A dam termite damage real-time monitoring and early warning method and system and a storage medium

Abstract

This invention relates to the field of dam safety monitoring technology, and discloses a method, system, and storage medium for real-time monitoring and early warning of ant infestations on dams. The method includes: constructing a dynamic noise feature space and a coupled feature model; real-time acquisition of acoustic and vibration signals and multi-dimensional environmental factor data; calling the noise space based on environmental data to adaptively filter the acoustic and vibration signals; using the coupled model to analyze the denoised signal and environmental data to obtain a preliminary risk level; selectively initiating active acoustic probing for physical verification based on this level, or performing inverse optimization of the noise model; and finally updating the risk level and outputting an early warning based on the verification results. This invention, by combining adaptive environmental filtering with active verification, solves the problems of low monitoring accuracy and poor reliability in existing technologies due to the inability to suppress strong time-varying noise and the lack of reliable verification methods for suspected signals, achieving high-precision and high-reliability intelligent monitoring and early warning of ant infestations on dams.

Description

Technical Field

[0001] This invention relates to the field of dam safety monitoring technology, specifically to a method, system, and storage medium for real-time monitoring and early warning of ant infestations on dams. Background Technology

[0002] Dikes, especially earth-rock dams, are crucial water conservancy infrastructure, and their structural safety directly affects the safety of life and property downstream. Termites are one of the main biological threats to dike safety. They bore into the dike interior, build nests, and create tunnels, forming penetrating seepage channels. In severe cases, this can lead to major risks such as piping, collapse, or even breaching of the dike. Therefore, long-term, effective, and non-destructive real-time monitoring and early warning of termite infestations on dikes is a key technical challenge in ensuring dike safety.

[0003] Acoustic vibration monitoring, as a non-destructive monitoring method, provides a feasible technical approach for early warning of termite infestations by deploying sensors within dams to capture the weak acoustic vibration signals generated by termite activity (such as feeding and movement). However, this method faces severe challenges in practical applications. The outdoor environment where dams are located is extremely complex, with strong background noise generated by wind, rain, water flow, surrounding traffic, and human activities. The intensity of these noises is much greater than that of termite activity signals, and their spectral characteristics and statistical regularities change dynamically with environmental factors (such as weather, temperature, and soil moisture), exhibiting strong time-varying and uncertainties.

[0004] Existing technologies often employ fixed bandpass filters or simple adaptive filtering algorithms when dealing with strong time-varying background noise. These methods struggle to track and adapt to complex environmental noise changes, resulting in unsatisfactory noise reduction. Weak ant infestation signals are frequently masked by residual noise, leading to numerous missed detections. Furthermore, certain sudden and rare environmental noise events (such as small animal activity or the movement of internal soil particles) may have signal characteristics similar to ant infestation signals. Existing technologies, lacking comprehensive perception and analysis capabilities of environmental conditions, struggle to effectively distinguish between them, resulting in high false alarm rates. This increases unnecessary manual verification costs and reduces the reliability of the monitoring system. Moreover, for suspected signals with low confidence levels, existing technologies lack effective, non-destructive secondary verification methods, creating a dilemma in risk assessment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method, system, and storage medium for real-time monitoring and early warning of ant infestations on dikes. This solves the problems of low accuracy and poor reliability of ant infestation early warning in traditional acoustic monitoring methods, which are unable to effectively suppress strong time-varying environmental noise and lack reliable verification methods for suspected signals.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a real-time monitoring and early warning method for ant infestations on dikes. This method establishes a complete monitoring and early warning process from environmental perception, adaptive filtering, coupling analysis to closed-loop response. The method includes:

[0007] First, an offline dynamic noise feature space is constructed for subsequent real-time monitoring, and an acoustic vibration spectrum environmental factor coupling feature model is constructed to comprehensively analyze acoustic vibration signals and environmental factors.

[0008] During the monitoring process, the system simultaneously collects real-time acoustic and vibration signals inside the dam and multi-dimensional environmental factor data reflecting the current environmental conditions, and performs necessary preprocessing on these raw data.

[0009] Subsequently, the core of the method lies in using real-time multidimensional environmental factor data to accurately call or synthesize a background noise power spectrum estimate that highly matches the current environment from a pre-constructed dynamic noise feature space. Based on this estimate, efficient environmental factor adaptive filtering is performed on the pre-processed acoustic and vibration signal to retain potential ant damage signals to the maximum extent while filtering out environmental background noise, thereby obtaining a denoised target signal.

[0010] Subsequently, the method utilizes a pre-trained acoustic and vibration spectrum environmental factor coupling feature model to perform in-depth coupling analysis on the target signal and the multi-dimensional environmental factor data synchronized with it, thereby obtaining a preliminary, quantitative ant infestation risk level.

[0011] Finally, for different initial risk levels, the method initiates an intelligent adaptive response and verification mechanism. For risks determined to have high confidence, the system triggers inverse optimization of the dynamic noise model, using the high-confidence event to improve the accuracy of the noise model. For risks determined to be suspected, active acoustic probing is initiated to verify the suspected area through physical detection. The system ultimately updates the ant infestation risk level based on the verification results and generates and outputs highly reliable early warning information.

[0012] In a preferred embodiment, to more accurately estimate background noise, instead of simply matching a single noise template when calling the dynamic noise feature space, multiple nearest-neighbor discrete environmental state units corresponding to the current real-time multidimensional environmental factor data are searched in the factor space, and multiple noise templates corresponding to these units are retrieved. Subsequently, these multiple noise templates are dynamically weighted and fused using a weighted interpolation function based on the distance between the real-time multidimensional environmental factor data and the center points of these units. This process can synthesize a background noise power spectrum estimate that more accurately reflects the current mixed environmental state than any single pre-stored template in real time, significantly improving the performance of adaptive filtering.

