Termite monitoring and killing method based on multi-modal sensing and dynamic threshold

Abstract

The invention relates to the technical field of termite monitoring and killing, in particular to a termite monitoring and killing method based on multi-modal sensing and a dynamic threshold value, and adopts the technical scheme that multi-source signal acquisition is adopted, and biological activity signals of a target area are synchronously acquired through a vibration sensor, a sonic sensor, a chemical gas sensor and a thermal imaging sensor; the anti-interference performance is improved, so that the false alarm rate is reduced; environmental parameter compensation is carried out, temperature, humidity and soil conductivity data are collected in real time, a dynamic calibration coefficient is generated to correct sensor original data, and environmental adaptability is improved; the method comprises the following steps: acquiring data of a current area, performing dynamic threshold calculation, inputting the acquired data into an LSTM-Transformer hybrid network, and outputting a self-adaptive early warning threshold of the current area so as to reduce the possibility of leak detection; through multi-stage early warning judgment, the real-time performance is improved, and the early warning delay is reduced; and starting a corresponding disposal process according to the early warning level, performing response execution, and narrowing the spraying range, thereby reducing pollution.

Description

Technical Field

[0001] The present invention relates to the technical field of termite monitoring and killing, and in particular to a termite monitoring and killing method based on multi-modal perception and dynamic threshold. Background Technique

[0002] Termites are insects that cause serious damage to buildings, furniture, crops, etc. Their harms are mainly reflected in the following aspects: Termites feed on wood, which will erode the wooden structure of houses, resulting in a decrease in the load-bearing capacity of buildings, and may even cause the collapse of houses, resulting in property losses and casualties; A large amount of excrement and termite galleries will be left by termite activities, which not only affect household hygiene, but also damage the appearance of wooden items such as furniture and floors; Although termites do not directly spread diseases themselves, their activities may carry pathogens such as bacteria and fungi, increasing the risk of human disease infection and even causing allergic reactions; Termites will attack the roots of crops, resulting in hindered plant growth and affecting yield and quality. At the same time, they will also damage forest resources and have a negative impact on the forestry economy.

[0003] Currently, the control measures for termites include eliminating attracting factors, placing traps, spraying insecticides, and fumigation treatment. However, for termite monitoring, a single sensor is vulnerable to interference, resulting in a high false alarm rate; the static threshold is rigid, with poor environmental adaptability, easily leading to missed detections; the real-time performance is insufficient, and the warning delay is high; and wide-area spraying causes large pollution.

[0004] In view of this, we propose a termite monitoring and killing method based on multi-modal perception and dynamic threshold to solve the existing problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a termite monitoring and killing method based on multi-modal perception and dynamic threshold to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A termite monitoring and killing method based on multi-modal perception and dynamic threshold, the operation steps include:

[0007] S1. Multi-source signal acquisition: Synchronously obtain biological activity signals in the target area through vibration sensors, acoustic sensors, chemical gas sensors, and thermal imaging sensors;

[0008] S2. Environmental parameter compensation: Real-time collect temperature, humidity, and soil conductivity data, and generate a dynamic calibration coefficient to correct the original sensor data;

[0009] S3. Dynamic threshold calculation: Input the collected data into the LSTM-Transformer hybrid network, and output the adaptive warning threshold T of the current area dynamic ;

[0010] S4. Multi - level warning determination: When the real - time density value D real satisfies 0.5T dynamic <D real ≤0.8T dynamic at this time, a primary warning is triggered; when D real >0.8T dynamic and it lasts for 3 sampling periods, a secondary warning is triggered; when the density change rate or the predicted density D predict >1.2T dynamic at this time, a high - level warning is triggered; where h is the unit time;

[0011] S5. Response execution: Start the corresponding disposal process according to the warning level, including data review, manual verification, or linkage with the killing device.

[0012] Furthermore, in S1: The vibration sensor is a broadband MEMS accelerometer with a sensitivity ≥ 5mV / g and a working frequency band covering 1 - 10kHz; the acoustic wave sensor is equipped with an adaptive band - stop filter to suppress 50Hz power frequency interference and wind noise; the chemical gas sensor uses a MOX semiconductor array with a detection limit for acetic acid ≤ 0.2ppm and a response time < 10s.

[0013] Furthermore, in S2: When the soil humidity > 40%, the sensitivity threshold of the vibration signal is automatically increased by 15% - 30%; calculate the influence of temperature on the diffusion rate of the pheromone gas based on the Arrhenius equation and correct the concentration detection value.

[0014] Furthermore, in S3, the specific implementation method of dynamic threshold calculation includes: using a temporal convolutional network to extract the periodic features of historical density data; fusing environmental parameters and real - time signals through a Transformer encoder to generate a spatio - temporal attention weight matrix; optimizing the threshold adjustment strategy based on a reinforcement learning framework.

