Termite monitoring system and method based on improved generative network and voice recognition

Abstract

The invention discloses a termite monitoring system and method based on an improved generative network and voice recognition. The method comprises the steps that S1, an original voice signal is converted into a digital voice signal through an analog-to-digital conversion module; s2, feature extraction processing is used for extracting spectrum features of the termite activity sound signals; s3, inputting the preprocessed digital sound signal into an energy-based model for processing; s4, inputting the enhanced termite activity sound signal into the improved generative adversarial network model, and outputting an anomaly detection result; s5, matching the dynamic behavior mode of the termite activity with a preset termite activity database to generate a monitoring result of the termite activity; and S6, the monitoring result is input to an intelligent decision support module, and the intelligent decision support module generates a termite prevention and control suggestion scheme based on the monitoring result. The accuracy and the response speed of the monitoring system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of termite monitoring, and in particular to a termite monitoring system and method based on improved generative networks and sound recognition. Background Art

[0002] With the rapid development of artificial intelligence and Internet of Things technologies, intelligent monitoring systems are being used more and more widely in agriculture and urban environmental management. Termite monitoring, as an important field of environmental protection and building maintenance, has always faced technical difficulties. The sound signals of termite activity are an important basis for identifying and monitoring termite activity. Its monitoring technology can help detect termite damage in a timely manner and provide data support for effective prevention and control.

[0003] At present, termite monitoring mainly relies on the following existing technologies: the first is a method based on manual observation, which determines termite activity by regularly checking the surface features of buildings or crops. Although this method is simple, it is time-consuming, inefficient, and cannot accurately detect hidden termite activities; the second is a method based on chemical induction, which releases chemical agents to trap termites. Although it can improve the detection efficiency to a certain extent, it will cause pollution to the environment and does not meet the requirements of green environmental protection; the third is a method based on traditional acoustic technology, which uses a single sound sensor to collect termite activity sound signals, but this technology is difficult to effectively remove background noise in complex environments, resulting in a high false alarm rate and insufficient accuracy.

[0004] At the same time, the existing technology has the following significant defects in practical applications:

[0005] On the one hand, the sound signals of termite activities are usually low-intensity and highly sparse, and are easily drowned out by environmental noise. Traditional acoustic monitoring technology lacks the ability to effectively process weak signals and cannot accurately extract the characteristic signals of termite activities. On the other hand, existing monitoring systems generally rely on fixed algorithm models, which are difficult to adapt to different environments and changes in termite behavior patterns, and lack flexibility and robustness.

[0006] In summary, the existing termite monitoring technology has significant deficiencies in signal processing, environmental adaptability and intelligent analysis. A new method is urgently needed to solve the above problems in order to achieve efficient, accurate and intelligent monitoring of termite activities. Summary of the invention

[0007] An object of the present invention is to provide a termite monitoring system and method based on improved generative networks and sound recognition, which significantly improves the accuracy and response speed of the monitoring system.

[0008] According to an embodiment of the present invention, a termite monitoring method based on an improved generative network and sound recognition includes the following steps:

[0009] S1, collecting the original sound signals generated by termite activities, and converting the original sound signals into digital sound signals through an analog-to-digital conversion module;

[0010] S2, performing signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the termite activity sound signal, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the termite activity sound signal;

[0011] S3, inputting the pre-processed digital sound signal into the energy-based model for processing, the energy-based model calculates the energy of the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals;

[0012] S4, inputting the enhanced termite activity sound signal into the improved generative adversarial network model, and outputting anomaly detection results;

[0013] S5, matching the dynamic behavior pattern of termite activity with a preset termite activity database to generate monitoring results of termite activity;

[0014] S6. Input the monitoring results into the intelligent decision support module, and the intelligent decision support module generates a termite control suggestion plan based on the monitoring results.

[0015] Optionally, the S1 specifically includes:

[0016] S11. The original sound signal generated by termite activity is collected by a sound sensor. The original sound signal is a time domain continuous signal x(t), where t represents a time variable, and x(t) includes the termite activity sound signal s(t) and the environmental noise signal n(t), satisfying the following relationship:

[0017] x(t)=s(t)+n(t);

[0018] Among them, s(t) is the sound signal generated by the target termite activity, and n(t) is the uncontrollable environmental noise signal;

[0019] S12. Performing analog-to-digital conversion on the collected original sound signal through an analog-to-digital conversion module, converting the time domain continuous signal x(t) into a discrete digital sound signal x[n], where n represents a sampling point index;

[0020] S13. The digital sound signal x[n] obtained after analog-to-digital conversion is stored and classified and marked, wherein the classification mark includes a termite activity sound signal and an ambient noise signal, for distinguishing termite activity sound signals from interference signals;

[0021] S14. Calibrate the sensitivity of the sensor signal during the acquisition process. The dynamic range of the original sound signal x(t) collected under different environmental conditions is adapted to the intensity range of the termite activity sound signal. The dynamic range adjustment parameter is G:

[0022]

[0023] Among them, x max Represents the maximum amplitude of the sound signal in the environment, s max Indicates the maximum amplitude of the sound signal of termite activity.

