Termite activity killing method and system based on image recognition and storage medium

Abstract

The invention relates to the technical field of termite prevention and control, and discloses a termite activity killing method and system based on image recognition and a storage medium, and the method comprises the steps: obtaining a time sequence image frame, calculating and generating an instantaneous activity vector field, and aggregating and updating a probabilistic pheromone potential energy diagram; topological structure extraction is carried out on the potential energy diagram to identify paths and nodes, and function classification is carried out on the nodes in combination with vector field divergence; generating a self-adaptive killing decision instruction based on comprehensive analysis of a group macroscopic state, a causal relationship between nodes and path network toughness; and the bait throwing execution unit is driven to execute throwing, and closed-loop confirmation is carried out on the killing effect by continuously monitoring key node parameters after throwing. According to the invention, through deep quantitative analysis of the behavior pattern of the termite group and the social network attribute, the accuracy of the killing strategy and the automatic verification of the self-adaptive decision and effect are realized, and the intelligent level and reliability of termite prevention and control are improved.

Description

Technical Field

[0001] This invention relates to the field of termite control technology, specifically to a method, system, and storage medium for termite activity extermination based on image recognition. Background Technology

[0002] As a social pest, termites are characterized by their covert and destructive activities, posing a serious threat to buildings, garden trees, and other structures. Existing termite control technologies mainly rely on chemical barrier methods and bait monitoring systems. In practical applications, the effectiveness of these traditional methods largely depends on the accurate assessment of termite activity and the precise selection of the timing and location for pesticide application.

[0003] However, current monitoring methods are often passive and discontinuous. For example, bait monitoring systems require regular manual inspections to confirm whether termites have invaded the monitoring device, resulting in long response cycles and potentially missing the optimal intervention opportunity. At the same time, the judgment of key information such as termite activity paths and nest entrances relies heavily on the professional experience of pest control personnel, lacking objective and quantitative data support, which makes the decision-making process subjective and uncertain. After baiting, the evaluation of the extermination effect also faces challenges, requiring a long waiting period before manual inspection is conducted again. It is impossible to achieve real-time, automated feedback and confirmation of the control effect. Therefore, existing technologies generally lack an effective means to conduct in-depth and dynamic analysis of termite colony behavior and make intelligent, adaptive decisions and closed-loop verification based on this analysis. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a termite control method, system, and storage medium based on image recognition. This solves the problems that existing termite control technologies rely heavily on static physical monitoring or manual judgment, lack the ability to conduct in-depth and quantitative analysis of termite colony behavior, cannot make adaptive strategy decisions based on the dynamic changes of termite social networks, and lack an objective and automated closed-loop verification mechanism for control effectiveness.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method, system, and storage medium for termite extermination based on image recognition.

[0006] The first aspect of this invention provides a termite extermination method based on image recognition. By continuously visually monitoring termite colonies, it achieves the analysis from individual microscopic movements to macroscopic behavioral patterns of the colony. Based on a comprehensive assessment of the social network structure, function, causal chain, and resilience of the colony, it formulates and executes a precise and adaptive extermination strategy, ultimately achieving closed-loop confirmation of the effect.

[0007] In one embodiment, the termite activity extermination method based on image recognition includes the following steps: First, acquiring time-series image frames of the monitoring area, and calculating and generating an instantaneous activity vector field representing the movement state of all individual termites in the area at a specific moment based on the time-series image frames.

[0008] Subsequently, based on the evolution of the instantaneous activity vector field over time, a probabilistic pheromone potential energy map is generated and dynamically updated. The probabilistic pheromone potential energy map reflects the historical cumulative intensity and distribution of termite activity paths. The update process considers the decay effect of pheromones over time and the real-time contribution of current activities to the potential energy.

[0009] Next, topological structure extraction is performed on the probabilistic pheromone potential energy map to identify paths and nodes in the termite network. Combined with the instantaneous activity vector field, the nodes are functionally classified. Specifically, the functional classification involves calculating the divergence of the instantaneous activity vector field in the neighborhood of each node, and classifying the node as a convergence point (C-class node) or a divergence point (D-class node) based on the sign and magnitude of the divergence.

[0010] Subsequently, based on a comprehensive analysis of the instantaneous activity vector field, the probabilistic pheromone potential energy map, and the functional classification of the nodes, an extermination decision command is generated. This comprehensive analysis includes three parallel evaluation dimensions: Assessing the macroscopic state of the group: By calculating the coherence of all vector directions in the instantaneous activity vector field, a global ordered parameter is obtained. The global ordered parameter characterizes the synchronicity and directional consistency of the group's motion. When the global ordered parameter reaches a preset threshold, it indicates that the group is in a state of coordinated motion, and one of the conditions for intervention is met at this time.

