Method and system for monitoring water and soil loss in karst region

Abstract

The invention discloses a karst region water and soil loss monitoring method and system. According to the monitoring method, a water and soil loss risk prediction result of a target area is obtained based on a deployed acoustic sensor array, the acoustic sensor array is deployed at a key monitoring point of the target area to be monitored, and acoustic wave signals of the earth surface and underground environment are collected at the same time; acoustic wave signals detected by the acoustic sensor array are collected, and the acoustic wave signals are converted into frequency domain characteristic data; inputting the frequency domain feature data into a pre-trained acoustic mode recognition model, and recognizing acoustic features related to water and soil loss; based on an output result of the acoustic mode recognition model, judging the water and soil loss type, degree and development trend of the monitored area, and generating a water and soil loss risk prediction result; according to the invention, real-time and continuous monitoring of the water and soil loss process in the karst region is realized through an acoustic monitoring means, the monitoring precision and early warning timeliness are improved, and a scientific basis is provided for water and soil conservation management.

Description

Technical Field

[0001] The invention belongs to the technical field of soil and water loss monitoring, and in particular relates to a method and system for monitoring soil and water loss in karst areas. Background Art

[0002] At present, the main problems in soil and water loss monitoring in karst areas include: traditional ground monitoring stations are sparse and cannot fully cover complex terrain; remote sensing images have limited penetration in areas with dense vegetation; the correlation between groundwater processes and surface erosion is difficult to quantify; and sudden erosion caused by intermittent heavy rainfall is difficult to capture in a timely manner.

[0003] In addition, existing monitoring methods are difficult to take into account the monitoring needs of the surface and underground environments at the same time, and lack the ability to monitor the soil erosion process in real time under the special geological structure of karst. Traditional monitoring methods often rely on manual on-site surveys or remote sensing image analysis at fixed time intervals. It is difficult to continuously monitor the soil erosion process and achieve early warning, resulting in serious monitoring lag problems and failure to provide timely and effective decision-making support for prevention and control measures. Summary of the invention

[0004] In response to the above problems, the present invention proposes a method and system for monitoring soil and water loss in karst areas. By deploying an acoustic sensor array to simultaneously collect sound wave signals from the surface and underground environments, and combining a deep learning algorithm to construct an acoustic pattern recognition model, the real-time dynamic monitoring and early warning of the soil and water loss process in the karst area can be achieved.

[0005] The specific technical solutions are as follows: The present invention provides a method for monitoring soil and water loss in a karst area. The monitoring method obtains a soil and water loss risk prediction result of a target area based on a deployed acoustic sensor array. The monitoring method comprises the following steps: Step S1, deploying an acoustic sensor array at key monitoring points in a target area to be monitored, wherein the acoustic sensor array includes a plurality of acoustic sensors for simultaneously collecting acoustic wave signals of the surface and underground environments.

[0006] The key monitoring points in the target area are divided into surface monitoring points and underground monitoring points. The surface monitoring points at least include: karst depressions and karst grooves; the underground monitoring points include: karst caves, cave entrances and underground river entrances.

[0007] Step S2, collecting the sound wave signals detected by the acoustic sensor array, and converting the sound wave signals into frequency domain feature data.

[0008] Step S3: input the frequency domain feature data into a pre-trained acoustic pattern recognition model, where the model is based on a deep learning algorithm and is used to identify acoustic features related to soil erosion.

[0009] Step S4, based on the output results of the acoustic pattern recognition model, determine the type, degree and development trend of soil erosion in the monitored area, and generate soil erosion risk prediction results.

[0010] Furthermore, the process of extracting the frequency domain feature data includes: The original sound wave signal is preprocessed, and the preprocessing includes denoising, filtering and signal enhancement; the preprocessed signal is subjected to time-frequency analysis to extract time domain features and frequency domain features.

[0011] The acoustic features of the sound wave signal are calculated, and the acoustic features include: Mel frequency cepstral coefficients MFCC, spectral flux, spectral centroid and spectral bandwidth; the statistical features of the sound wave signal are extracted, and the statistical features include: mean, variance, skewness, kurtosis and energy distribution.