[0024] Optionally, the S2 specifically includes:

[0025] S21. Perform signal enhancement processing on the collected digital sound signal x[n], and increase the intensity of the termite activity sound signal through dynamic gain to obtain the digital sound signal x after signal enhancement. enh [n];

[0026] S22. The digital sound signal after signal enhancement x enh [n] Perform noise filtering and use adaptive filtering algorithm to separate the termite activity sound signal and the environmental noise signal to obtain the digital sound signal x after noise filtering filtered [n];

[0027] S23. Use short-time Fourier transform to filter the noise of the digital sound signal x filtered [n] Perform feature extraction processing, convert the signal into a time-frequency spectrum representation, and generate the signal's spectrum feature matrix X(k,m):

[0028]

[0029] Among them, k is the frequency index, m is the time frame index, and N represents the number of sampling points per frame.

[0030] Optionally, the S3 specifically includes:

[0031] S31. The pre-processed digital sound signal x filtered [n] is input into the improved energy-based model to construct the energy function E(x) with multi-level sparsity constraints. filtered ; θ), used to calculate the energy of the sound signal of termite activity:

[0032]

[0033] Among them, x filtered [n] is the input noise-filtered digital sound signal, G(z;θ) is the generated signal, W is the weighting matrix used to adjust the spectral sparsity of the signal, ||·|| 1 Represents weighted l1 -norm, ||·|| 0 Indicates l 0 -norm, λ 1 and λ 2 are the sparsity regularization weights, z is the noise distribution, and θ is the model parameter;

[0034] S32. Using the energy function E(x filtered ; θ) is optimized to identify abnormal features in the sound signals of termite activity by minimizing the value of the objective function:

[0035]

[0036] Among them, by adjusting the weight of the generated signal G(z;θ) and the sparsity regularization term, the high sparsity characteristics of the termite activity sound signal are enhanced, while the environmental noise signal is suppressed;

[0037] S33. Output enhanced termite activity sound signal x enhanced [n], the enhanced termite activity sound signal is expressed by the following relationship:

[0038]

[0039] in, is the residual noise component calculated by the energy-based model, x enhanced [n] contains the sparse feature components of the target termite activity sound signal.

[0040] Optionally, the S4 specifically includes:

[0041] S41. The improved generative adversarial network model includes an improved generator and an improved discriminator. The improved generator learns the feature distribution of the termite activity sound signal based on the multimodal feature input to generate a simulated termite activity sound signal G(z; θ g ), the improved discriminator can distinguish the real sound signal from the simulated sound signal, and evaluate the abnormality of the termite activity sound signal. The overall optimization goal of the improved generative adversarial network is:

[0042]

[0043] Where x is the enhanced termite activity sound signal, z is the noise input of the generator, G(z;θ g ) is the simulated termite activity sound signal generated by the generator, θ g is the parameter set of the generator, D(x) is the authenticity score output by the discriminator, which is used to evaluate whether the input signal is a real signal, and p real is the distribution of the real signal, used to train the discriminator, p zis the distribution of noise input, used as the input of the generator, A(x,G(z;θ g )) is an anomaly evaluation function that measures the difference between the true signal x and the generated signal G(z;θ g ) is the difference in feature distribution, λ is the weight of the anomaly assessment regularization term, which is used to balance the optimization objectives of the generation task and the anomaly detection task;

[0044] S42. The improved generator receives the enhanced multimodal features of the termite activity sound signal, including the time-spectral feature, the frequency sparsity feature, and the environmental context parameter ψ, and the improved generator generates a more accurate simulated termite activity sound signal by jointly learning the multimodal features:

[0045] G(z;θ g )=f(z,X(k,m),ψ;θ g );

[0046] Where X(k,m) is the time-frequency feature matrix, which is used to describe the distribution of sound signals in the time and frequency dimensions. k is the frequency index, m is the time frame index, which indicates the distribution of different time frames in the feature matrix, and ψ is the environmental context parameter, which is used to assist the generator in simulating termite sound signals more accurately. f(z,X(k,m),ψ;θ g ) is the generating function of the generator, θ g is the parameter set of the generator;

[0047] S43. Improve the discriminator to perform authenticity discrimination and abnormality evaluation at the same time, and adjust the network parameters θ in combination with the multi-objective optimization mechanism d :

[0048]

[0049] Among them, θ d is the parameter set of the discriminator, which is used to learn the distribution difference between the real signal and the generated signal. R(A(x)) is the regularization term for the anomaly evaluation, which is used to constrain the anomaly evaluation result of the discriminator. γ is the regularization weight coefficient, which is used to adjust the influence of the anomaly detection task on the optimization target of the discriminator.

[0050] S44. When the improved generator generates the simulated termite activity sound signal, it dynamically combines the abnormal feature information fed back by the improved discriminator to adaptively adjust the generation process:

[0051]

[0052] Among them, η is the learning rate of the generator, is the gradient of the generator loss function to its parameters, which is used to guide the generator optimization direction, A(x,G(z;θ g )) is abnormal evaluation feedback;

[0053] S45. The improved discriminator outputs the authenticity score D(x) and the anomaly detection result A(x) of the termite activity sound signal, and compares the anomaly evaluation value with the set threshold τ. When A(x)>τ, the signal is judged to be abnormal, and the weight parameters of the improved generator and the improved discriminator are dynamically adjusted.