[0011] Quantifying causal relationships between nodes: Extract the divergence time series of the classified C-type nodes and D-type nodes, and calculate the time-delay cross-correlation between C-type nodes and D-type nodes. By analyzing the time-delay cross-correlation function, confirm whether there is a stable recruitment-response causal chain from D-type nodes to C-type nodes, which is used to verify the effectiveness of the critical path.

[0012] Determining the resilience of the path network: After detecting an external disturbance event, by monitoring the changes in the probabilistic pheromone potential energy diagram, the formation time required for the blocked path to recover or for a new detour path to form is calculated. Based on the length of the formation time, the network is determined to be either a highly resilient network capable of rapid self-repair or a fragile network prone to collapse.

[0013] Based on the results of the above comprehensive analysis, specific extermination decision instructions are generated. For example, strategy selection is made according to the resilience determination results of the path network: if it is determined to be a fragile network, instructions are generated to execute single-point, full-volume deployment on a single critical path; if it is determined to be a highly resilient network, instructions are generated to execute multi-point, micro-volume, and gradual deployment on multiple related paths.

[0014] Finally, the bait delivery execution unit is driven to perform the delivery, and after the delivery, the extermination effect is confirmed by continuously monitoring the divergence time series of the C-type nodes. When the divergence value of all C-type nodes is reduced to below the preset extermination confirmation threshold and continues for a preset time, the termite activity extermination system based on image recognition generates an extermination confirmation signal, indicating that the activity at the nest entrance has been suppressed.

[0015] A second aspect of the present invention provides a termite control system based on image recognition, comprising an image acquisition unit, a bait delivery execution unit, and a data processing unit. The image acquisition unit is used to acquire time-series image frames of a monitoring area, the bait delivery execution unit is used to perform a delivery action at a designated location, and the data processing unit is connected to the image acquisition unit and the bait delivery execution unit respectively, and is configured to execute a termite control method based on image recognition.

[0016] A third aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a termite extermination method based on image recognition.

[0017] This invention provides a method, system, and storage medium for termite control based on image recognition. It offers the following advantages: 1. This invention generates an instantaneous activity vector field by calculation, and based on this, aggregates and updates a probabilistic pheromone potential energy map. Combined with the divergence analysis of the vector field, it can extract the topological structure and classify the node functions of the termite activity network in the monitoring area. This method quantifies the collective behavior into a structured network model, accurately identifying key functional units such as nest entrances, food sources, and main paths. Compared with traditional methods that rely on macroscopic activity signs, it improves the objectivity and accuracy of target identification.

[0018] 2. This invention constructs a multi-dimensional decision-making model by comprehensively analyzing the macro-state of the colony, the causal relationships between nodes, and the resilience of the path network. This model not only confirms the synergy of the colony's behavior and the effectiveness of the path, but also assesses the robustness of the termite social network. Based on the assessment results, it adaptively selects either a single-point intensive or multi-point gradual deployment strategy. This mechanism enables the extermination plan to be dynamically adjusted for termite colonies with different characteristics, improving the targeting and effectiveness of the strategy.

[0019] 3. This invention establishes an automated closed-loop feedback and effect confirmation mechanism by continuously monitoring the divergence time series of key functional nodes after baiting. By quantifying the changes in the activity intensity of key nodes, it objectively determines whether the extermination operation is successful, replacing the traditional manual observation and delayed judgment. This achieves real-time, quantitative, and automated verification of the extermination effect, ensuring the integrity and reliability of the entire prevention and control process. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is an overall flowchart of the method of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See attached document Figure 1 The present invention provides a termite control system based on image recognition, including an image acquisition unit, a data processing unit, a bait delivery execution unit, and a communication and network module.

[0023] In one specific embodiment, the image acquisition unit functions to provide the data processing unit with raw video stream data containing termite activity information required for subsequent analysis.

[0024] The image acquisition unit may include one or more image sensors, which may be industrial-grade charge-coupled devices or complementary metal-oxide-semiconductor sensors, to ensure image clarity under different lighting conditions.

[0025] To capture the rapid, minute movements of termites, the image sensor's frame rate is configured to be no less than a preset frame rate threshold. The resolution of the image sensor is configured to be no less than a preset resolution threshold to ensure the accuracy of identifying the morphology and movement of a single termite.

[0026] Image sensors are installed in the monitoring area, such as the entrance to a termite nest or a predetermined location at a key node in an ant trail. The installation height and tilt angle of the image sensors are adjusted to ensure that the field of view of the image sensors can completely cover the target area to be monitored.