[0013] In a preferred embodiment, to improve the accuracy of risk assessment, during coupling analysis, the system first extracts a set of key acoustic and vibrational features (such as spectral centroid, spectral entropy, and key frequency band energy percentage) from the denoised target signal. Then, these acoustic and vibrational features are cascaded with synchronously acquired multidimensional environmental factor data at the feature level to construct a higher-dimensional, more information-rich coupled feature vector. Inputting this coupled feature vector into the model for inference enables the model to fully consider the current environmental conditions when making decisions, thereby more accurately distinguishing between termite activity and noise interference in different environments.

[0014] In a preferred embodiment, the active acoustic detection and verification mechanism is a key innovation of this invention in reducing the false alarm rate. This mechanism is triggered when the system outputs a preliminary risk level of "suspected risk." The system controls a sensor node in the monitoring network as a transmitting source node to emit a series of coded acoustic signals with excellent autocorrelation characteristics into the dam medium. Simultaneously, one or more nearby receiving source nodes are responsible for receiving the response signals after penetrating or diffracting the suspected risk area. By performing matched filtering on the response signals, the actual propagation delay and amplitude of the coded acoustic signals can be accurately obtained. Since termite nests can cause abnormal physical properties of the medium (such as cavities or looseness), altering the sound wave propagation path or energy, the measured propagation delay and amplitude can be compared with pre-calibrated health status benchmarks to determine whether there are medium anomalies caused by termite infestation in the area. If an anomaly is detected, the risk level is upgraded to high risk; otherwise, it is determined to be an environmental disturbance, and the risk level is downgraded to safe.

[0015] In a preferred embodiment, the dynamic noise model inverse optimization mechanism enables the invention to possess self-learning and adaptive evolution capabilities. When the initial risk level output by the system is "high confidence risk," the system has sufficient reason to believe that the power spectrum of the currently denoised target signal is the actual termite activity signal component. Therefore, the system will extract this "actual signal component" from the power spectrum of the preprocessed original acoustic vibration signal, and the remaining part can be regarded as a highly accurate estimate of the real background noise in the current environment. The system then uses this high-precision estimate of the real background noise to incrementally update the noise template corresponding to the current real-time multidimensional environmental factor data in the dynamic noise feature space, thereby enabling the noise model to continuously iterate and optimize during operation.

[0016] A second aspect of the present invention provides a real-time monitoring and early warning system for ant infestations on dikes, the system being used to implement the aforementioned method, characterized in that it includes:

[0017] The model initialization module is responsible for training the coupled feature model of acoustic and vibration spectrum environmental factors and constructing a dynamic noise feature space for real-time noise reduction, whether the system is deployed or offline.

[0018] The data acquisition and preprocessing module is responsible for acquiring the acoustic and vibration signals and multi-dimensional environmental factor data of the dam in real time and synchronously at the monitoring site.

[0019] The environmental factor adaptive filtering module receives real-time data, and based on multi-dimensional environmental factor data, calls the dynamic noise feature space to obtain background noise estimation, performs adaptive spectral subtraction and noise reduction on the acoustic and vibration signals, and outputs the denoised target signal.

[0020] The coupling analysis and preliminary risk assessment module extracts the acoustic and vibration characteristics of the target signal, couples it with environmental data, and uses the coupling characteristic model to perform inference and output a preliminary ant infestation risk level.

[0021] The adaptive response and early warning generation module serves as the system's decision-making and execution center. Based on the initial risk level, it schedules and executes adaptive responses such as reverse optimization or proactive investigation, and generates and issues early warning information based on the final, verified risk level.

[0022] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the real-time monitoring and early warning method for ant infestation on dikes as described in any of the preceding claims.

[0023] This invention provides a method, system, and storage medium for real-time monitoring and early warning of ant infestations on dikes. It offers the following advantages:

[0024] 1. This invention achieves accurate filtering of complex time-varying noise by constructing a dynamic noise feature space and synthesizing precise background noise power spectrum estimates in real time using multi-dimensional environmental factor data. This provides high-quality target signals for subsequent risk analysis, fundamentally reducing false alarms and missed alarms caused by noise interference, and significantly improving the accuracy of preliminary risk assessment.

[0025] 2. This invention introduces an active acoustic detection and verification mechanism. When a suspected risk is detected, the system actively emits coded sound waves and analyzes their propagation characteristics to perform secondary confirmation of the medium anomaly through physical detection. This effectively eliminates interference from ambiguous signals, ensuring that the final output warning information has a very high degree of confidence.

[0026] 3. The designed dynamic noise model inverse optimization mechanism can use confirmed high-confidence risk events to incrementally update the noise template in the dynamic noise feature space. This enables the system to have self-learning and evolution capabilities, continuously adapting to long-term environmental changes (such as seasonal changes and soil moisture changes), thereby ensuring long-term stability and high efficiency of monitoring performance and avoiding performance degradation due to model aging. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0028] Figure 2 This is a schematic diagram of the system functional architecture of the present invention;

[0029] Figure 3 This is a schematic diagram of the adaptive filtering of environmental factors according to the present invention;

[0030] Figure 4 This is a schematic diagram of the coupling feature model analysis of the present invention;

[0031] Figure 5 This is a schematic diagram illustrating the adaptive response and optimization of the present invention. Detailed Implementation

[0032] 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.

[0033] See attached document Figure 1 This invention provides a method for real-time monitoring and early warning of ant infestation in dikes, comprising the following steps:

[0034] S100. Construct a coupled feature model of dynamic noise feature space and acoustic vibration spectrum environmental factors for ant infestation monitoring.