[0054] Optionally, the S5 specifically includes:

[0055] S51. Based on the enhanced termite activity sound signal x enhanced [n] The anomaly detection results A(x) generated by the improved generative adversarial network model are used to construct the dynamic behavior pattern of termite activity. The dynamic behavior pattern M(t) includes the spectral feature distribution F(k,m), activity intensity S(t) and spatial diffusion trend R(t) of termite activity at time t.

[0056] S52. Compare the constructed dynamic behavior pattern M(t) with the preset termite activity database D ref Matching, database D ref Store typical termite activity behavior feature template T i , including its corresponding spectral feature distribution F i (k,m), activity intensity range S i and the diffusion range R i , the matching condition is defined as:

[0057]

[0058] Where ΔF=||F(k,m)-F i (k,m)|| 2 represents the difference in the distribution of spectral characteristics, ΔS=|S(t)-S i | represents the difference in activity intensity, ΔR = |R(t)-R i | represents the difference in diffusion range, ∈ F ,∈ S ,∈ R is the set matching threshold;

[0059] S53. Generate monitoring results R of termite activity based on the matching results monitor ,The monitoring results include the real-time location, activity intensity and diffusion range of termite activities;

[0060] S54. Generate the monitoring result R monitor Compare with the actual collected signal characteristics to verify its accuracy and adjust the matching conditions∈ F ,∈ S and ∈ R .

[0061] A termite monitoring system based on improved generative network and sound recognition is used to perform a termite monitoring method based on improved generative network and sound recognition, the system comprising:

[0062] A sound signal collection module is configured to collect original sound signals generated by termite activities and convert the collected original sound signals into digital sound signals through an analog-to-digital conversion module. The sound signal collection module includes a sensor array and a signal conversion device to meet the sound collection requirements under different environmental noise conditions;

[0063] A sound signal processing module is configured to perform signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the sound signal of termite activity, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the sound signal of termite activity to form a signal feature matrix suitable for subsequent analysis;

[0064] An energy-based model processing module is configured to input the pre-processed digital sound signal into the energy-based model for analysis. The energy-based model calculates the energy distribution of the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals;

[0065] An improved generative adversarial network processing module is configured to receive the enhanced termite activity sound signal and input it into an improved generative adversarial network model. The improved generative adversarial network processing module includes an improved generator and an improved discriminator. The improved generator generates a simulated signal by learning the characteristic distribution of the termite activity sound signal. The improved discriminator distinguishes between the real signal and the simulated signal, evaluates the degree of abnormality, and outputs an abnormality detection result.

[0066] A dynamic behavior pattern construction and matching module is configured to construct a dynamic behavior pattern of termite activity based on the enhanced termite activity sound signal and the abnormality detection result, wherein the dynamic behavior pattern includes the frequency spectrum characteristic distribution, activity intensity and spatial diffusion trend of termite activity, and matches it with a preset termite activity database to generate monitoring results of termite activity, including real-time location, activity intensity and diffusion range;

[0067] The intelligent decision support module is configured to receive the monitoring results and generate a termite control suggestion plan based on the monitoring results.

[0068] The beneficial effects of the present invention are:

[0069] (1) The present invention improves the generative adversarial network model by inputting multimodal features into the generator, thereby achieving high-fidelity modeling of termite activity sound signals. The improved discriminator adopts a dual optimization mechanism, combines authenticity discrimination and abnormality evaluation, effectively distinguishes termite activity sound signals from environmental noise, and accurately detects abnormal signals. By introducing a multi-objective optimization algorithm and anomaly feature feedback mechanism, the authenticity of the generator's output signal and the sensitivity of the discriminator's anomaly detection are significantly improved.

[0070] (2) The present invention introduces multi-level sparsity constraints into the energy-based model. The regularization method that combines spectral sparsity and time-domain sparsity significantly improves the feature extraction capability of termite activity sound signals. The model can automatically adjust the sparsity weight through the energy function, optimize the target features in the signal, and suppress the interference of noise components. It shows stronger robustness when processing low-intensity and high-sparseness signals.

[0071] (3) The present invention constructs a dynamic behavior pattern of termite activity to dynamically associate the spectral feature distribution, activity intensity and diffusion trend of termite activity sound signals, thereby achieving accurate modeling of termite activity behavior. At the same time, in combination with a preset termite activity database, a multi-dimensional matching algorithm is used to perform template matching on real-time monitoring signals to quickly generate monitoring results of termite activity, including real-time location, activity intensity and diffusion range. Dynamic behavior modeling and prediction can be achieved based on real-time data, significantly improving the accuracy and response speed of the monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0073] Figure 1 This is a flow chart of a termite monitoring system and method based on improved generative network and sound recognition proposed by the present invention. DETAILED DESCRIPTION

[0074] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0075] refer to Figure 1 , a termite monitoring method based on improved generative network and sound recognition, comprising the following steps:

[0076] S1, collecting the original sound signals generated by termite activities, and converting the original sound signals into digital sound signals through an analog-to-digital conversion module;

[0077] S2, performing signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the termite activity sound signal, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the termite activity sound signal;

[0078] S3, inputting the pre-processed digital sound signal into the energy-based model for processing, the energy-based model performs energy calculation on the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals;

[0079] S4, inputting the enhanced termite activity sound signal into the improved generative adversarial network model, and outputting anomaly detection results;

[0080] S5, matching the dynamic behavior pattern of termite activity with a preset termite activity database to generate monitoring results of termite activity;

[0081] S6. Input the monitoring results into the intelligent decision support module, and the intelligent decision support module generates a termite control suggestion plan based on the monitoring results.