[0027] When the image acquisition unit includes multiple image sensors, the termite activity extermination system based on image recognition also includes a synchronization controller. The synchronization controller is used to provide a unified clock signal or trigger signal to all image sensors to ensure that all image sensors acquire images in a strictly synchronized manner, such as frame-synchronized acquisition, to generate time-aligned multi-view image data.

[0028] The image acquisition unit is wired to the data processing unit via a data interface. The image acquisition unit processes the acquired continuous time-series image frames. Real-time transmission to the data processing unit, where The two-dimensional spatial coordinates of the image, For timestamps.

[0029] In one specific embodiment, the data processing unit functions to receive time-series image frames from the image acquisition unit. It executes a series of preset algorithm processes to generate extermination decision instructions, and sends the extermination decision instructions to the decoy deployment execution unit.

[0030] The physical implementation of the data processing unit can be an industrial control computer, an edge computing server, or an embedded system containing a specific computing chip.

[0031] The data processing unit includes one or more central processing units, one or more graphics processing units, and non-volatile memory and volatile memory.

[0032] The non-volatile memory is used to store the operating system, program code for executing the method of the present invention, and historical data for analysis and feedback.

[0033] The data processing unit is equipped with a first data interface for communication with the image acquisition unit. This first data interface can be one or more gigabit Ethernet interfaces, USB 3.0 interfaces, or Camera Link interfaces, used for high-speed reception of time-series image frames. .

[0034] The data processing unit is also equipped with a second data interface for communicating with the decoy deployment execution unit. The second data interface can be one or more serial communication interfaces, Modbus communication interfaces, or digital / analog I / O interfaces for sending extermination decision commands.

[0035] The data processing unit stores data in non-volatile memory and volatile memory and executes a series of functional modules by the central processing unit and the graphics processing unit. These modules are specifically used to implement the analysis and decision-making process of the present invention.

[0036] In one embodiment, the data processing unit includes: The image preprocessing module is used to process the received time-series image frames. Perform denoising, background modeling, and distortion correction operations; The instantaneous activity vector field generation module is used to calculate and generate instantaneous activity vector fields based on preprocessed image frames. ; The group macroscopic state assessment module is used to calculate and generate instantaneous activity vector fields. Globally ordered parameters ; The PPM aggregation module is used to generate instantaneous activity vector fields. Over time The evolution of the probabilistic pheromone potential energy map is aggregated and dynamically updated. ; The path function parsing module is used to analyze the potential energy map of free pheromones. Topological structure extraction is performed, and this is combined with the generation of instantaneous active vector fields. Calculate divergence This is used to determine the function of a node (C class, D class). The causal chain quantization module is used to extract the divergence time series of C-type and D-type nodes. and And calculate time-delay cross-correlation. ; The network resilience assessment module is used to assess network resilience in response to detected disturbance events. Afterwards, quantification Network strain and resilience recovery time Used to determine network resilience ( or ); The strategy decision-making module integrates the outputs of the module group macro-state assessment module, path function analysis module, causal chain quantification module, and network resilience assessment module, and generates the final deployment strategy based on the preset decision logic. The instruction generation module is used to convert the strategy output by the strategy decision module into physical instructions that can be recognized and executed by the decoy deployment execution unit; The closed-loop feedback module is used to continuously monitor the C-type node divergence output by the path function parsing module after the bait delivery execution unit has performed the delivery. This is used to confirm the effectiveness of the extermination process.

[0037] In one specific embodiment, the bait delivery execution unit is electrically connected to the data processing unit. The function of the bait delivery execution unit is to receive the extermination decision instruction from the data processing unit and accurately deliver a predetermined dose of bait at a designated physical coordinate point within the monitoring area.

[0038] The bait delivery execution unit includes a multi-axis motion mechanism, a bait dispensing module, and a local controller.

[0039] In one embodiment, the multi-axis motion mechanism is a Cartesian coordinate robot. The multi-axis motion mechanism has at least two translational degrees of freedom (X-axis and Y-axis), and its motion range covers the entire monitoring area of ​​the image acquisition unit. The multi-axis motion mechanism is driven by a servo motor or a stepper motor and is equipped with a position encoder to ensure that the positioning accuracy of the multi-axis motion mechanism is not lower than a preset accuracy threshold.

[0040] In another embodiment, the multi-axis motion mechanism is a multi-joint robotic arm used to perform delivery actions within a monitoring area with complex obstacles or non-planar structures.

[0041] The bait dispensing module is mounted on the end effector of the multi-axis motion mechanism. The bait dispensing module includes a bait storage bin, a precision metering pump, and a dispensing nozzle.

