an infrared thermal imaging device, and a GPS locator. The patrol drone captures video information through the camera and the infrared thermal imaging device, and obtains the location of termite damage through the GPS locator. The patrol drone sends the video information and the location of termite damage to the backend software system in real time, or the patrol drone analyzes and identifies termite damage information through its own processor and uploads the termite damage information identification results to the backend software system. The termite extermination drone is equipped with a termite extermination device. The termite extermination device on the drone kills termites by spraying powder, bait or liquid, or by emitting laser or ultrasonic waves, and uploads the termite extermination data to the background software system. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog is equipped with a second camera, a second infrared thermal imaging device, and a second GPS locator. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog captures video information via the second camera and the second infrared thermal imaging device. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog uses GPS... The locator 2 obtains the location of termite infestation. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog is equipped with a termite extermination device or a laser emitting device or an ultrasonic emitting device. The termite extermination device carried by the wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog includes at least one of the following: a powder, bait, or liquid applicator; a robotic arm capable of dispensing bait; a laser emitting device; or an ultrasonic emitting device. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog re-inspects the termite infestation location discovered by the patrol drone and performs termite extermination based on the termite inspection data. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog also collects termite infestation information in confined spaces. The wheeled or tracked unmanned vehicle, self-propelled robot, or robot dog uploads the termite infestation information and termite extermination data to the background software system. The unmanned vessel is equipped with a camera to collect information on termite damage. The unmanned vessel is also equipped with a termite extermination device, which is a termite attractant lamp. The unmanned vessel uploads the termite damage information and termite extermination data to the background software system. The nest detector is used to detect underground termite nests, and it uploads termite damage information to the background software system.

3. A multi-dimensional sensing land, water, and air IoT-based precision termite control system according to claim 1 or 2, characterized in that, The fixed equipment includes at least one of the following: termite trapping lamp, termite monitoring device; The termite trapping lamp integrates a termite attracting lamp, a weather station, and a camera. The weather station transmits weather information to the background software system. When termites swarm, the termite attracting lamp attracts winged adult termites into the lamp. The camera identifies and counts termite species. The termite trapping lamp uploads termite damage information and termite extermination data to the background software system. The termite monitoring device includes a shell, bait to attract termites, and camera three. The termite monitoring device is installed on the surface or underground. Camera three identifies termite species and obtains termite damage information. The termite monitoring device uploads termite damage information and termite extermination data to the background software system.

4. A multi-dimensional sensing method for precise termite control via the Internet of Things (IoT) covering land, water, and air, characterized in that... The method includes: S101 collects termite damage information and meteorological elements through multiple devices on land, sea, and air, and marks termite-infested areas using GPS positioning. S102, uploads termite damage information, meteorological elements and termite-infested areas to the backend software system. The backend software system includes a termite infestation big data center and a termite intelligent monitoring platform. The termite infestation big data center uses a layered architecture to store data. The termite infestation big data center includes a data access layer, an entity class library and a business logic layer, and supports NoSQL features to store variable fields in JSON format. S103, the termite intelligent monitoring platform uses a convolutional neural network model to identify collected termite damage information to obtain identification results. The convolutional neural network model includes convolutional layers and pooling layers. The convolutional layers calculate termite damage feature values ​​by sliding the convolutional kernel over the input termite damage information, and the pooling layers perform feature dimensionality reduction on the termite damage feature values ​​through average pooling or max pooling. The identification results are compared with a termite image database to determine the termite species and quantity. S104, the termite intelligent monitoring platform combines meteorological elements, historical termite outbreak information, and termite-infested areas. It establishes a quantitative relationship between meteorological elements and termite damage characteristic values ​​through a multivariate statistical regression model, and uses a Bayesian neural network model to predict termite swarming paths, outbreak times, and scale. When the termite damage characteristic values ​​exceed the preset damage threshold, the termite intelligent monitoring platform issues an early warning and obtains the warning results. S105, the background software system dispatches termite extermination devices carried by multiple equipment on land, sea and air according to the early warning results, so as to carry out termite extermination in the early warning termite-infested areas and obtain termite extermination data. S106, the background software system optimizes the convolutional neural network model in step S103 and the Bayesian neural network model in step S104 based on the termite extermination data.