[0012] Furthermore, the acoustic features are calculated using short-time Fourier transform (STFT): ;in, is the original sound wave signal, Indicated in In the analysis window, the position in the window is The sample value of It is the sample index in the window, i.e. the number of FFT points, ranging from 0 to N-1; is the time frame index, indicating the analysis window; is the frame shift, which indicates the sample number offset between adjacent windows; k is the frequency index, which indicates the frequency point in the short-time Fourier transform STFT; is a window function; Represents the short-time Fourier transform coefficient of the kth frequency point on the mth time frame; The calculation formula of Mel frequency cepstral coefficient MFCC is: ;in, represents the number of Mel filter banks, represents the cepstral coefficient index; The calculation formula of spectral centroid SC is: ; The calculation formula of spectral flux SF is: ; Sound wave energy distribution The calculation formula is: ;in, represents the frequency band index, and Indicates The lower and upper frequency indices of the frequency band.

[0013] Furthermore, the acoustic pattern recognition model is a hybrid model of a deep convolutional neural network and a long short-term memory network, and its structure includes: an input layer, a convolution layer, a pooling layer, a long short-term memory network layer, a fully connected layer and a softmax layer; The input layer receives an acoustic feature vector, which is a multidimensional array including time domain features, frequency domain features, acoustic features and statistical features; the time domain features include the time series data of the original signal after preprocessing; the frequency domain features are the spectrum data obtained by short-time Fourier transform; the acoustic features include Mel frequency cepstral coefficients MFCC, spectral flux SF, spectral centroid SC and spectral bandwidth SBW; the statistical features include the mean, variance, skewness, kurtosis and energy distribution of each frequency band of the signal; The convolution layer extracts the local pattern of acoustic features, the pooling layer reduces the feature dimension and improves the model robustness, the long short-term memory network layer captures the temporal dependency of acoustic features, and the fully connected layer and softmax layer output the probability distribution of soil erosion type and degree; The mathematical expression of the convolutional layer operation is: ,in: is the output of the previous layer; is the convolution kernel weight; is the bias term; is the rectified linear unit activation function; Represents the convolution operation; Input gate of long short-term memory network LSTM: ; Forget Gate: , output gate: ; Candidate memory cells: ; Memory unit update: ; Hide status updates: ; in: is the input of the current time step, is the hidden state of the previous time step, is the memory cell state at the previous time step, , , , is the weight matrix, , , , is the bias vector, is the hyperbolic tangent activation function, is the Hadamard product, using element-by-element multiplication; Represents the sigmoid activation function; The expression of the output layer is: ;in, is the output of the LSTM layer; is the output layer weight matrix; is the output layer bias vector; is the normalized exponential function.

[0014] Furthermore, the acoustic pattern recognition model is trained using a cross entropy loss function: ;in, is the total number of soil erosion types, is the true label, Predict probabilities for the model.