[0042] The bait storage chamber is used to hold slow-release bait with a preset physical form to be deployed. The precision metering pump is connected to the bait storage chamber to precisely control the extrusion volume of the bait. The dispensing nozzle is connected to the outlet of the precision metering pump to apply the bait to the target surface.

[0043] The local controller is equipped with a data interface for communicating with the second data interface of the data processing unit.

[0044] The local controller is used to receive extermination decision commands, which at least include the target deployment coordinates. and target delivery volume Target delivery coordinates The physical coordinates are determined by the data processing unit based on the analysis of the PPM aggregation module.

[0045] Upon receiving the extermination decision command, the local controller performs the following operations: Parse instructions; Generate motion control signals for the multi-axis motion mechanism to drive the dispensing nozzle to move to the target dispensing coordinates. This also generates a drive signal for the precision metering pump, enabling the precision metering pump to accurately extrude a volume of... The bait.

[0046] After the deployment action is completed, the local controller sends a status confirmation signal to the data processing unit through the data interface, which triggers the closed-loop feedback module in the data processing unit to enter the subsequent monitoring and confirmation stage.

[0047] In one specific embodiment, the communication and network module may be integrated into the data processing unit, or electrically connected to the data processing unit as a separate physical unit.

[0048] The function of the communication and network module is to enable bidirectional data exchange between the data processing unit and the remote monitoring center, and to provide the network support required for data transmission between the image acquisition unit, data processing unit and bait delivery execution unit within the image recognition-based termite control system.

[0049] In one embodiment, the communication network within the image recognition-based termite control system is divided into a data acquisition network and an equipment control network.

[0050] The data acquisition network is used for data transmission between the image acquisition unit and the data processing unit. It is implemented via a high-bandwidth wired connection and employs a high-throughput data transmission protocol. The data acquisition network is specifically designed to carry large-volume time-series image frames generated by the image acquisition unit. .

[0051] The equipment control network is used for communication between the data processing unit and the bait delivery execution unit. The equipment control network adopts industrial fieldbus protocols, such as the Modbus RTU protocol based on RS-485 serial communication, or the Modbus TCP / IP protocol based on Ethernet. The equipment control network is specifically used to carry the extermination decision instructions generated by the data processing unit, which have small data volume but high real-time requirements, and to receive the status confirmation signals returned by the bait delivery execution unit.

[0052] The communication and network module includes a wide area network (WAN) interface for connecting to a remote monitoring center. The WAN interface can be an Ethernet interface, a Wi-Fi module, or a cellular communication module.

[0053] The communication and network module supports the TCP / IP protocol stack and is used to upload specific data generated by the data processing unit to the remote monitoring center. This data includes: probabilistic pheromone potential energy maps generated by the PPM aggregation module. Historical snapshots and time series of C-type node divergence generated by the closed-loop feedback module The data upload uses standard application layer protocols, such as HTTPS or MQTT, and also includes decision logs generated by the strategy decision module.

[0054] Meanwhile, the communication and network module is used to receive remote configuration instructions or firmware update packages from the remote monitoring center and send these instructions to the data processing unit to remotely adjust the parameters of the termite activity extermination system based on image recognition.

[0055] In one specific embodiment, the data processing unit stores data in non-volatile memory and volatile memory, and the image recognition-based termite extermination method is executed by the central processing unit and the graphics processor.

[0056] See attached document Figure 2 The present invention also provides a termite extermination method based on image recognition, which is initiated by an image preprocessing module that receives raw time-series image frames from an image acquisition unit. .

[0057] The image preprocessing module first processes the time-series image frames. Gaussian filtering is applied to remove random noise, resulting in a denoised image frame. .

[0058] Subsequently, the image preprocessing module employs an adaptive Gaussian mixture model technique, based on... Dynamically update background model Through the denoised image frame With background model Perform difference analysis to extract the foreground mask image. Among them, the foreground mask image The non-zero pixel regions in the image correspond to individual termites that are moving.

[0059] The image preprocessing module further calculates the camera intrinsic parameter matrix from the image acquisition unit. and distortion coefficient Foreground mask image Perform distortion correction.

[0060] Finally, the image preprocessing module applies the precalibrated perspective transformation matrix. The coordinates of the distortion-corrected image Converted to world coordinates consistent with the physical plane of the monitored area Processed foreground mask image It is transmitted to the instantaneous activity vector field generation module.

[0061] The instantaneous motion vector field generation module receives two consecutive preprocessed foreground images. and .

[0062] The instantaneous activity vector field generation module uses the Farnebäck dense optical flow algorithm to calculate... All non-zero pixels in The displacement generates a pixel-level displacement field. ,in, and They are and Displacement in the direction.