[0082] In this implementation, S1 specifically includes:

[0083] S11. The original sound signal generated by termite activity is collected by a sound sensor. The original sound signal is a time domain continuous signal x(t), where t represents the time variable. x(t) includes the termite activity sound signal s(t) and the environmental noise signal n(t), satisfying the following relationship:

[0084] x(t)=s(t)+n(t);

[0085] Among them, s(t) is the sound signal generated by the target termite activity, and n(t) is the uncontrollable environmental noise signal;

[0086] S12. Performing analog-to-digital conversion on the collected original sound signal through an analog-to-digital conversion module, converting the time domain continuous signal x(t) into a discrete digital sound signal x[n], where n represents a sampling point index;

[0087] S13. The digital sound signal x[n] obtained after analog-to-digital conversion is stored and classified and marked, the classification mark includes a termite activity sound signal and an ambient noise signal, which is used to distinguish between termite activity sound signals and interference signals;

[0088] S14. Calibrate the sensitivity of the sensor signal during the acquisition process. The dynamic range of the original sound signal x(t) collected under different environmental conditions is adapted to the intensity range of the termite activity sound signal. The dynamic range adjustment parameter is G:

[0089]

[0090] Among them, x max Represents the maximum amplitude of the sound signal in the environment, s max Indicates the maximum amplitude of the sound signal of termite activity.

[0091] In this implementation, S2 specifically includes:

[0092] S21. Perform signal enhancement processing on the collected digital sound signal x[n], and increase the intensity of the termite activity sound signal through dynamic gain to obtain the digital sound signal x after signal enhancement. enh [n];

[0093] S22. The digital sound signal after signal enhancement x enh [n] Perform noise filtering and use an adaptive filtering algorithm to separate the termite activity sound signal and the environmental noise signal to obtain the digital sound signal x after noise filtering. filtered [n];

[0094] S23. Use short-time Fourier transform to filter the noise of the digital sound signal x filtered [n] Perform feature extraction processing, convert the signal into a time-frequency spectrum representation, and generate the signal's spectrum feature matrix X(k,m):

[0095]

[0096] Among them, k is the frequency index, m is the time frame index, and N represents the number of sampling points per frame.

[0097] In this implementation, S3 specifically includes:

[0098] S31. The pre-processed digital sound signal x filtered [n] is input into the improved energy-based model to construct the energy function E(x) with multi-level sparsity constraints. filtered ; θ), used to calculate the energy of the sound signal of termite activity:

[0099]

[0100] Among them, x filtered [n] is the input noise-filtered digital sound signal, G(z;θ) is the generated signal, W is the weighting matrix used to adjust the spectral sparsity of the signal, ||·|| 1 Represents weighted l 1 -norm, ||·|| 0 Indicates l 0 -norm, λ 1 and λ 2are the sparsity regularization weights, z is the noise distribution, and θ is the model parameter;

[0101] S32. Using the energy function E(x filtered ; θ) is optimized to identify abnormal features in the sound signals of termite activity by minimizing the value of the objective function:

[0102]

[0103] Among them, by adjusting the weights of the generated signal G(z;θ) and the sparsity regularization term, the high sparsity characteristics of the termite activity sound signal are enhanced, while the environmental noise signal is suppressed;

[0104] S33. Output enhanced termite activity sound signal x enhanced [n], the enhanced termite activity sound signal is expressed by the following relationship:

[0105]

[0106] in, is the residual noise component calculated by the energy-based model, x enhanced [n] contains the sparse feature components of the target termite activity sound signal.

[0107] In this implementation, S4 specifically includes:

[0108] S41. The improved generative adversarial network model includes an improved generator and an improved discriminator. The improved generator learns the feature distribution of the termite activity sound signal based on the multimodal feature input to generate a simulated termite activity sound signal G(z; θ g ), the improved discriminator can distinguish the real sound signal from the simulated sound signal, and evaluate the abnormality of the termite activity sound signal. The overall optimization goal of the improved generative adversarial network is:

[0109]

[0110] Where x is the enhanced termite activity sound signal, z is the noise input of the generator, G(z;θ g ) is the simulated termite activity sound signal generated by the generator, θ g is the parameter set of the generator, D(x) is the authenticity score output by the discriminator, which is used to evaluate whether the input signal is a real signal, and p real is the distribution of the real signal, used to train the discriminator, p z is the distribution of noise input, used as the input of the generator, A(x,G(z;θ g )) is an anomaly evaluation function that measures the difference between the true signal x and the generated signal G(z;θ g) is the difference in feature distribution, λ is the weight of the anomaly assessment regularization term, which is used to balance the optimization objectives of the generation task and the anomaly detection task;