[0015] Furthermore, the training steps of the acoustic pattern recognition model are: Step S31, collecting acoustic wave samples generated by different types of soil erosion processes, including acoustic wave samples under surface runoff, rainfall erosion, rock collapse, underground dissolution and underground river flow scenarios; Step S32, labeling the sound wave samples and establishing a corresponding relationship between the acoustic features and the type and degree of soil erosion; Step S33, using principal component analysis PCA and t-distributed random neighbor embedding t-SNE dimensionality reduction technology to extract the distinguishing features of the sound wave sample; The expression of the principal component analysis PCA is: ;in, is the original acoustic feature matrix, is the principal component loading matrix, is the feature matrix after dimensionality reduction; Calculated from the eigenvectors: ;in, is the covariance matrix of the original features, is the eigenvalue diagonal matrix; Use t-distributed random neighbor embedding to calculate conditional probabilities in high-dimensional space: ; Conditional probability in low-dimensional space: ; The optimization goal is to minimize the KL divergence: ; Step S34, using batch gradient descent method to optimize model parameters, the learning rate is initially set to 0.001, and a learning rate decay strategy is used; Cluster analysis uses the K-means algorithm: ;in, For sample Belongs to cluster indicator variable of Cluster the center of is the number of clusters, is the sample size; Step S35, using early stopping method to avoid overfitting, and stopping training when the performance of the validation set no longer improves; Furthermore, the method also includes a sound wave propagation model correction step: establishing a sound wave propagation model in a karst area, taking into account the influencing factors of topography, rock material, vegetation coverage and meteorological conditions, using a known sound source to conduct a test, and obtaining the propagation characteristics of sound waves in a karst environment; based on the test results, correcting the sound wave propagation model parameters, and using the corrected propagation model to improve the accuracy of sound source positioning and signal analysis; The sound wave propagation model is based on the wave equation: ;in, is the sound pressure, is the speed of sound, is the Laplace operator; Under complex terrain conditions, the ray tracing method is used to calculate the sound wave propagation path: ;in, is the sound wave ray path, is the path length parameter, is the refractive index of sound waves, which is related to the sound velocity of the medium. , is the reference sound speed; Sound wave attenuation model considering terrain and atmospheric conditions: ;in, is the sound pressure level at the receiving point; is the sound pressure level of the sound source; is the distance from the sound source to the receiving point; is the atmospheric absorption coefficient; is the attenuation caused by terrain; Attenuation caused by atmospheric conditions.

[0016] Further, in step S4, based on the output result of the acoustic pattern recognition model, a soil erosion risk prediction result is generated, and a soil erosion risk prediction score is generated. It is expressed as: ; For the Weight coefficient of soil erosion class; For the The probability of soil erosion occurring is For the the severity of soil erosion; Set multi-level warning thresholds based on risk scores, divided into four warning levels: minor, moderate, severe and catastrophic; When the risk score exceeds the warning threshold, the system automatically triggers the corresponding level of warning signal; According to different warning levels, corresponding emergency response measures are formulated, including increasing monitoring frequency, notifying relevant departments and activating emergency plans.

[0017] Furthermore, the method further comprises the step of calculating the amount of soil loss based on the acoustic characteristics: Calculate soil loss using acoustic signal characteristics monitored by acoustic sensors , whose expression is: ;in, is the amount of soil loss, in tons / hectare; \kappa is the calibration coefficient related to soil type; and is the start and end time of monitoring; is the frequency-dependent soil loss transfer function; is the sound wave signal intensity in the time-frequency domain.

[0018] The conversion function Determined according to different soil types and particle size distribution:

[0019] in, is the base conversion factor; For the Weight coefficient of soil particle size; For the The characteristic frequency of soil particle size; is the frequency distribution width; M is the number of soil particle size classifications.

[0020] The acoustic signal strength It is positively correlated with the scouring intensity during soil erosion and is calculated by the power spectral density of short-time Fourier transform: ;in, For the On the time frame The short-time Fourier transform coefficients of the frequency points are Corresponding to the frequency index , Corresponds to the timeframe index ; The system calculates the amount of soil loss With preset threshold The comparison results are used to adjust the soil and water loss risk prediction score. The risk score increases ,in is the adjustment factor, This is the largest amount of soil loss in history.

[0021] Based on the same inventive concept, the present invention provides a karst area soil erosion monitoring system for executing the monitoring method of the present invention, wherein the monitoring system comprises: a signal acquisition unit, a signal processing unit, a pattern recognition unit and a risk assessment unit connected in sequence; Furthermore, the signal acquisition unit is used to receive the sound wave signal collected by the acoustic sensor array and convert the sound wave signal into frequency domain feature data; The acoustic sensor array includes a plurality of acoustic sensors for collecting acoustic wave signals in a target area. The acoustic sensor array is arranged at surface monitoring points and underground monitoring points. The surface monitoring points include at least karst depressions and karst grooves. The underground monitoring points include karst caves, cave entrances, and underground river entrances. Furthermore, the signal processing unit is used to perform preprocessing, time-frequency analysis and feature extraction on the frequency domain feature data to obtain a multidimensional feature vector including time domain features, frequency domain features, acoustic features and statistical features; Further, the pattern recognition unit includes a pre-trained acoustic pattern recognition model for receiving the multi-dimensional feature vector and outputting recognition results of the type, degree and development trend of soil and water loss; Furthermore, the risk assessment unit is used to calculate the soil and water loss risk score based on the output result of the acoustic pattern recognition model, determine the risk level according to a preset early warning threshold, and generate a soil and water loss risk prediction result.