[0063] Instantaneous activity vector field generation module for displacement field Clustering is performed to merge the neighboring pixel displacements belonging to the same termite individual, and the number of neighboring pixel displacements is calculated. Location of the centroid of an individual termite Average speed and average direction .

[0064] Among them, the speed magnitude and direction Calculated using the following formula: ; ; in, and It is the first The average displacement corresponding to each individual termite.

[0065] The instantaneous activity vector field generation module will generate all the data in the instantaneous activity vector field generation module. Recognized at all times The vector information of individual termites is combined to output an instantaneous activity vector field. , It is transmitted to the group macro-state assessment module, PPM aggregation module, and path function parsing module.

[0066] Group macro-state assessment module receives Extract the total number of vectors at the current moment. and the direction of each vector .

[0067] The macroscopic state assessment module of the population calculates the global ordered parameters. The calculation formula is: ; in: Is The total number of activity vectors detected at any given time It is the first A vector in Direction and angle at any moment It is the imaginary unit.

[0068] The group macro-state assessment module will calculate the With the preset ordered state threshold If a comparison is made, > If the condition is met, the group is determined to be in an ordered state; otherwise, it is determined to be in a random state. The group macro-state assessment module outputs this state signal to the strategy decision module.

[0069] PPM aggregation module receives The PPM aggregation module maintains a two-dimensional floating-point matrix, i.e., a probabilistic pheromone potential energy map, in both the non-volatile and volatile memory of the data processing unit. .

[0070] The PPM aggregation module updates at each time step according to the following recursive formula. renew : ; in: It is the potential energy diagram from the previous moment; It is a decay factor less than 1, used to simulate the volatilization of pheromones; Is The increment contributed by the current activity at any given moment.

[0071] Activity increment The calculation formula is: ; in: Indicates the position of the center of mass Falling on the probabilistic pheromone potential energy map In the matrix All vectors within a grid cell A set; It is the first The magnitude of the velocity of each vector.

[0072] The PPM aggregation module will update the probabilistic pheromone potential energy map. The output is sent to the path function parsing module and the network resilience assessment module.

[0073] The path function parsing module receives a probabilistic pheromone potential energy map. and instantaneous activity vector field .

[0074] The path function parsing module first analyzes the probabilistic pheromone potential energy map. Apply a thresholding process to extract high-resolution images. The high-value region was then processed using the Zhan-Suen skeleton extraction algorithm to extract the high-value region. The value region is refined into a path skeleton with a width of one pixel.

[0075] The path parsing module analyzes the path skeleton, identifies the intersection points (nodes) and connecting paths (edges), and uses the probabilistic pheromone potential energy map. The integral value on the edge is used to classify the edge into class A or class B. Class A is a high-traffic main road, and class B is a low-traffic reconnaissance path.

[0076] Subsequently, the path parsing module analyzes the neighborhood of each identified node. Inside, calculation divergence On discrete grids, divergence is calculated using the central difference method: ; in: and It is a vector field exist and Components in direction; and It is the spatial step size of the grid.

[0077] The path parsing module performs a sliding window time-series average of the divergence value of each node to obtain... .

[0078] The path parsing module is based on The values ​​are used to classify nodes by function: like < ,in, If the threshold is positive, it is classified as a Class C node, i.e., a termite gathering point or nest entrance; like It is determined to be a Class D node, that is, a source node or a food source.

[0079] The path analysis module outputs lists of C-type and D-type nodes to the causal chain quantization module and the policy decision module, and outputs the divergence time series of the C-type nodes. The output is sent to the closed-loop feedback module.

[0080] The causal chain quantization module receives the classification results of C-class and D-class nodes and the real-time divergence time series. and .

[0081] Causal chain quantization module calculation and Normalized time-delay cross-correlation function between : ; in: It is a time lag; It is the calculation of expected value; , These are the mean values ​​of their respective time series; , It is the standard deviation of their respective time series.

[0082] Causal chain quantization module in Search within the interval >0 The peak value, if at If a peak value exceeding the threshold is detected at a value greater than 0, then it is determined that... and There exists a causal chain from recruitment to response, and the causal chain quantification module outputs this causal chain confirmation signal to the strategy decision module.

[0083] Network resilience assessment module receives probabilistic pheromone potential energy diagram The network resilience assessment module is used to assess network resilience in response to a detected disturbance event. Start up afterward.

[0084] Network resilience assessment module receives probabilistic pheromone potential energy diagram The network resilience assessment module is used to assess network resilience in response to a detected disturbance event. After startup, disturbance event For example, to assess the disruption of termite activity paths caused by bait deployment, the network resilience assessment module first performs topological analysis on the probabilistic pheromone potential energy map to identify areas disrupted by high-level pheromone activity. The path network formed by the value region.