[0035] S200 synchronously collects real-time acoustic and vibration signals and multi-dimensional environmental factor data inside the dam, and preprocesses the collected data;

[0036] S300. Based on the real-time multidimensional environmental factor data, the dynamic noise feature space is called to perform environmental factor adaptive filtering on the preprocessed acoustic vibration signal to obtain the noise-reduced target signal.

[0037] S400. Using the acoustic and vibration spectrum environmental factor coupling characteristic model, the target signal and its synchronous multidimensional environmental factor data are analyzed to obtain a preliminary ant infestation risk level.

[0038] S500. Based on the preliminary ant infestation risk level, selectively perform inverse optimization of the dynamic noise model, or initiate active acoustic detection to verify the ant infestation risk, and update the ant infestation risk level based on the verification results, and finally generate and output early warning information.

[0039] The method of this invention can be executed by a real-time monitoring and early warning system for ant infestations on dikes. This monitoring and early warning system can physically be a monitoring, control, and data processing terminal. The monitoring, control, and data processing terminal can be a high-performance computing device, such as a server or workstation cluster, internally configured with a processor, memory, and a graphics processing unit (GPU) for accelerating parallel computing, and running program instructions to implement the monitoring and early warning method of this invention. Functionally, the monitoring and early warning system can be divided into multiple cooperating software modules, which can run on the monitoring, control, and data processing terminal, specifically including: a model initialization and construction module, a data acquisition and preprocessing module, an environmental factor adaptive filtering module, a coupling analysis and preliminary risk assessment module, and an adaptive response and early warning generation module.

[0040] The model initialization building module is used to execute step S100;

[0041] The data acquisition and preprocessing module is used to execute step S200;

[0042] An environmental factor adaptive filtering module is used to execute step S300;

[0043] The coupling analysis and initial risk assessment module is used to execute step S400;

[0044] The adaptive response and early warning generation module is used to execute step S500.

[0045] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the real-time monitoring and early warning method for ant infestation on dikes as described in any of the foregoing method embodiments.

[0046] In one specific embodiment, the computer-readable storage medium may be a non-volatile storage medium, such as a read-only memory (ROM), flash memory, hard disk (HDD), or solid-state drive (SSD) in a server or terminal device.

[0047] When a computer program stored on the medium is loaded and executed by one or more processors, the processors are able to perform the above method steps.

[0048] In step S100, the method first involves constructing a dynamic noise feature space offline. This feature space is the core basis for implementing environmental factor adaptive filtering (EAAF) in the subsequent step S300, and its purpose is to establish a deterministic mapping relationship between the spectral characteristics of the dam background noise and different environmental factor states.

[0049] The construction process takes place on a dam confirmed to be free of termite activity, or in an area with similar geological and hydrological characteristics. To ensure the integrity of the constructed space, the system requires long-term, continuous data collection via pre-set sensor nodes. This collection period covers typical environmental changes the dam may experience, such as different seasons, diurnal temperature variations, rainfall before and after precipitation, and reservoir water level fluctuations. During the data collection, the system operates at a pre-set sampling frequency. Synchronous recording of background noise time-domain signal And the strictly corresponding multidimensional environmental factor vector .

[0050] To form a structured feature space that can be queried, it is necessary to discretize the continuous multidimensional environmental factor data.

[0051] Specifically, the continuous value range of each environmental factor (such as temperature, humidity, and pore water pressure) is divided into several non-overlapping sub-intervals. This represents a specific multidimensional environmental state. That is, it is composed of specific sub-intervals to which each environmental factor belongs. In this way, the continuous environmental state space is mapped into a set of finite, discrete state units.

[0052] For each collected paragraph under a specific environmental condition The system first performs a short-time Fourier transform (STFT) on the background noise signal to obtain its time-frequency spectrum. Subsequently, in order to obtain a spectral fingerprint that can stably characterize the noise properties under this environmental state, the system averages the time-frequency spectra of all collected samples under this state over time to calculate the multidimensional environmental state. The corresponding average noise power spectral density is denoted as

[0053] ;

[0054] in, Indicates frequency, In a multidimensional environment The noise signal collected in the time frame Spectral amplitude at that location This is the total number of time frames used for averaging. This calculation process effectively smooths out instantaneous noise fluctuations and extracts statistically significant steady-state noise spectrum characteristics dominated by current environmental factors.

[0055] Finally, the dynamic noise feature space It is constructed as a structured dataset, such as a lookup table or a key-value database. Within this space, each discrete multidimensional environment state... As a unique key, its corresponding average noise power spectral density This feature space, once constructed, is stored in the memory of the monitoring and early warning system, providing fast and accurate data support for real-time noise estimation and adaptive filtering in the subsequent online monitoring phase.

[0056] In step S100, in addition to constructing the dynamic noise feature space, an offline training model of acoustic-vibration spectrum environmental factor coupling feature (SE-CFM) is also included. The fundamental purpose of this model is to use machine learning methods to deeply learn and solidify the intrinsic correlation between the acoustic-vibration signal features generated by termite activity and specific combinations of environmental factors, thereby achieving high-precision termite infestation identification in real-time monitoring.

[0057] The training process relies on a pre-collected and labeled comprehensive training dataset. This dataset contains a large number of samples, each consisting of three parts: a raw acoustic-vibration signal, a synchronized multi-dimensional environmental factor vector, and so on. and a clear binary label (For example, a value of 1 indicates that termite activity is confirmed within the signal segment, and a value of 0 indicates that no termite activity is confirmed.) These samples were obtained from termite breeding box simulations in a controlled laboratory environment, as well as from field collections in dam areas where termite infestation has been confirmed through drilling and other methods.