[0022] Furthermore, the risk assessment unit also includes: an early warning release module, which is used to automatically generate and release early warning information according to the risk score, and transmit the soil and water loss risk prediction result to the monitoring center.

[0023] Compared with the prior art, the present invention has the following beneficial effects: The present invention realizes real-time monitoring and early warning of soil erosion by deploying acoustic sensor arrays at key monitoring points in karst areas, collecting acoustic wave signals from the surface and underground environments at the same time, and combining the acoustic pattern recognition model of the deep learning algorithm. This method overcomes the limitations of traditional monitoring methods in karst areas and can effectively identify different types of soil erosion phenomena, including surface runoff, rainfall erosion, rock collapse, underground dissolution and underground river flow, and predict their development trends. Through the correction of the acoustic wave propagation model and the multi-level early warning mechanism, the monitoring accuracy and early warning timeliness are improved, providing a scientific basis and technical support for soil and water conservation in karst areas, and has strong practicality and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1The present invention is a flow chart of a method for monitoring soil and water loss in karst areas.

[0025] Figure 2 The present invention is a schematic diagram of the composition of a karst area soil and water loss monitoring system. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Example 1 like Figure 1 As shown, a method for monitoring soil and water loss in karst areas of the present invention is provided. The monitoring method obtains soil and water loss risk prediction results of a target area based on a deployed acoustic sensor array. The monitoring method comprises the following steps: Step S1, deploying an acoustic sensor array at key monitoring points in a target area to be monitored, wherein the acoustic sensor array includes a plurality of acoustic sensors for simultaneously collecting acoustic wave signals of the surface and underground environments.

[0028] The key monitoring points in the target area are divided into surface monitoring points and underground monitoring points. The surface monitoring points at least include: karst depressions and karst grooves; the underground monitoring points include: karst caves, cave entrances and underground river entrances.

[0029] Taking a typical karst area in a certain province as an example, a monitoring range of 25 square kilometers was selected, and an acoustic sensor array was deployed in the area; specifically, three high-sensitivity microphone sensors were deployed at each of the five surface monitoring points (including three karst depressions and two karst grooves), and the sampling frequency was set to 48kHz; two waterproof acoustic sensors were deployed at each of the three underground monitoring points (including one inside a karst cave, one cave entrance, and one underground river entrance), and the sampling frequency was set to 44.1kHz. The sensors were connected to the data collection center via a wireless network to achieve real-time data transmission.

[0030] Step S2, collecting the sound wave signals detected by the acoustic sensor array, and converting the sound wave signals into frequency domain feature data.

[0031] Take a rainstorm process as an example to collect acoustic wave signals. The rainstorm process lasted for 4 hours, with a cumulative rainfall of 85mm. During the rainstorm, the acoustic sensor array recorded about 14,400 seconds of acoustic wave data. The collected raw acoustic wave signals were preprocessed, including using a bandpass filter (cutoff frequency of 100Hz-15kHz) to remove environmental noise, applying an adaptive noise reduction algorithm to suppress wind noise interference, and enhancing effective signals through dynamic range compression. After preprocessing, the continuous acoustic wave signal was divided into 10-second analysis windows, with an overlap rate of 50% for each window, and a total of about 2,880 signal segments were obtained. Short-time Fourier transform (STFT) was used to perform time-frequency analysis on each signal segment, with the window length set to 2048 points and the frame shift set to 512 points to obtain frequency domain feature data.

[0032] Step S3: input the frequency domain feature data into a pre-trained acoustic pattern recognition model, where the model is based on a deep learning algorithm and is used to identify acoustic features related to soil erosion.