[0015] Furthermore, in S4, the response measures for the secondary warning include: controlling the drone equipped with a thermal imager to perform grid scanning on the warning area to generate a heat map of the ant - nest probability; using the YOLO - Ant algorithm to identify the suspected ant - nest area with a positioning accuracy error ≤ 0.5m; automatically generating a work order containing coordinates, risk level, and disposal suggestions and transmitting it to the operation and maintenance terminal through LoRaWAN.

[0016] Furthermore, in S5, the execution logic of the linkage killing device for the high - level warning includes: calculating the chemical agent spraying path according to the density diffusion direction; adopting pressure - gradient spraying control with the chemical agent concentration in the central area being 3 times that of the edge area; starting density decay monitoring for 24 hours after killing, then triggering secondary killing.

[0017] Furthermore, it also includes a self-learning optimization module: building a false alarm case library in the cloud, generating adversarial samples through contrastive learning, and updating the edge detection model quarterly; adopting a federated learning framework to aggregate model parameters in multiple deployment areas and generate a global optimization model.

[0018] Furthermore, it also includes a reliability guarantee mechanism: deploying redundant sensor nodes and using a voting mechanism to eliminate abnormal data; when the node communication is interrupted, switching to an offline threshold calculation mode and using the local data of the last 7 days to maintain the basic warning function.

[0019] Furthermore, on the basis of the reliability guarantee mechanism, it further includes steps to enhance data reliability: deploying reference sensor nodes and periodically sending test signals to verify the integrity of the monitoring network; when the node data deviates from the cluster median by ±2σ, automatically triggering a self-check program and switching to an alternative sampling strategy, where σ is the standard deviation of the data points deviating from the average value.

[0020] Furthermore, it also includes risk prediction expansion: building a regional ant nest association topology based on a graph neural network, combining meteorological data to predict the termite migration path in the next 72 hours; outputting a risk heat map and marking the priority treatment areas.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] The present invention uses multi-source signal acquisition to synchronously obtain the biological activity signals in the target area through vibration sensors, acoustic sensors, chemical gas sensors, and thermal imaging sensors, improving the anti-interference ability and thus reducing the false alarm rate; performing environmental parameter compensation, collecting temperature, humidity, and soil conductivity data in real time, generating a dynamic calibration coefficient to correct the original sensor data, and improving the environmental adaptability; and performing dynamic threshold calculation, inputting the collected data into an LSTM-Transformer hybrid network, and outputting the adaptive warning threshold of the current area to reduce the possibility of missed detection; improving the real-time performance through multi-level warning determination and reducing the warning delay; starting the corresponding disposal process according to the warning level, performing response execution, and narrowing the spraying range to reduce pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic flowchart of a termite monitoring and killing method based on multi-modal perception and dynamic threshold of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0025] Embodiment 1

[0026] As Figure 1As shown in the figure, a termite monitoring and killing method based on multi-modal perception and dynamic threshold, the operation steps include:

[0027] S1. Multi-source signal acquisition: Synchronously obtain the biological activity signals of the target area through vibration sensors, acoustic wave sensors, chemical gas sensors and thermal imaging sensors;

[0028] S2. Environmental parameter compensation: Real-time collect temperature, humidity and soil conductivity data, and generate dynamic calibration coefficients to correct the original sensor data;

[0029] S3. Dynamic threshold calculation: Input the collected data into the LSTM-Transformer hybrid network, and output the adaptive warning threshold T of the current area dynamic ;

[0030] S4. Multi-level warning determination: When the real-time density value D real satisfies 0.5T dynamic <D real ≤0.8T dynamic at that time, trigger the primary warning; when D real >0.8T dynamic and lasts for 3 sampling periods, trigger the intermediate warning; when the density change rate or the predicted density D predict >1.2T dynamic at that time, trigger the advanced warning; where h is the unit time;

[0031] S5. Response execution: Start the corresponding disposal process according to the warning level, including data review, manual verification or linkage with the killing device.

[0032] The working principle of a termite monitoring and killing method based on multi-modal perception and dynamic threshold according to Embodiment 1 is:

[0033] In S1: The vibration sensor is a broadband MEMS accelerometer with a sensitivity ≥ 5mV / g and a working frequency band covering 1 - 10kHz; The acoustic wave sensor is equipped with an adaptive band-stop filter to suppress 50Hz power frequency interference and wind noise; The chemical gas sensor uses a MOX semiconductor array with a detection limit for acetic acid ≤ 0.2ppm and a response time < 10s.