[0085] The network resilience assessment module monitors blocked legacy paths. of The decay time is measured, and the image recognition-based termite control system is simultaneously monitored to determine whether a new detour path has been created. Detour route It is based on the probabilistic pheromone potential energy map It is formed naturally through dynamic updates.

[0086] If an image recognition-based termite control system creates a new detour path... The network resilience assessment module calculates new paths. of The time required for the value to reach a stable state.

[0087] The network resilience assessment module will compare the network to a preset resilience time threshold: If the formation time is less than the toughness time threshold, or the formation time is extremely short, the network formed by the path is determined to be a high-toughness network. If the formation time is greater than or equal to the toughness time threshold or path The network that failed to form is determined to be a fragile network.

[0088] The network resilience assessment module outputs the network resilience classification results or vulnerable networks to the policy decision module.

[0089] The network resilience assessment module monitors blocked legacy paths. of Value decay time Simultaneously monitor whether the image recognition-based termite control system forms a new detour path. .

[0090] If an image recognition-based termite control system creates a new detour path... The network resilience assessment module calculates new paths. of The time required for the value to reach a stable state .

[0091] The network resilience assessment module will With a preset resilience time threshold Comparison: like < or Extremely short, classified as a highly resilient network. ; like ≥ or Failed to form, identified as a fragile network .

[0092] The network resilience assessment module classifies network resilience results. or Output to the strategy decision module.

[0093] The strategy decision-making module integrates input signals from multiple modules: the ordered state signal from the group macro-state assessment module, the A-class path and C / D-class node positions from the path function analysis module, the causal chain confirmation signal from the causal chain quantification module, and the network resilience classification results from the network resilience assessment module. .

[0094] The strategy decision-making module first performs a logical judgment: (Condition 1) > (The group is orderly); (Condition 2) The target A-type path has been confirmed by the causal chain.

[0095] The strategy decision module bases its decisions on the network resilience classification results only if both of the above conditions are met simultaneously. Execution strategy selection: like = (For fragile networks), the strategy decision module outputs strategy A: execute a single-point, sufficient bait deployment at the center point of path A.

[0096] like = (For a highly resilient network), the strategy decision module outputs strategy B: perform multi-point, micro-scale, and gradual bait deployment on multiple related Class A paths.

[0097] The strategy decision module outputs the selected strategy (A or B) and the corresponding deployment parameters (target coordinates, deployment quantity) to the instruction generation module.

[0098] The instruction generation module receives the deployment strategy and converts it into physical instructions that can be recognized by the local controller of the decoy deployment execution unit.

[0099] The closed-loop feedback module is activated after receiving a status confirmation signal from the decoy deployment execution unit.

[0100] The closed-loop feedback module continuously monitors the divergence time series of all Class C nodes provided by the path function parsing module. .

[0101] The closed-loop feedback module will With a preset elimination confirmation threshold (This value is close to zero) for comparison.

[0102] If all C-class nodes The values ​​are all lower than And it continues for more than a preset confirmation time. Then the closed-loop feedback module generates an extermination confirmation signal and stores the extermination confirmation signal in the decision log or sends it to the remote monitoring center.

[0103] In one specific embodiment, the data storage and management functions are performed by the data processing unit and the non-volatile memory within the data processing unit, and optionally in collaboration with a remote monitoring center via a communication and network module.

[0104] The non-volatile memory in the data processing unit is used to store the configuration data necessary for the operation of the termite extermination method based on image recognition and the analysis data generated during the operation.

[0105] Configuration data includes: the attenuation factor used by the PPM aggregation module. The ordered state threshold used in the group macro-state assessment module The divergence threshold used by the path function parsing module ; and the resilience time threshold used in the network resilience assessment module. .

[0106] The analysis data includes: the instantaneous activity vector field generated by the instantaneous activity vector field generation module at a specific time point. The sampled data; the globally ordered parameters generated by the population macroscopic state assessment module. The time series.

[0107] The analysis also includes: probabilistic pheromone potential energy maps generated by the PPM aggregation module. Periodic snapshots; divergence time series of C-type and D-type nodes generated by the path function parsing module. and .

[0108] The analysis also includes: time-delay cross-correlation calculated by the causal chain quantization module. Results; Network resilience classification results generated by the network resilience assessment module The decision log generated by the strategy decision-making module; and the extermination confirmation signal generated by the closed-loop feedback module.

[0109] In one embodiment, a database management system runs on the data processing unit for structured storage of the analyzed data. The database management system may include a time-series database specifically designed for efficient storage and retrieval of time-series data with timestamps, such as... , and .