[0058] For each sample in the training dataset, the system first processes its original acoustic and vibration signals using the same Environmental Factor Adaptive Filtering (EAAF) process as in the online monitoring phase to obtain the denoised target signal. This ensures that the data the model faces during training and inference has a consistent feature distribution. Next, a set of acoustic vibration features that effectively characterize the signal properties are extracted from the power spectrum of the target signal, forming an acoustic vibration feature vector. This vector includes, but is not limited to, the spectral centroid reflecting the center of the signal frequency component distribution. Spectral entropy, a measure of the spectral complexity or uncertainty of a signal. And the proportion of energy in the key frequency bands of termite activity. .

[0059] Subsequently, the extracted acoustic vibration feature vectors Its synchronized multidimensional environmental factor vector At the feature level, concatenation is performed to fuse them into a higher-dimensional coupled feature vector. The mathematical representation of this vector is: This coupling step is one of the key aspects of this invention. It enables the model not only to learn the patterns of the acoustic and vibration signals themselves, but also to learn under what environmental conditions these patterns are more likely to occur, thereby greatly enhancing the model's ability to identify ant damage signals in complex environments.

[0060] The offline training model for the coupled feature vectors of acoustic and vibrational spectrum environmental factors employs algorithms capable of effectively processing these coupled feature vectors and learning nonlinear mapping relationships, such as Gradient Boosting Decision Tree (GDBT), Random Forest, or, for scenarios requiring the capture of temporal information, Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) units. After dividing the entire labeled dataset into training, validation, and test sets, the model is trained on the training set. The training process iteratively optimizes and adjusts the model's internal parameters to minimize the loss function between the model's predicted output and the true label. The loss function typically uses binary cross-entropy:

[0061] ;

[0062] in, The total number of training samples, It is the first The true label of each sample, and The model is for the first Each sample is predicted to have the probability of being an ant infestation. After training, the model's generalization ability is evaluated using an independent test set. Once the model's performance reaches the preset accuracy and recall standards, its final structure and parameters will be solidified and deployed into the monitoring and early warning system for real-time and efficient ant infestation risk probability inference in step S400.

[0063] In step S200, the method of the present invention transitions from the offline preparation stage to the real-time online monitoring stage. Multiple integrated acoustic and environmental sensor nodes deployed inside the dam structure, particularly in areas with high ant infestation rates or critical structural locations, are activated and begin performing high-frequency, synchronous data acquisition tasks. The core of this step is acquiring raw data streams that accurately reflect the microscopic activities inside the dam and its surrounding physical environment. To ensure strict temporal consistency and comparability between acoustic and vibration signals and multidimensional environmental factors, all sensor nodes are calibrated using a unified clock synchronization mechanism (e.g., based on Network Time Protocol (NTP) or higher-precision GPS pulse-second signal (PPS)) to ensure that all collected samples are accompanied by a high-precision, unified timestamp.

[0064] During the data acquisition process, each node uses a preset, uniform sampling frequency that was determined during the S100 model construction phase. The work is carried out. Nodes synchronously acquire two types of data: the first type is the raw acoustic vibration time-domain signal captured by high-sensitivity acoustic vibration sensors (such as piezoelectric ceramic sensors or MEMS microphones). The second category consists of multi-dimensional environmental factor vectors collected by environmental sensor arrays integrated on nodes. This vector Specifically, this includes key physical quantities reflecting the acoustic properties of the soil medium, namely soil temperature. Soil moisture pore water pressure and soil electrical conductivity .

[0065] The acquired raw data stream must undergo rigorous preprocessing before entering subsequent analysis modules. First, the raw acoustic and vibration signals... A DC component removal operation is performed to eliminate zero-point drift caused by the sensor itself or the circuitry. Next, a digital bandpass filter is applied to the signal. The passband range of this filter is carefully set based on the characteristic frequency band of ant infestation activity (e.g., a specific kHz band) determined in S100. Its purpose is to filter out low-frequency geological disturbances (such as water flow and wind-induced vibrations) and high-frequency electronic white noise unrelated to ant infestation activity to the greatest extent possible, thereby achieving an initial signal-to-noise ratio improvement without compromising the effective signal.

[0066] Finally, to ensure the numerical stability of subsequent algorithms and eliminate the influence of different physical dimensions, the system performs numerical analysis on the filtered acoustic-vibration signal and the multidimensional environmental factor vector. Each component in the model is normalized. For example, min-max scaling is used to linearly map the data to the interval [0,1] or [-1,1]. This step affects the performance of signal processing algorithms such as spectral subtraction in S300, and the coupling feature model of acoustic-vibration spectrum environmental factors in S400. Fairly handling the weights of different features is crucial. After preprocessing, the system outputs a clean, aligned, and normalized data stream for deep adaptive filtering in step S300.

[0067] Reference Appendix Figure 3 In this step S300, the system receives the preprocessed real-time acoustic vibration signal from step S200, along with a strictly synchronized multi-dimensional environmental factor vector. The core task of this step is to utilize this real-time, continuous-valued multidimensional environmental factor vector. Query the dynamic noise feature space constructed in step S100, which uses discrete states as keys. This allows us to obtain the most suitable background noise power spectrum estimate under the current environment. .

[0068] Due to the real-time multidimensional environmental factor vector Since it is continuously changing, its value (e.g., temperature 22.5℃, humidity 73%) is almost impossible to precisely match a specific state discretely defined in S100. (e.g., (20℃, 25℃) and (70%, 80%)). Therefore, the system needs to perform an efficient state matching and template estimation process.