[0033] The extracted frequency domain feature data include: 13th order Mel frequency cepstral coefficients MFCC, spectral flux SF, spectral centroid SC and spectral bandwidth SBW, as well as statistical features (mean, variance, skewness, kurtosis and energy distribution of 10 frequency bands). For a 10-second signal segment, a total of 68-dimensional feature vectors were extracted. These feature vectors were input into a pre-trained acoustic pattern recognition model for analysis. The model adopts a CNN-LSTM hybrid architecture. The CNN part contains 3 convolutional layers (the convolution kernel sizes are 3×3, 5×5 and 7×7, and the number of filters is 32, 64 and 128, respectively), each of which is followed by a maximum pooling layer; the LSTM part contains 2 bidirectional LSTM layers, each containing 128 units; the fully connected layer contains 256 neurons, and the Dropout rate is 0.5 to prevent overfitting; the output layer uses a softmax function to map to the probability of 5 types of soil erosion (surface runoff, rainfall erosion, rock collapse, underground dissolution and underground river flow).

[0034] The model used 30,000 annotated sound wave samples from 10 different karst regions during the training phase, and the model performance was evaluated by 5-fold cross-validation, ultimately achieving a classification accuracy of 92.7% on the test set. The model parameters were updated using the Adam optimizer, with an initial learning rate set to 0.001, a 10% decay every 20 epochs, a batch size of 64, and a total of 100 epochs. To improve the generalization ability of the model in real environments, data augmentation techniques were used during training, including adding Gaussian white noise with different signal-to-noise ratios, random time stretching, and frequency offset.

[0035] Step S4, based on the output results of the acoustic pattern recognition model, determine the type, degree and development trend of soil erosion in the monitored area, and generate soil erosion risk prediction results.

[0036] For this rainstorm process, the results of the acoustic pattern recognition model analysis showed that: strong surface runoff signals were detected at the depression monitoring point with a confidence level of 0.95; moderate rainfall scouring signals were detected in the karst grooves with a confidence level of 0.87; small-scale rock collapse signals were detected at the entrance of a karst cave with a confidence level of 0.83; and a signal of a significant increase in underground river flow was detected at the entrance of the underground river channel with a confidence level of 0.91. Based on these recognition results, the risk scoring formula was used to calculate the soil and water loss risk of this rainstorm process.

[0037] For surface runoff: , , For rainfall washoff: , For rockfall: For underground dissolution: , For underground river flows: ; According to the risk scoring formula: Risk=0.552.

[0038] According to the preset warning thresholds (mild: 0.0-0.3; moderate: 0.3-0.5; severe: 0.5-0.7; catastrophic: >0.7), the soil erosion risk score of this rainstorm process is 0.552, which is a severe warning level. The system automatically generates warning information and sends it to the local soil and water conservation management department, and recommends the following emergency measures: increase the monitoring frequency to once every 30 minutes; send staff to the rock collapse site for on-site investigation; check the drainage system of the community downstream of the underground river; prepare temporary drainage facilities and sandbags and other protective materials.

[0039] During the calibration of the sound wave propagation model, a simulated sound source with a known position (including a frequency sweep signal of 20Hz-20kHz) was used for testing, and the sound pressure level attenuation was measured at different distances (10m, 50m, 100m, and 200m). The attenuation coefficient in the sound wave propagation model was adjusted based on the local terrain data and rock material parameters (mainly limestone, with a sound speed of about 6000m / s). The test results show that under conditions of 70% humidity and 25°C temperature, the actual attenuation coefficient of surface sound wave propagation is about 1.2 times the theoretical value, while the attenuation coefficient of sound wave propagation in underground caves is about 0.8 times the theoretical value. Based on these test data, the sound wave propagation model parameters were calibrated, the sound source positioning accuracy was improved, and the average positioning error was reduced from the original ±15m to ±5m.