[0111] S42. The improved generator receives the enhanced multimodal features of the termite activity sound signal, including the time-spectral feature, the frequency sparsity feature, and the environmental context parameter ψ, and the improved generator generates a more accurate simulated termite activity sound signal by jointly learning the multimodal features:

[0112] G(z;θ g )=f(z,X(k,m),ψ;θ g );

[0113] Where X(k,m) is the time-frequency feature matrix, which is used to describe the distribution of sound signals in the time and frequency dimensions. k is the frequency index, m is the time frame index, which indicates the distribution of different time frames in the feature matrix, and ψ is the environmental context parameter, which is used to assist the generator in simulating termite sound signals more accurately. f(z,X(k,m),ψ;θ g ) is the generating function of the generator, θ g is the parameter set of the generator;

[0114] S43. Improve the discriminator to perform authenticity discrimination and abnormality evaluation at the same time, and adjust the network parameters θ in combination with the multi-objective optimization mechanism d :

[0115]

[0116] Among them, θ d is the parameter set of the discriminator, which is used to learn the distribution difference between the real signal and the generated signal. R(A(x)) is the regularization term for the anomaly evaluation, which is used to constrain the anomaly evaluation result of the discriminator. γ is the regularization weight coefficient, which is used to adjust the influence of the anomaly detection task on the optimization target of the discriminator.

[0117] S44. When the improved generator generates the simulated termite activity sound signal, it dynamically combines the abnormal feature information fed back by the improved discriminator to adaptively adjust the generation process:

[0118]

[0119] Among them, η is the learning rate of the generator, is the gradient of the generator loss function to its parameters, which is used to guide the generator optimization direction, A(x,G(z;θ g )) is abnormal evaluation feedback;

[0120] S45. The improved discriminator outputs the authenticity score D(x) and the anomaly detection result A(x) of the termite activity sound signal, and compares the anomaly evaluation value with the set threshold τ. When A(x)>τ, the signal is judged to be abnormal, and the weight parameters of the improved generator and the improved discriminator are dynamically adjusted.

[0121] In this implementation, S5 specifically includes:

[0122] S51. Based on the enhanced termite activity sound signal x enhanced [n] The anomaly detection results A(x) generated by the improved generative adversarial network model are used to construct the dynamic behavior pattern of termite activity. The dynamic behavior pattern M(t) includes the spectral feature distribution F(k,m), activity intensity S(t) and spatial diffusion trend R(t) of termite activity at time t.

[0123] S52. Compare the constructed dynamic behavior pattern M(t) with the preset termite activity database D ref Matching, database D ref Store typical termite activity behavior feature template T i , including its corresponding spectral feature distribution F i (k,m), activity intensity range S i and the diffusion range R i , the matching condition is defined as:

[0124]

[0125] Where ΔF=||F(k,m)-F i (k,m)|| 2 represents the difference in the distribution of spectral characteristics, ΔS = |S(t)-S i | represents the difference in activity intensity, ΔR = |R(t)-R i | represents the difference in diffusion range, ∈ F ,∈ S ,∈ R is the set matching threshold;

[0126] S53. Generate monitoring results R of termite activity based on the matching results monitor ,The monitoring results include the real-time location, activity intensity and diffusion range of termite activities;

[0127] S54. Generate the monitoring result R monitor Compare with the actual collected signal characteristics to verify its accuracy and adjust the matching conditions∈ F ,∈ S and ∈ R .

[0128] A termite monitoring system based on improved generative network and sound recognition is used to perform a termite monitoring method based on improved generative network and sound recognition. The system includes:

[0129] A sound signal collection module is configured to collect original sound signals generated by termite activities and convert the collected original sound signals into digital sound signals through an analog-to-digital conversion module. The sound signal collection module includes a sensor array and a signal conversion device to meet the sound collection requirements under different environmental noise conditions;

[0130] A sound signal processing module is configured to perform signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the sound signal of termite activity, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the sound signal of termite activity to form a signal feature matrix suitable for subsequent analysis;

[0131] An energy-based model processing module is configured to input the pre-processed digital sound signal into the energy-based model for analysis. The energy-based model calculates the energy distribution of the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals;

[0132] An improved generative adversarial network processing module is configured to receive the enhanced termite activity sound signal and input it into an improved generative adversarial network model. The improved generative adversarial network processing module includes an improved generator and an improved discriminator. The improved generator generates a simulated signal by learning the characteristic distribution of the termite activity sound signal. The improved discriminator distinguishes between the real signal and the simulated signal, evaluates the degree of abnormality, and outputs an abnormality detection result.

[0133] A dynamic behavior pattern construction and matching module is configured to construct a dynamic behavior pattern of termite activity based on the enhanced termite activity sound signal and the abnormal detection result. The dynamic behavior pattern includes the frequency spectrum characteristic distribution, activity intensity and spatial diffusion trend of termite activity, and matches it with a preset termite activity database to generate monitoring results of termite activity, including real-time location, activity intensity and diffusion range;

[0134] The intelligent decision support module is configured to receive the monitoring results and generate a termite control suggestion plan based on the monitoring results.