[0069] In one specific embodiment, the system first processes the multidimensional environmental factor vector. Each continuous component in the data is rapidly quantized. Specifically, the system will quantize the real-time temperature value. Humidity value These are mapped to the corresponding sub-intervals defined in S100, respectively. This is achieved by mapping the multidimensional environmental factor vectors... Perform this operation on all components, and the system will generate a continuous multidimensional environmental factor vector. It is mapped to the discrete state unit that is closest to it in the discrete feature space. Subsequently, the system uses this As an index, in the dynamic noise feature space A high-speed search is performed to directly extract the pre-calculated and stored average noise power spectral density. This spectrum It was then designated as the background noise estimate for the current moment. .

[0070] In another, more refined embodiment, an interpolation method is employed to obtain a smoother and more accurate noise estimate. When When a discrete state element is entered into, the system not only searches for that element, but also searches for its representation in the multidimensional environmental factor space. The system retrieves the discrete states of K-Nearest Neighbors and simultaneously extracts multiple noise templates corresponding to these neighboring states. Then, it uses an interpolation function... (For example, based on) To these (the inverse distance-weighted average of the distances to the nearest neighbor centers) for this The noise templates are dynamically weighted and fused. This process synthesizes a new background noise estimate in real time. This template reflects the noise characteristics of the real environment, which is currently in a discrete state, more accurately than any single pre-stored template.

[0071] Regardless of the method used, the final output of this step is... It is a high-fidelity background noise power spectrum estimate that is closely related to current environmental factors (such as temperature, humidity, and water pressure).

[0072] After obtaining a high-fidelity background noise power spectrum estimate of the current environment... Then, the system immediately processes the pre-processed real-time acoustic and vibration signals from the S200 (referred to here as...). The system performs noise reduction processing. First, it processes the real-time acoustic and vibration signal. Perform a short-time Fourier transform (STFT) to convert it from the time domain to the time-frequency domain, obtaining its complex spectrum. ,in For frequency, This is a time frame. The system then calculates its real-time signal power spectrum. .

[0073] The core of this invention lies in performing a spectral subtraction method based on environmental factors. The system performs real-time signal power spectrum... With the acquired time frame Aligned background noise power spectrum estimation Subtraction is performed. To achieve robust noise reduction and suppress artifacts, this embodiment employs an improved spectral subtraction algorithm, which introduces two key control parameters: an over-subtraction factor. and a spectral lower limit factor .

[0074] Over-reduction factor (Usually set to a value greater than or equal to 1) is used to amplify the estimated noise power spectrum by a certain proportion before subtraction. The purpose of this operation is to compensate for possible noise underestimation when constructing the noise template in S100, and to deal with short-term, unmodeled noise fluctuations in the environment, thereby eliminating background noise more thoroughly.

[0075] Spectral lower limit factor A small positive number close to zero is used to set a spectral floor. During spectral subtraction, the signal energy in some frequency compartments may become negative or zero after subtracting the noise energy. Setting this value directly to zero would introduce harsh, unnatural musical noise during time-domain reconstruction. By introducing... This invention ensures that the power spectrum of the denoised signal will not be lower than a tiny percentage of the original signal power spectrum at any frequency point (i.e., This greatly suppresses the generation of musical noise and preserves the naturalness of the signal.

[0076] Therefore, the power spectrum of the target signal after noise reduction It is calculated using the following formula:

[0077] ;

[0078] The power spectrum of the target signal is calculated. Then, the system takes the square root to obtain the amplitude of the target signal. During the signal reconstruction stage, this invention preserves the phase information of the original noisy signal. This is based on a widely validated assumption that additive noise primarily affects the amplitude of a signal rather than its phase, and that the intelligibility of a signal is mainly carried in its phase information. The system will then determine the amplitude of the new target signal. With the original phase By combining these methods, the complex spectrum of the target signal can be reconstructed. .

[0079] Finally, the complex spectrum is obtained by performing the inverse short-time Fourier transform (ISTFT). Transforming back to the time domain yields the final, denoised target signal in the time domain. Time-domain target signal It greatly suppresses the background noise that changes with the environment, highlights the acoustic and vibration characteristics of possible ant infestations, and provides high-quality data input for the high-precision analysis in step S400.

[0080] Reference Appendix Figure 4 In step S400, the core task of the system is to construct in real time a coupled feature vector that can comprehensively reflect the current acoustic and vibration activities and their physical environment. This vector serves as the direct input for the inference process of the acoustic-vibration spectrum-environmental factor coupled feature model (SE-CFM) in subsequent step S400. This construction process is one of the key innovations of this invention in achieving high-precision recognition, and it consists of two stages: acoustic-vibration feature extraction and multi-dimensional feature coupling.

[0081] First, the system processes the denoised time-domain target signal output in step S300. Real-time time-frequency analysis is performed. Specifically, the system performs... Perform a Short Time Fourier Transform (STFT) to obtain its value in the current time frame. Target signal power spectrum Subsequently, the system calculates and extracts a set of predefined acoustic-vibration features from the power spectrum, designed to efficiently characterize ant infestation activities, thus forming a real-time acoustic-vibration feature vector. In this embodiment, the vector Specifically, it includes:

[0082] Spectral centroid This feature calculates the centroid or weighted average frequency of the power spectrum of the current frame. It reflects the main distribution location of signal energy in the frequency domain. Ant damage activities (such as gnawing and vibration) often have specific frequency domain clustering, and this feature can effectively capture such changes.

[0083] Spectral entropy This feature measures the complexity or flatness of the power spectrum. A highly ordered, peaked signal (such as a specific communication signal of termites) has low spectral entropy, while a disordered, flat signal (such as residual white noise) has high spectral entropy.

[0084] Key frequency band energy percentage Based on offline analysis in S100, one or more key frequency bands where termite activity is most active (e.g., arrive The system calculates and normalizes the total energy within these frequency bands (Hz). This is an extremely effective and robust feature for identifying ant infestation activity.