5. The multi-dimensional sensing method for precise termite control via land, water, and air IoT as described in claim 4, characterized in that, The multi-device system in step S101 includes at least one of the following: patrol drone, extermination drone, self-propelled robot, robot dog, unmanned boat, nest finder, wheeled or tracked unmanned vehicle. Among them, patrol drones prioritize patrolling large areas without tree cover, while wheeled or tracked unmanned vehicles, self-propelled robots, and robotic dogs conduct secondary inspections of forests, shrublands, and small indoor spaces. Unmanned boats automatically navigate to designated locations according to instructions from the backend system, then activate termite-attracting lights, and nest detectors simultaneously probe underground termite nest structures. The multiple devices on land, sea, and air achieve location synchronization through GIS data services. These GIS data services support tagging and geofencing functions, enabling the delineation and adjustment of dynamic patrol ranges, and real-time adjustment of the scheduling priority of multiple devices on land, sea, and air based on the termite risk level.

6. The multi-dimensional sensing method for precise termite control via land, water, and air IoT as described in claim 4, characterized in that, The optimization of the convolutional neural network model in step S103 includes local reparameterization: for termite image recognition tasks, a mixed-scale Gaussian prior distribution is introduced into the weights of the convolutional layers, and the posterior distribution is approximated through variational inference; during the training process of the convolutional neural network model, local reparameterization is used to replace global parameter sampling, and the mean and variance of the output termite damage feature values ​​are directly calculated to reduce computational costs; at the same time, the pooling layer captures the local invariance of termite morphology through the receptive field mechanism, and combines image enhancement technology to preprocess blurred termite images, thereby improving the robustness of the convolutional neural network model in recognizing subtle features including termite swarming holes and mud lines.

7. The multi-dimensional sensing method for precise termite control via land, water, and air IoT as described in claim 4, characterized in that, The linkage prediction mechanism between the multivariate statistical regression model and the Bayesian neural network model in step S104 includes: the multivariate statistical regression model establishes a linear relationship between meteorological elements and termite damage characteristic values ​​using the least squares method, and outputs a preliminary prediction interval; the Bayesian neural network model refines the preliminary prediction interval by calculating the probability distribution of termite swarming paths through variational inference and Monte Carlo sampling; if the confidence interval output by the multivariate statistical regression model exceeds the preset confidence interval range, the high uncertainty mode of the Bayesian neural network model is triggered, the weights are regularized using a mixed-scale Gaussian prior, and the prediction results are dynamically adjusted in conjunction with historical termite outbreak damage information; finally, a fan-shaped termite risk map is displayed through the visualization interface of the termite intelligent monitoring platform, with different colors used to mark the termite risk level.

8. The multi-dimensional sensing method for precise termite control via land, water, and air IoT as described in claim 5, characterized in that, The feedback control mechanism for the multi-device termite extermination operation involving land, sea, and air in step S105 includes: after the extermination drone accurately applies pesticides based on the early warning results, wheeled or tracked unmanned vehicles, self-propelled robots, or robot dogs carrying high-definition cameras re-inspect the extermination area and collect actual termite extermination data; if residual termite activity is detected, the wheeled or tracked unmanned vehicles, self-propelled robots, or robot dogs use robotic arms to deliver bait or spray powder for secondary treatment; simultaneously, unmanned boats use imaging equipment to inspect the water-facing slope of the dam and provide real-time feedback on the inspection results to the background software system; all termite extermination data is stored through a static file server and linked to a GIS geofence database to form a mapping relationship between termite extermination data and termite spatial location, which is used to optimize subsequent scheduling strategies.

9. The multi-dimensional sensing method for precise termite control via land, water, and air IoT according to claim 4, characterized in that, The high-availability architecture of the termite infestation big data center includes: using sharding technology to distribute termite monitoring data for storage, supporting physical replication between master and slave databases and off-site disaster recovery; the data access layer maps database tables to objects through entity class libraries, and the business logic layer optimizes query results through caching; for unstructured data, a lightweight distributed file system is used to achieve file synchronization and load balancing, while supporting NoSQL features to store variable fields in JSON format, ensuring the stability of the backend software system when data surges.

10. A multi-dimensional sensing method for precise termite control via land, water, and air IoT as described in claim 4, characterized in that, The model optimization in step S106 specifically includes an incremental learning mechanism: the background software system compares the early warning results with the actual termite damage information of the termite outbreak, and calculates the prediction error; if the prediction error exceeds the preset prediction error threshold, the incremental learning process is started, and the newly collected termite damage information and meteorological elements are used as incremental samples, and the weight parameters are locally updated through the reparameterization technology of the Bayesian neural network model; at the same time, the feature extraction layer of the convolutional neural network model adopts a dynamic convolution kernel adjustment strategy, and optimizes the convolution step size and filling rules according to the seasonal activity characteristics of termites, so that the convolutional neural network model can continuously adapt to the morphological changes of termites in different geographical environments.

Images