[0034] Verify the termite activity characteristics through multi-dimensional signal cross-validation: The vibration sensor detects the specific frequency (2 - 8kHz) of termites gnawing on wood; The acoustic wave sensor captures the acoustic fingerprint map of group activities (frequency domain resolution ≤ 0.1Hz); The chemical sensor detects characteristic gases such as acetic acid and nonenyl alcohol released by termites (detection limit ≤ 0.5ppm); The thermal imaging sensor identifies abnormal temperature rise of the termite nest (accuracy is ±0.3℃). Eliminate false alarms of a single sensor through multi-source data fusion and improve the recognition accuracy of termite activities.

[0035] In S2: When the soil humidity > 40%, the vibration signal sensitivity threshold is automatically increased by 15% - 30%; calculate the influence of temperature on the diffusion rate of pheromone gas based on the Arrhenius equation, and correct the concentration detection value.

[0036] The corrected concentration detection value is where C raw is the original concentration detection value; E a is the Arrhenius activation energy, with a default value of 45 kJ / mol; R is the molar gas constant, with the unit kJ / (mol·K); T ref is the reference temperature, T ref = 298 K; T real is the absolute temperature of the environment, with the unit K.

[0037] In S3, the specific implementation method of dynamic threshold calculation includes: using a temporal convolutional network to extract the periodic features of historical density data; fusing environmental parameters and real-time signals through a Transformer encoder to generate a spatio-temporal attention weight matrix; optimizing the threshold adjustment strategy based on a reinforcement learning framework.

[0038] The dynamic warning threshold is adaptively adjusted by region, and the formula is: T dynamic = T base ×(1 + α·ΔE), where T base is the base threshold, ΔE is the environmental change factor, α is the learning rate coefficient, and α is optimized through reinforcement learning.

[0039] Proximal Policy Optimization (PPO) is a reinforcement learning method that balances performance improvement and training stability by restricting the policy update amplitude. Its core mechanisms include truncated objective functions and importance sampling, and it is applicable to continuous and discrete action space tasks. The reward function is a core concept in reinforcement learning, used to quantify the immediate value of an agent taking a certain action in a specific state, thereby guiding it to learn the optimal policy. In the PPO algorithm, the reward function is closely related to policy optimization and achieves efficient optimization through policy gradient updates and value function estimation. The present invention uses the PPO algorithm to optimize the threshold adjustment strategy, and the reward function is designed as:

[0040] In S4, the response measures for intermediate warnings include: controlling a drone equipped with a thermal imager to perform grid scanning on the warning area to generate an ant nest probability heat map; using the YOLO-Ant algorithm to identify suspected ant nest areas with a positioning accuracy error ≤ 0.5 m; automatically generating a work order containing coordinates, risk levels, and disposal suggestions, and transmitting it to the operation and maintenance terminal through LoRaWAN.

[0041] In S5, the execution logic of the linkage killing device for advanced warning includes: calculating the chemical agent spraying path according to the density diffusion direction, and the path planning satisfies: coverage radius ≥ 1.5 × predicted diffusion distance; adopting pressure gradient spraying control, and the chemical agent concentration in the central area is 3 times that in the edge area; after killing, start density decay monitoring for 24 hours, then trigger secondary killing.

[0042] In the three-level warning system, the primary warning is regarded as a potential risk, and the triggering condition is: the real-time density value is greater than 50% of the dynamic threshold; the response measure is: start high-frequency data collection, that is, increase the sampling rate to 10Hz, and push the warning to the local terminal. The intermediate warning is regarded as a confirmed threat, and the triggering condition is: the density value is greater than 80% of the dynamic threshold for 3 consecutive cycles; the response measure is: activate the drone re-inspection, generate a probability distribution map of the ant nest, and push the work order to the operation and maintenance personnel. The advanced warning is regarded as an emergency spread, and the triggering condition is: the density growth rate is greater than 5% / hour or it is predicted to exceed the threshold by 120% in the next 6 hours; the response measure is: the linkage killing system automatically sprays slow-release chemical agents to block the diffusion path.

[0043] The present invention also includes a self-learning optimization module: building a false alarm case library in the cloud, generating adversarial samples through contrastive learning, and updating the edge detection model every quarter; adopting a federated learning framework to aggregate the model parameters of multiple deployment areas to generate a global optimization model.

[0044] The present invention also includes a reliability guarantee mechanism: deploying redundant sensor nodes, and adopting a voting mechanism to eliminate abnormal data; when the node communication is interrupted, switch to the offline threshold calculation mode, and use the local data of the last 7 days to maintain the basic warning function.

[0045] The voting mechanism adopted to eliminate abnormal data is the two-out-of-three consistency principle. The two-out-of-three consistency principle is a redundancy voting mechanism, mainly used to improve the accuracy and reliability of system decisions. Its core rule is: when at least two of the three redundant signals have the same output, it is determined that the valid condition is satisfied.

[0046] On the basis of the reliability guarantee mechanism, it further includes steps to enhance data reliability: deploying reference sensor nodes, periodically sending test signals to verify the integrity of the monitoring network; when the node data deviates from the cluster median by ±2σ, automatically trigger a self-check program and switch to an alternative sampling strategy, where σ is the standard deviation of the data point deviating from the average value.