[0135] Embodiment 1:

[0136] Example In June 2024, a historical building management center received a report that a century-old wooden structure in a humid southern region was suspected of being infested by termites. The building management party hoped to avoid destructive inspections and, at the same time, quickly and accurately locate the termite activity areas to formulate targeted prevention and control measures. For this reason, the building management center decided to use the termite monitoring system of the present invention for a comprehensive inspection.

[0137] The system was deployed at 22:00 on the night of June 15, 2024. Fifty high-sensitivity sound sensors were arranged at the main wooden structure parts of the building, including the floor, columns, beams, and walls. After the system was started, it collected the termite activity sound signals inside the building and converted the original sound signals into digital sound signals through an analog-to-digital conversion module.

[0138] At 1:37 am on June 16, 2024, the system detected a sound source under the floor in the northwest corner. The signal anomaly score gradually increased, jumping from 0.3 of the normal value to 2.1, exceeding the set anomaly threshold of 1.5. System analysis showed that the characteristic frequency of the sound source was 4.8 kHz, and the amplitude was about 3.2 dB, which highly coincided with the typical termite activity signal.

[0139] Through energy-based model analysis, the sound signals in this area showed obvious sparsity characteristics in the high-frequency band, and the noise components in the low-frequency band were effectively removed. The enhanced signals showed that there was obvious termite activity in this area. The generative adversarial network was further used to confirm that the signal obtained a credibility score of 95% in the authenticity discrimination, and the anomaly evaluation value was 2.7, much higher than the average value of 0.9. The source of the abnormal signal was confirmed to be termite activity.

[0140] The system constructed the dynamic behavior pattern of this area. The data showed that from 2:00 am to 4:00 am, the termite activity intensity under the floor in the northwest corner increased from 3.2 dB to 5.6 dB, and the activity range expanded from the center of the floor to the surrounding by about 1.8 square meters. At the same time, the system detected that weak abnormal sound signals also began to appear at the bottom of another adjacent column, with an activity intensity of 2.1 dB. By matching with the preset database, the system confirmed that the activity characteristics conform to the typical behavior pattern of termite colonies spreading from a single point.

[0141] Finally, the system generated a monitoring report, including the following key information:

[0142] Abnormal area: The center point under the floor in the northwest corner, with an area of 2.8 square meters, spreading to the bottom of the adjacent column;

[0143] Activity intensity: Increased from 3.2 dB to 5.6 dB;

[0144] Diffusion trend: Extending to the south, with an estimated potential diffusion range of 6 square meters;

[0145] Time node: The anomaly began at 1:37 am on June 16, and the peak intensity occurred at 4:12.

[0146] After receiving the system report, the management immediately dispatched technicians to the abnormal area for verification. Without dismantling the structure, the technicians used a micro camera probe to explore under the floor and found a termite colony nest with an area of ​​about 0.4 square meters. The activity area of ​​the nest was highly consistent with the system monitoring results. Further testing confirmed that there were preliminary signs of excavation of termite tunnels at the bottom of the column.

[0147] In order to verify the effectiveness of the present invention, the management party simultaneously used traditional sound detection technology to monitor the same area. The comparison results are shown in Table 1 below:

[0148] Table 1 Comparison data between the termite monitoring method based on improved generative network and sound recognition and the traditional method

[0149]

[0150] Through the system monitoring of the present invention, the building successfully discovered and located the termite activity area at an early stage, significantly reduced the risk of spread, and avoided potential structural damage. Compared with traditional methods, the termite monitoring method of the present invention shows higher real-time, accuracy and reliability, providing important technical support for pest control in similar scenarios.

[0151] The present invention improves the generative adversarial network model by inputting multimodal features into the generator, thereby achieving high-fidelity modeling of termite activity sound signals. The improved discriminator adopts a dual optimization mechanism, combines authenticity discrimination and abnormality assessment, effectively distinguishes termite activity sound signals from environmental noise, and accurately detects abnormal signals. By introducing a multi-objective optimization algorithm and an abnormal feature feedback mechanism, the authenticity of the generator's output signal and the sensitivity of the discriminator's abnormality detection are significantly improved.

[0152] The present invention introduces multi-level sparsity constraints into the energy-based model, and significantly improves the feature extraction capability of termite activity sound signals through a regularization method that combines spectral sparsity and time-domain sparsity. The model can automatically adjust the sparsity weight through the energy function, optimize the target features in the signal, and suppress the interference of noise components, showing stronger robustness when processing low-intensity and high-sparseness signals.

[0153] The present invention dynamically associates the spectral feature distribution, activity intensity and diffusion trend of termite activity sound signals by constructing a dynamic behavior pattern of termite activity, thereby realizing accurate modeling of termite activity behavior. At the same time, in combination with a preset termite activity database, a multi-dimensional matching algorithm is used to perform template matching on real-time monitoring signals to quickly generate monitoring results of termite activity, including real-time position, activity intensity and diffusion range. Dynamic behavior modeling and prediction can be realized based on real-time data, thereby significantly improving the accuracy and response speed of the monitoring system.