[0085] Extracting acoustic vibration feature vectors Simultaneously, the system acquires the pre-processed data transmitted from S200 and synchronizes it with the current time frame. Strictly synchronized multidimensional environmental factor vector This vector Includes the current soil temperature Soil moisture pore water pressure and soil electrical conductivity .

[0086] Subsequently, the system performs the crucial feature coupling step. (See attached...) Figure 4 As shown, the system will use acoustic vibration feature vectors at the feature level. With multidimensional environmental factor vector Perform cascading operations to fuse them into a single, high-dimensional coupled feature vector. Its mathematical representation is:

[0087] ;

[0088] The innovation of this coupling operation lies in the fact that it provides information to the subsequent SE-CFM model, enabling the model to learn deeper and more complex rules, such as a certain acoustic vibration feature. When the soil is dry ( (At lower levels) it may be harmless, but when the soil is moist ( A higher reading (likely indicating an ant infestation) is highly likely a sign of ant damage. This is how the signal is constructed. This greatly enhances the feature recognition ability and the model's anti-interference ability, laying a solid foundation for the S400's subsequent accurate reasoning and discrimination.

[0089] High-dimensional coupled feature vectors were constructed in real time. Then, the system immediately invokes the acoustic and vibration spectrum environmental factor coupling feature model that was trained offline and deployed in step S100. The system will do this The vector is used as input to perform a real-time forward propagation (inference) computation.

[0090] In the reasoning process, the acoustic and vibration spectrum environmental factors are coupled with the characteristic model. (Whether it's a structure like GBDT, Random Forest, or RNN) it utilizes the complex nonlinear mapping relationships learned during the training phase to... The coupling relationship between the acoustic and vibrational characteristics and environmental factors is comprehensively evaluated. This calculation process is efficient and aims to output a continuous probability value in real time. The value is between [0,1].

[0091] This probability value Strictly corresponding to the definition of training data labels in S100, its physical meaning is: it quantifies the confidence level of the current acoustic vibration signal being judged as real ant infestation activity under the current environmental conditions. The closer a value is to 1, the more consistent the model considers the combination of the currently monitored signal pattern and the environment to be with the known ant damage activity patterns in the training data; conversely, the closer a value is to 0, the more it indicates a tendency towards background noise or harmless disturbances.

[0092] To facilitate the decision-making and execution of subsequent S500 steps, the system will use a set of preset multi-level thresholds to evaluate the continuous probability value. The data is discretized to map it to a preliminary ant infestation risk level. For example, the system can set a suspected threshold. and a high-risk threshold (For example These thresholds are empirically set and solidified in the system based on the model's performance on the validation set during the S100 training phase (e.g., by analyzing ROC curves to balance recall and precision).

[0093] Based on this, the preliminary risk level determination logic is as follows:

[0094] if The initial risk level is assessed as safe or risk-free.

[0095] if The initial risk level is classified as suspected risk.

[0096] if The initial risk level is determined to be highly suspected or highly confident risk.

[0097] This initial ant infestation risk level will not immediately trigger the final alarm. Instead, it is passed to the adaptive response and verification module of S500 as the final output of step S400, so that the system can make the next intelligent decision.

[0098] Reference Appendix Figure 5 When the preliminary ant infestation risk level output by step S400 is a highly suspected or high-confidence risk (i.e. The system triggers this high-confidence response mechanism. The core purpose of this mechanism is to leverage this highly deterministic ant infestation signal discovery to, in turn, refine the dynamic noise feature space constructed in S100. Perform an online, incremental optimization.

[0099] The starting point of this logic is: the coupling characteristic model of acoustic and vibration spectrum environmental factors. The noise-reduced time-domain target signal output by the S300 was confirmed with high confidence. (and its target signal power spectrum) When the activity is genuine ant infestation, the system can detect it. It is considered the best estimate of the true signal components at the current moment.

[0100] The system will then perform a reverse noise stripping process. It simultaneously holds two key data points:

[0101] Real-time signal power spectrum after S200 preprocessing: ;

[0102] The power spectrum of the target signal that was noise-reduced by S300 and identified as a high-confidence signal by S400: .

[0103] The system is based on the assumption that the signal and noise are approximately added together on the power spectrum (i.e. By performing a reverse subtraction, a new estimate of the true background noise at the current moment is calculated.

[0104] ;

[0105] this This represents the current real-time environmental factors. Below is the power spectrum of the actual background noise occurring within the dam. This measured value is likely to be higher than the historical average value stored in S100. It can better reflect the current noise situation.

[0106] Subsequently, the system performs a dynamic noise feature space analysis. Incremental updates. The system first updates based on the current multidimensional environmental factor vector. It was located in the dynamic noise feature space. The corresponding discrete state unit The system then employs a smooth update strategy, such as an exponential moving average, to update the data stored in that unit. Average noise power spectral density

[0107] ;

[0108] in, yes The original average noise power spectral density stored in it, It is a small learning rate (e.g., 0.01) that controls the magnitude of the update of the original template by new samples. This mechanism allows for dynamic noise feature space. Over time, the system slowly and steadily adapts to long-term shifts in the dam environment (such as seasonal variations and changes in soil density), rather than being drastically altered by a single event. Through this closed-loop feedback, the system utilizes high-confidence signal discovery to optimize its noise model, enabling more precise filtering in the future (S300), thereby improving the accuracy of the next analysis (S400).

[0109] Reference Appendix Figure 5 When the initial ant infestation risk level output by step S400 is a suspected risk, that is, the ant infestation probability output by S400... satisfy This indicates that the signal characteristics captured by passive monitoring are rather vague, or that the confidence level is insufficient to directly determine it as high-risk. To avoid missed detections (if it is genuine ant infestation) and false alarms (if it is rare noise interference), the system will trigger this active acoustic detection and verification mechanism at this time.