[0110] A database management system may also include a relational database for storing configuration data, decision logs, and... The metadata of the snapshot.

[0111] For original time-series image frames with large data volumes Used for system debugging or historical backtracking, this data is stored in a file format and stored in the non-volatile memory of the data processing unit, while the index information is recorded in a relational database.

[0112] The data processing unit also executes data lifecycle management strategies, such as locally stored raw time-series image frames. After a preset storage period, such as 30 days, the data is automatically deleted or overwritten to ensure the recycling of storage space, while decision logs and extermination confirmation signals are set to be stored permanently or long-term.

[0113] The communication and network module is used to upload specific data in non-volatile memory to a remote monitoring center according to preset rules, such as timed or event-triggered events, to achieve remote data backup and long-term archiving.

[0114] Data storage and management also include access control mechanisms to ensure that read and write operations on configuration data and analysis data require authorization and verification.

[0115] In one specific embodiment, the hardware interface between the image acquisition unit and the data processing unit is established through a first data interface. The image acquisition unit transmits time-series image frames through the first data interface according to a predetermined protocol. It is transmitted to the data processing unit in real time.

[0116] The hardware interface between the data processing unit and the bait delivery execution unit is established through a second data interface. The data processing unit sends the extermination decision command to the local controller of the bait delivery execution unit through the second data interface in accordance with a predetermined protocol.

[0117] The communication and network module connects to the main processor of the data processing unit via the internal bus of the data processing unit or an external Ethernet interface, and is used to perform the uploading of analysis data and the receiving of remote configuration commands.

[0118] Within the data processing unit, the image preprocessing module, instantaneous activity vector field generation module, group macroscopic state assessment module, PPM aggregation module, path function parsing module, causal chain quantization module, network resilience assessment module, policy decision-making module, instruction generation module, and closed-loop feedback module are all software modules that execute on the processor of the data processing unit.

[0119] Data exchange between these software modules is achieved through a shared memory region in the volatile memory of the data processing unit. For example, the image preprocessing module outputs a foreground mask image. The data is written to the first shared memory buffer; the instantaneous activity vector field generation module reads data from the first shared memory buffer and outputs the instantaneous activity vector field. Write to the second shared memory buffer; the group macro-state assessment module, PPM aggregation module, and path function parsing module read from the second shared memory buffer. Data serves as input, and so on. The execution order and data synchronization between modules are managed through the operating system's thread scheduling and synchronization primitives.

[0120] The instruction generation module encapsulates the generated physical instructions into data frames conforming to the Modbus RTU protocol through the serial communication driver of the operating system of the data processing unit. The data frames are then sent to the local controller through the physical layer of the second data interface.

[0121] When integrating an image recognition-based termite control system, the image acquisition unit and the bait delivery execution unit are first installed in the monitoring area. During installation, it is ensured that the physical movement range of the multi-axis motion mechanism of the bait delivery execution unit coincides with the physical area covered by the field of view of the image acquisition unit.

[0122] After integrating the termite control system based on image recognition, a coordinate system calibration process must be performed. This process is used to establish a precise mapping relationship between the image coordinate system of the image acquisition unit and the world coordinate system of the multi-axis motion mechanism of the bait delivery execution unit.

[0123] The mapping relationship can be a 3x3 perspective transformation matrix. Perspective transformation matrix The perspective transformation matrix was calculated by placing multiple calibration points with known physical coordinates in the monitoring area, obtaining the corresponding image pixel coordinates, and then fitting the data using the least squares method. It is stored in the non-volatile memory of the data processing unit.

[0124] When an image recognition-based termite control system is running, the deployment location calculated by the strategy decision module must be subjected to a perspective transformation matrix before being used by the instruction generation module to generate extermination decision instructions. Coordinate transformation is performed to convert image coordinates into precise physical world coordinates that can be executed by multi-axis motion mechanisms. .

[0125] After the integration and calibration of the image recognition-based termite control system are completed, a closed-loop testing process is executed. The closed-loop testing process includes: At any location within the monitoring area Place a physical marker.

[0126] The image acquisition unit acquires images of the physical markers, and the data processing unit calculates the image coordinates. .

[0127] The instruction generation module uses the perspective transformation matrix. Image coordinates Convert to physical coordinates It then generates movement commands and sends them to the decoy deployment execution unit.

[0128] The decoy delivery unit drives the dispensing nozzle to move to the physical coordinates. Place.

[0129] Confirm the physical tip and physical markings of the dispensing nozzle. Their physical positions coincide, with the deviation within the preset allowable error. Within this range, it is used to verify the accuracy of system calibration and integration.