[0154] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A termite monitoring method based on improved generative network and sound recognition, characterized in that: The steps include: S1, collecting the original sound signals generated by termite activities, and converting the original sound signals into digital sound signals through an analog-to-digital conversion module; S2, performing signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the termite activity sound signal, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the termite activity sound signal; S3, inputting the pre-processed digital sound signal into the energy-based model for processing, the energy-based model performs energy calculation on the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals; S4, inputting the enhanced termite activity sound signal into the improved generative adversarial network model, and outputting anomaly detection results; S5, matching the dynamic behavior pattern of termite activity with a preset termite activity database to generate monitoring results of termite activity; S6. Input the monitoring results into the intelligent decision support module, and the intelligent decision support module generates a termite control suggestion plan based on the monitoring results.

2. A termite monitoring method based on improved generative network and sound recognition according to claim 1, characterized in that: The S1 specifically includes: S11. The original sound signal generated by termite activity is collected by a sound sensor. The original sound signal is a time domain continuous signal x(t), where t represents a time variable, and x(t) includes the termite activity sound signal s(t) and the environmental noise signal n(t), satisfying the following relationship: x(t)=s(t)+n(t); Among them, s(t) is the sound signal generated by the target termite activity, and n(t) is the uncontrollable environmental noise signal; S12. Performing analog-to-digital conversion on the collected original sound signal through an analog-to-digital conversion module, converting the time domain continuous signal x(t) into a discrete digital sound signal x[n], where n represents a sampling point index; S13. The digital sound signal x[n] obtained after analog-to-digital conversion is stored and classified and marked, wherein the classification mark includes a termite activity sound signal and an ambient noise signal, for distinguishing termite activity sound signals from interference signals; S14. Calibrate the sensitivity of the sensor signal during the acquisition process. The dynamic range of the original sound signal x(t) collected under different environmental conditions is adapted to the intensity range of the termite activity sound signal. The dynamic range adjustment parameter is G: Among them, x max Represents the maximum amplitude of the sound signal in the environment, s max Indicates the maximum amplitude of the sound signal of termite activity.

3. A termite monitoring method based on improved generative network and sound recognition according to claim 1, characterized in that: The S2 specifically includes: S21. Perform signal enhancement processing on the collected digital sound signal x[n], and increase the intensity of the termite activity sound signal through dynamic gain to obtain the digital sound signal x after signal enhancement. enh [n]; S22. The digital sound signal after signal enhancement x enh [n] Perform noise filtering and use adaptive filtering algorithm to separate the termite activity sound signal and the environmental noise signal to obtain the digital sound signal x after noise filtering filtered [n]; S23. Use short-time Fourier transform to filter the noise of the digital sound signal x filtered [n] Perform feature extraction processing, convert the signal into a time-frequency spectrum representation, and generate the signal's spectrum feature matrix X(k,m): Among them, k is the frequency index, m is the time frame index, and N represents the number of sampling points per frame.

4. The termite monitoring method based on improved generative network and sound recognition according to claim 1, characterized in that: The S3 specifically includes: S31. The pre-processed digital sound signal x filtered [n] is input into the improved energy-based model to construct the energy function E(x) with multi-level sparsity constraints. filtered ; θ), used to calculate the energy of the sound signal of termite activity: Among them, x filtered [n] is the input noise-filtered digital sound signal, G(z;θ) is the generated signal, W is the weighting matrix used to adjust the spectral sparsity of the signal, ||·||1 represents the weighted l1-norm, ||·||0 represents the l0-norm, λ1 and λ2 are the sparsity regularization weights, z is the noise distribution, and θ is the model parameter; S32. Using the energy function E(x filtered ; θ) is optimized to identify abnormal features in the sound signals of termite activity by minimizing the value of the objective function: Among them, by adjusting the weights of the generated signal G(z;θ) and the sparsity regularization term, the high sparsity characteristics of the termite activity sound signal are enhanced, while the environmental noise signal is suppressed; S33. Output enhanced termite activity sound signal x enhanced [n], the enhanced termite activity sound signal is expressed by the following relationship: in, is the residual noise component calculated by the energy-based model, x enhanced [n] contains the sparse feature components of the target termite activity sound signal.