[0110] The core of this mechanism is to temporarily switch the relevant sensor nodes from a purely passive listening mode to an active transmission and reception mode in order to obtain direct evidence about the physical properties of the dam medium.

[0111] First, the system associates a command with a sensor node located at the suspected risk signal source (e.g., determined by the node with the strongest signal in the S400), making it act as a transmitter (TX). This TX node emits a pre-set high-frequency coded acoustic signal with excellent autocorrelation characteristics into the dam medium. In this embodiment, the signal The preferred signals are linear frequency modulated (Chirp) signals or M-sequence encoded signals, which have high time compression ratios and noise immunity, making them suitable for detection in strongly attenuated soil media.

[0112] Meanwhile, one or more neighboring nodes located near the TX node are designated as receiver sources (RX). These RX nodes maintain a high-speed acquisition mode, synchronously recording the high-frequency coded acoustic signal. The response signal received after penetrating or diffracting through a suspected risk area. .

[0113] The system then responds to the received signal Perform matched filtering. Matched filtering is an optimal detection technique in signal processing; it receives the response signal. With high-frequency coded acoustic signals The temporal deconvolution copy is used for convolution operations. Because... The excellent autocorrelation characteristics of the signal, the output of the matched filter The accurate propagation delay of the signal It exhibits a very sharp peak.

[0114] The verification logic of this step is that termite nests physically exhibit structural anomalies in the dam medium, typically being cavitary, loose, or high-water-content areas, whose acoustic impedance is distinctly different from the surrounding dense soil.

[0115] Reference Establishment: During the S100 offline phase or when the dam is in a healthy state, the reference acoustic propagation delay between all TX-RX node pairs has been pre-calibrated and stored. and reference signal amplitude .

[0116] Real-time comparison: The system will accurately measure the propagation delay. Propagation delay of reference sound wave Compare them.

[0117] Increased latency: If the suspected risk area does indeed contain anthill voids, the high-frequency coded acoustic signal... The sound wave will be forced to detour around the cavity, resulting in a longer propagation path and thus increasing the measured propagation time delay of the reference sound wave. Significantly greater than .

[0118] Amplitude attenuation: The interface of the anthill cavity causes strong scattering and energy attenuation of sound waves, resulting in an increase in the amplitude of the matched filter peak. Significantly lower than the reference signal amplitude .

[0119] The system is based on the time delay deviation and amplitude attenuation rate To calculate a structural anomaly index. If this index exceeds a preset verification threshold... If the system detects an abnormality, it determines that the suspected risk has been actively verified and upgrades its risk level to high risk. If the active investigation does not find any significant anomalies, the suspected risk is downgraded to safe or low-risk disturbance.

[0120] In this final step of S500, the system comprehensively assesses and updates the preliminary risk level output by S400 to form a final, highly reliable risk judgment, and generates structured early warning information accordingly. This step is the decision output point of the entire monitoring and analysis process, ensuring that only verified or highly confident events are reported.

[0121] The logic of this update and generation process is strictly defined based on the output of the preceding steps:

[0122] For a safe status: If the initial risk level output by S400 is safe (i.e., probability of ant infestation). If the S500's adaptive response module is not triggered, the system will directly confirm the final risk level as safe in this step. The system will not generate any warnings, will maintain silent monitoring, and will proceed to the next data acquisition and analysis cycle.

[0123] For high-confidence risk states: If the initial risk level output by S400 is highly suspected (i.e., The system then performed inverse optimization of the noise model. This operation itself implicitly confirms the high-confidence judgment. Therefore, in this step, the system formally upgrades the preliminary level to the final risk level: high risk.

[0124] For a suspected risk status: This is the most crucial part of the logical judgment in this step. If the S400 output is a suspected risk (i.e., ...), then... The system must rely on the results of active acoustic probing verification to make a final decision.

[0125] Scenario A (Verification Successful): If the structural anomaly index calculated by the active detection exceeds the preset verification threshold. This indicates the physical characteristics of sound wave propagation (such as time delay deviation). The system confirmed the presence of a media anomaly in the area. Therefore, the suspected risk was upgraded to a final risk level: High Risk (Active Verification).

[0126] Case B (Verification Failure): If the actively detected structural anomaly index does not exceed the verification threshold. This indicates that the suspected signal is likely a rare environmental noise or interference that was not completely filtered out by the S300. The system has downgraded this suspected risk to the final risk level: Safe (Disturbance verified).

[0127] After determining the final, more reliable ant infestation risk level, and only if this final level is determined to be high risk or high risk (actively verified), the system will generate a final warning. The system will generate a structured warning data packet, which will be pushed to the monitoring and warning system's application platform for manual review and on-site handling by operations and maintenance personnel. This structured warning information specifically includes the following fields:

[0128] [Event ID]: A unique alert identifier.

[0129] [Final Risk Level]: High risk or high risk (active verification).

[0130] [Location Information]: The sensor node number that triggered the alarm or its coordinates.

[0131] [Timestamp]: The time when the anomaly was first detected.

[0132] [Confidence Level]: The probability value output by S400

[0133] [Verification Method]: Indicate whether it is passive analysis confirmation or active exploration confirmation.

[0134] [Key Evidence]: If it is an active verification, please include key indicators, such as the measured time delay deviation. .

[0135] Through this series of comprehensive judgments and information generation, the method of this invention not only improves the accuracy of ant infestation monitoring, but also greatly reduces the false alarm rate of the system through the adaptive verification mechanism of S500, realizing intelligent, efficient and reliable early warning of ant infestation on dikes.