Claims

1. A method for termite control based on image recognition, characterized in that, Includes the following steps: S1. Acquire time-series image frames of the monitored area; S2. Calculate and generate an instantaneous activity vector field based on the time-series image frames; S3. Based on the instantaneous activity vector field, aggregate to generate and dynamically update the probabilistic pheromone potential energy map; S4. Extract the topology of the probabilistic pheromone potential energy map to identify paths and nodes, and classify the nodes by function in conjunction with the instantaneous activity vector field. S5. Based on a comprehensive analysis of the instantaneous activity vector field, the probabilistic pheromone potential energy map, and the functional classification of the nodes, generate an extermination decision command; The comprehensive analysis includes assessing the macroscopic state of the group, quantifying the causal relationships between nodes, and determining the resilience of the path network; S6. Execute the extermination decision command, drive the bait deployment execution unit to perform deployment, and after deployment, continuously monitor the parameters related to the node to confirm the extermination effect in a closed loop.

2. The termite control method based on image recognition according to claim 1, characterized in that, In step S3, the specific steps for aggregating, generating, and dynamically updating the probabilistic pheromone potential energy map include: S301. Apply an attenuation factor to the probabilistic pheromone potential energy map of the previous time step to attenuate it. S302. Calculate the activity increment based on the velocity magnitudes of each vector in the instantaneous activity vector field at the current moment; S303. Add the attenuated probabilistic pheromone potential energy map to the activity increment to obtain the probabilistic pheromone potential energy map at the current time.

3. The termite control method based on image recognition according to claim 1, characterized in that, In step S4, the specific steps for classifying the functions of the nodes include: S401. Calculate the divergence of the instantaneous activity vector field in the neighborhood of each node; S402. Based on the sign and magnitude of the divergence, classify the nodes into Class C nodes and Class D nodes.

4. The termite control method based on image recognition according to claim 1, characterized in that, In step S5, the specific steps for assessing the macroscopic state of the group include: S501a. Calculate the global ordered parameters based on the instantaneous active vector field; S502a. The global ordered parameter is compared with a preset ordered state threshold as one of the triggering conditions for generating the extermination decision command.

5. The termite control method based on image recognition according to claim 3, characterized in that, The specific steps for quantifying the causal relationships between nodes in step S5 include: S501b, Extract the divergence time series of the C-type nodes and the D-type nodes; S502b: Calculate the time-delay cross-correlation between the divergence time series to confirm the causal relationship between the D-type nodes and the C-type nodes in recruiting responses.

6. The termite control method based on image recognition according to claim 1, characterized in that, In step S5, the specific steps for determining the resilience of the path network include: S501c. After detecting a disturbance event, monitor the change in the probabilistic pheromone potential energy map to calculate the formation time of the new path; S502c. Based on the formation time of the new path, determine whether the network formed by the path is a highly resilient network or a fragile network.

7. The termite control method based on image recognition according to claim 1, characterized in that, In step S5, the specific steps for generating the extermination decision command also include: Strategy selection is based on the resilience of the path network: If the network is a fragile network, then a decision to eliminate the virus by executing a single-point, sufficient-quantity extermination command is generated on a single Class A path. If the network is a highly resilient network, then it generates extermination decision instructions that are executed on multiple Class A paths in a multi-point, micro-scale, and gradual manner.

8. The termite control method based on image recognition according to claim 3, characterized in that, In step S6, the step of confirming the closed-loop effect of the extermination specifically includes: S601. Continuously monitor the divergence time series of all C-type nodes; S602. When the divergence value of all C-type nodes is lower than the preset extermination confirmation threshold and continues for a preset duration, an extermination confirmation signal is generated.

9. A termite control system based on image recognition, used to execute the termite control method based on image recognition according to any one of claims 1-8, characterized in that, include: The image acquisition unit is used to acquire time-series image frames of the monitored area; The bait delivery execution unit is used to receive the extermination decision instruction generated by the data processing unit, and execute the bait delivery action at the specified physical coordinates within the monitoring area according to the instruction; The data processing unit, connected to the image acquisition unit and the bait deployment execution unit, is used to analyze and classify the termite activity network by generating an instantaneous activity vector field and a probabilistic pheromone potential energy map based on the time-series image frames acquired by the image acquisition unit; and to generate extermination decision instructions to drive the bait deployment execution unit based on a comprehensive analysis of the macroscopic state of the colony, the causal relationship between nodes, and the network resilience; and to conduct closed-loop confirmation of the extermination effect through continuous monitoring after deployment.

10. A storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a termite control method based on image recognition as described in any one of claims 1-8.

Images