5. The termite monitoring method based on improved generative network and sound recognition according to claim 1, characterized in that: The S4 specifically includes: S41. The improved generative adversarial network model includes an improved generator and an improved discriminator. The improved generator learns the feature distribution of the termite activity sound signal based on the multimodal feature input to generate a simulated termite activity sound signal G(z; θ g ), the improved discriminator can distinguish the real sound signal from the simulated sound signal, and evaluate the abnormality of the termite activity sound signal. The overall optimization goal of the improved generative adversarial network is: Where x is the enhanced termite activity sound signal, z is the noise input of the generator, G(z;θ g ) is the simulated termite activity sound signal generated by the generator, θ g is the parameter set of the generator, D(x) is the authenticity score output by the discriminator, which is used to evaluate whether the input signal is a real signal, and p real is the distribution of the real signal, used to train the discriminator, p z is the distribution of noise input, used as the input of the generator, A(x,G(z;θ g )) is an anomaly evaluation function that measures the difference between the true signal x and the generated signal G(z;θ g ) is the difference in feature distribution, λ is the weight of the anomaly assessment regularization term, which is used to balance the optimization objectives of the generation task and the anomaly detection task; S42. The improved generator receives the enhanced multimodal features of the termite activity sound signal, including the time-spectral feature, the frequency sparsity feature, and the environmental context parameter ψ, and the improved generator generates a more accurate simulated termite activity sound signal by jointly learning the multimodal features: G(z;θ g )=f(z,X(k,m),ψ;θ g ); Where X(k,m) is the time-frequency feature matrix, which is used to describe the distribution of sound signals in the time and frequency dimensions. k is the frequency index, m is the time frame index, which indicates the distribution of different time frames in the feature matrix, and ψ is the environmental context parameter, which is used to assist the generator in simulating termite sound signals more accurately. f(z,X(k,m),ψ;θ g ) is the generating function of the generator, θ g is the parameter set of the generator; S43. Improve the discriminator to perform authenticity discrimination and abnormality evaluation at the same time, and adjust the network parameters θ in combination with the multi-objective optimization mechanism d : Among them, θ d is the parameter set of the discriminator, which is used to learn the distribution difference between the real signal and the generated signal. R(A(x)) is the regularization term for the anomaly evaluation, which is used to constrain the anomaly evaluation result of the discriminator. γ is the regularization weight coefficient, which is used to adjust the influence of the anomaly detection task on the optimization target of the discriminator. S44. When the improved generator generates the simulated termite activity sound signal, it dynamically combines the abnormal feature information fed back by the improved discriminator to adaptively adjust the generation process: Among them, η is the learning rate of the generator, is the gradient of the generator loss function to its parameters, which is used to guide the generator optimization direction, A(x,G(z;θ g )) is abnormal evaluation feedback; S45. The improved discriminator outputs the authenticity score D(x) and the anomaly detection result A(x) of the termite activity sound signal, and compares the anomaly evaluation value with the set threshold τ. When A(x)>τ, the signal is judged to be abnormal, and the weight parameters of the improved generator and the improved discriminator are dynamically adjusted.

6. The termite monitoring method based on improved generative network and sound recognition according to claim 1, characterized in that: The S5 specifically includes: S51. Based on the enhanced termite activity sound signal x enhanced [n] The anomaly detection results A(x) generated by the improved generative adversarial network model are used to construct the dynamic behavior pattern of termite activity. The dynamic behavior pattern M(t) includes the spectral feature distribution F(k,m), activity intensity S(t) and spatial diffusion trend R(t) of termite activity at time t. S52. Compare the constructed dynamic behavior pattern M(t) with the preset termite activity database D ref Matching, database D ref Store typical termite activity behavior feature template T i , including its corresponding spectral feature distribution F i (k,m), activity intensity range S i and the diffusion range R i , the matching condition is defined as: Where ΔF=||F(k,m)-F i (k,m)||2 represents the difference in the distribution of spectral characteristics, ΔS=|S(t)-S i | represents the difference in activity intensity, ΔR = |R(t)-R i | represents the difference in diffusion range, ∈ F ,∈ S ,∈ R is the set matching threshold; S53. Generate monitoring results R of termite activity based on the matching results monitor ,The monitoring results include the real-time location, activity intensity and diffusion range of termite activities; S54. Generate the monitoring result R monitor Compare with the actual collected signal characteristics to verify its accuracy and adjust the matching conditions∈ F ,∈ S and ∈ R .

7. A termite monitoring system based on improved generative network and sound recognition, used to execute a termite monitoring method based on improved generative network and sound recognition as claimed in any one of claims 1 to 6, characterized in that: The system comprises: A sound signal collection module is configured to collect original sound signals generated by termite activities and convert the collected original sound signals into digital sound signals through an analog-to-digital conversion module. The sound signal collection module includes a sensor array and a signal conversion device to meet the sound collection requirements under different environmental noise conditions; A sound signal processing module is configured to perform signal enhancement processing, noise filtering processing and feature extraction processing on the collected digital sound signal, wherein the signal enhancement processing is used to enhance the intensity of the sound signal of termite activity, the noise filtering processing is used to remove the environmental noise signal, and the feature extraction processing is used to extract the frequency spectrum characteristics of the sound signal of termite activity to form a signal feature matrix suitable for subsequent analysis; An energy-based model processing module is configured to input the pre-processed digital sound signal into the energy-based model for analysis. The energy-based model calculates the energy distribution of the digital sound signal through an energy function, identifies abnormal sound signals and outputs enhanced termite activity sound signals; An improved generative adversarial network processing module is configured to receive the enhanced termite activity sound signal and input it into an improved generative adversarial network model. The improved generative adversarial network processing module includes an improved generator and an improved discriminator. The improved generator generates a simulated signal by learning the characteristic distribution of the termite activity sound signal. The improved discriminator distinguishes between the real signal and the simulated signal, evaluates the degree of abnormality, and outputs an abnormality detection result. A dynamic behavior pattern construction and matching module is configured to construct a dynamic behavior pattern of termite activity based on the enhanced termite activity sound signal and the abnormality detection result, wherein the dynamic behavior pattern includes the frequency spectrum characteristic distribution, activity intensity and spatial diffusion trend of termite activity, and matches it with a preset termite activity database to generate monitoring results of termite activity, including real-time location, activity intensity and diffusion range; The intelligent decision support module is configured to receive the monitoring results and generate a termite control suggestion plan based on the monitoring results.

Images