Claims

1. A method for real-time monitoring and early warning of ant infestation on dikes, characterized in that, Includes the following steps: S100. Construct a dynamic noise feature space and a coupled feature model of acoustic and vibration spectrum environmental factors for ant infestation monitoring. The construction of the dynamic noise feature space includes: offline acquisition of dam background noise signals under different environmental conditions; dividing the continuous value domain of multidimensional environmental factor data into multiple discrete environmental state units; calculating the average noise power spectrum corresponding to each environmental state unit, storing the average noise power spectrum as a noise template, and establishing a mapping relationship between the environmental state unit and the noise template. S200 synchronously collects real-time acoustic and vibration signals and real-time multi-dimensional environmental factor data inside the dam, and preprocesses the collected data; S300. Based on the real-time multidimensional environmental factor data, the dynamic noise feature space is called to perform environmental factor adaptive filtering on the preprocessed acoustic vibration signal to obtain the noise-reduced target signal. S400. Using the aforementioned acoustic and vibration spectrum environmental factor coupling characteristic model, the target signal and synchronized multidimensional environmental factor data are analyzed to obtain a preliminary ant infestation risk level. S500. Based on the preliminary ant infestation risk level, selectively perform inverse optimization of the dynamic noise model, or initiate active acoustic detection to verify the ant infestation risk, and update the ant infestation risk level based on the verification results, and finally generate and output early warning information.

2. The method for real-time monitoring and early warning of ant infestation in dikes according to claim 1, characterized in that, In step S300, calling the dynamic noise feature space based on the real-time multidimensional environmental factor data includes: In the dynamic noise feature space, find the discrete environmental state units that are the nearest neighbors of the real-time multidimensional environmental factor data in the factor space; Retrieve multiple noise templates corresponding to multiple environmental state units; Based on the real-time multidimensional environmental factor data and the weighted interpolation function of the center points of multiple environmental state units, the multiple noise templates are dynamically weighted and fused to synthesize the background noise power spectrum estimate.

3. The method for real-time monitoring and early warning of ant infestation on dikes according to claim 1, characterized in that, The analysis of the target signal and synchronized multidimensional environmental factor data in step S400 includes: Extract a set of acoustic vibration features from the target signal; The acoustic vibration features and the multidimensional environmental factor data are cascaded at the feature level to construct a coupled feature vector. The coupled feature vector is input into the acoustic and vibration spectrum environmental factor coupled feature model for inference to obtain the ant infestation risk probability corresponding to the preliminary ant infestation risk level.

4. The method for real-time monitoring and early warning of ant infestation in dikes according to claim 1, characterized in that, The S500 step of initiating active acoustic detection to verify the risk of ant infestation includes: When the preliminary ant infestation risk level is suspected risk, the control source node transmits coded acoustic signals to the dam medium; The receiving source node receives the response signal after the coded acoustic signal penetrates the suspected risk area; Perform matched filtering on the response signal to obtain the propagation delay and amplitude of the encoded acoustic signal; The propagation delay and amplitude are compared with the pre-stored reference values ​​to determine whether there is a medium abnormality caused by ant damage.

5. The method for real-time monitoring and early warning of ant infestation in dikes according to claim 1, characterized in that, The inverse optimization of the dynamic noise model in step S500 includes: When the initial ant infestation risk level is a high confidence risk, the power spectrum of the denoised target signal is regarded as the real signal component. The real signal component is extracted from the power spectrum of the preprocessed acoustic and vibration signal to obtain an estimate of the real background noise under the current environment; Using the estimated real background noise value, the noise template corresponding to the current real-time multidimensional environmental factor data in the dynamic noise feature space is incrementally updated.

6. The method for real-time monitoring and early warning of ant infestation in dikes according to claim 4, characterized in that, The step S500, which updates the ant infestation risk level based on the verification results, includes: When the active acoustic detection and verification determines that there is a medium anomaly, the suspected risk level is upgraded to high risk; If the active acoustic detection verification does not determine the presence of a medium anomaly, the suspected risk level will be downgraded to safe.

7. The method for real-time monitoring and early warning of ant infestation in dikes according to claim 2, characterized in that, The environmental factor adaptive filtering of the preprocessed acoustic vibration signal in step S300 includes: Calculate the real-time signal power spectrum of the preprocessed acoustic vibration signal; The result of subtracting the background noise power spectrum estimate from the real-time signal power spectrum by multiplying it by an over-subtraction factor; A spectral floor is set, which is the product of the real-time signal power spectrum and a spectral lower limit factor, to obtain the power spectrum of the denoised target signal.

8. A real-time monitoring and early warning system for dam ant infestation, applied to the real-time monitoring and early warning method for dam ant infestation as described in any one of claims 1-7, characterized in that, include: The model initialization building module is used for offline training of the coupled feature model of acoustic and vibration spectrum environmental factors and to construct a dynamic noise feature space; The data acquisition and preprocessing module is used to acquire real-time acoustic and vibration signals and synchronous multi-dimensional environmental factor data within the dam monitoring area; An environmental factor adaptive filtering module is used to access the dynamic noise feature space based on the multidimensional environmental factor data, obtain the background noise power spectrum estimate corresponding to the current environment, and perform adaptive spectrum subtraction and noise reduction on the acoustic vibration signal to obtain the denoised target signal. The coupling analysis and preliminary risk assessment module is used to extract the acoustic and vibration characteristics of the target signal, couple them with the multidimensional environmental factor data, construct a coupling feature vector, and input the coupling feature vector into the coupling feature model for reasoning to obtain a preliminary ant infestation risk level. The adaptive response and early warning generation module is used to execute an adaptive response and generate an ant infestation early warning based on the preliminary ant infestation risk level.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images