[0047] The present invention also includes risk prediction expansion: constructing a regional ant nest association topology based on a graph neural network, combining meteorological data to predict the termite migration path in the next 72 hours; outputting a risk heat map and marking the priority processing area, where the predicted density growth rate > 5% / day.

[0048] The above specific embodiments are only several preferred embodiments of the present invention. Based on the technical solution of the present invention and the relevant revelations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A termite monitoring and killing method based on multi-modal perception and dynamic threshold, characterized in that The operation steps include: S1. Multi-source signal acquisition: Synchronously obtain the biological activity signals of the target area through vibration sensors, acoustic wave sensors, chemical gas sensors, and thermal imaging sensors; S2. Environmental parameter compensation: Real-time collect temperature, humidity, and soil conductivity data, and generate dynamic calibration coefficients to correct the original sensor data; S3. Dynamic threshold calculation: Input the collected data into the LSTM-Transformer hybrid network to output the adaptive warning threshold T of the current area dynamic ; S4. Multi - level warning determination: When the real - time density value D real satisfies 0.5T dynamic <D real ≤0.8T dynamic at this time, a primary warning is triggered; When D real >0.8T dynamic and it lasts for 3 sampling periods, a secondary warning is triggered; When the density change rate or the predicted density D predict >1.2T dynamic at this time, a high - level warning is triggered; where h is the unit time; S5. Response execution: Initiate corresponding disposal processes according to the warning level, including data review, manual verification, or linkage with killing devices.

2. The termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, wherein In S1: The vibration sensor is a broadband MEMS accelerometer with a sensitivity ≥5mV / g and a working frequency band covering 1 - 10kHz; The acoustic wave sensor is equipped with an adaptive band-stop filter to suppress 50Hz power frequency interference and wind noise; The chemical gas sensor uses a MOX semiconductor array with a detection limit for acetic acid ≤0.2ppm and a response time <10s.

3. The termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, wherein In S2: When the soil humidity > 40%, the vibration signal sensitivity threshold is automatically increased by 15% - 30%; Calculate the influence of temperature on the diffusion rate of pheromone gas based on the Arrhenius equation, and correct the concentration detection value.

4. A termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, characterized in that, In S3, the specific implementation method of dynamic threshold calculation includes: Using a temporal convolutional network to extract the periodic features of historical density data; Fusing environmental parameters and real-time signals through a Transformer encoder to generate a spatio-temporal attention weight matrix; Optimizing the threshold adjustment strategy based on a reinforcement learning framework.

5. A termite monitoring and killing method based on multimodal perception and dynamic threshold according to claim 1, characterized in that In S4, the response measures for medium-level warnings include: Controlling a drone equipped with a thermal imager to conduct grid scanning of the warning area to generate a heat map of the ant nest probability; Using the YOLO-Ant algorithm to identify suspected ant nest areas with a positioning accuracy error ≤0.5m; Automatically generate a work order containing coordinates, risk levels, and disposal suggestions, and transmit it to the operation and maintenance terminal through LoRaWAN.

6. The termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, wherein, In S5, the execution logic of the linkage killing device for advanced warning includes: calculating the chemical agent spraying path according to the density diffusion direction; adopting pressure gradient spraying control, with the chemical agent concentration in the central area being 3 times that in the edge area; starting density decay monitoring for 24 hours after killing, if the decay rate < then trigger secondary killing.

7. A method for termite monitoring and killing based on multi-modal perception and dynamic threshold according to claim 1, characterized in that, It also includes a self-learning optimization module: Build a false alarm case library in the cloud, generate adversarial samples through contrastive learning, and update the edge detection model every quarter; Adopt a federated learning framework to aggregate the model parameters of multiple deployment areas to generate a global optimization model.

8. A termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, characterized in that, It also includes a reliability guarantee mechanism: Deploy redundant sensor nodes and use a voting mechanism to eliminate abnormal data; When the node communication is interrupted, switch to an offline threshold calculation mode and use the local data of the last 7 days to maintain the basic warning function.

9. The termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 8, characterized in that, On the basis of the reliability guarantee mechanism, it further includes steps to enhance data reliability: Deploy reference sensor nodes and periodically send test signals to verify the integrity of the monitoring network; When the node data deviates from the cluster median by ±2σ, automatically trigger a self-check program and switch to an alternative sampling strategy, where σ is the standard deviation of the data points deviating from the average value.

10. A termite monitoring and killing method based on multi-modal perception and dynamic threshold according to claim 1, characterized in that, It also includes risk prediction expansion: Build a regional ant nest association topology based on a graph neural network, combine meteorological data to predict the termite migration path in the next 72 hours; Output a risk heat map and mark the priority processing areas.

Images