[0040] In this monitoring example, the system worked continuously for 7 days, covering the complete monitoring cycle before, during and after the rainstorm. By analyzing the temporal changes of the acoustic wave characteristics during the rainstorm, it was found that there was a time lag of about 15 minutes between the surface runoff intensity and the rainfall intensity, and there was a time lag of about 40 minutes between the change in underground river flow and the surface runoff intensity. These temporal relationships provide a basis for formulating more accurate warning times. After the 7-day monitoring period, the technicians conducted an on-site inspection and found that the soil erosion area predicted by the system was 89% consistent with the actual situation, proving the effectiveness and accuracy of the monitoring method.

[0041] The process of extracting the frequency domain feature data includes: The original sound wave signal is preprocessed, and the preprocessing includes denoising, filtering and signal enhancement; the preprocessed signal is subjected to time-frequency analysis to extract time domain features and frequency domain features.

[0042] The acoustic features of the sound wave signal are calculated, and the acoustic features include: Mel frequency cepstral coefficients MFCC, spectral flux, spectral centroid and spectral bandwidth; the statistical features of the sound wave signal are extracted, and the statistical features include: mean, variance, skewness, kurtosis and energy distribution.

[0043] The acoustic features are calculated using short-time Fourier transform (STFT): ;in, is the original sound wave signal, Indicated in In the analysis window, the position in the window is The sample value of It is the sample index in the window, i.e. the number of FFT points, ranging from 0 to N-1; is the time frame index, indicating the analysis window; is the frame shift, which indicates the sample number offset between adjacent windows; k is the frequency index, which indicates the frequency point in the short-time Fourier transform STFT; is a window function; Represents the short-time Fourier transform coefficient of the kth frequency point on the mth time frame; The calculation formula of Mel frequency cepstral coefficient MFCC is: ;in, represents the number of Mel filter banks, represents the cepstral coefficient index; The calculation formula of spectral centroid SC is: ; The calculation formula of spectral flux SF is: ; Sound wave energy distribution The calculation formula is: ;in, represents the frequency band index, and Indicates The lower and upper frequency indices of the frequency band.

[0044] The acoustic pattern recognition model is a hybrid model of a deep convolutional neural network and a long short-term memory network, and its structure includes: an input layer, a convolution layer, a pooling layer, a long short-term memory network layer, a fully connected layer and a softmax layer; The input layer receives an acoustic feature vector, which is a multidimensional array including time domain features, frequency domain features, acoustic features and statistical features; the time domain features include the time series data of the original signal after preprocessing; the frequency domain features are the spectrum data obtained by short-time Fourier transform; the acoustic features include Mel frequency cepstral coefficients MFCC, spectral flux SF, spectral centroid SC and spectral bandwidth SBW; the statistical features include the mean, variance, skewness, kurtosis and energy distribution of each frequency band of the signal; The convolution layer extracts the local pattern of acoustic features, the pooling layer reduces the feature dimension and improves the model robustness, the long short-term memory network layer captures the temporal dependency of acoustic features, and the fully connected layer and softmax layer output the probability distribution of soil erosion type and degree; The mathematical expression of the convolutional layer operation is: ,in: is the output of the previous layer; is the convolution kernel weight; is the bias term; is the rectified linear unit activation function; Represents the convolution operation; Input gate of long short-term memory network LSTM: ; Forget Gate: , output gate: ; Candidate memory cells: ; Memory unit update: ; Hide status updates: ; in: is the input of the current time step, is the hidden state of the previous time step, is the memory cell state at the previous time step, , , , is the weight matrix, , , , is the bias vector, is the hyperbolic tangent activation function, is the Hadamard product, using element-by-element multiplication; Represents the sigmoid activation function; The expression of the output layer is: ;in, is the output of the LSTM layer; is the output layer weight matrix; is the output layer bias vector; is the normalized exponential function.

[0045] The acoustic pattern recognition model is trained using the cross entropy loss function: ;in, is the total number of soil erosion types, is the true label, Predict probabilities for the model.

[0046] The training steps of the acoustic pattern recognition model are: Step S31, collecting acoustic wave samples generated by different types of soil erosion processes, including acoustic wave samples under surface runoff, rainfall erosion, rock collapse, underground dissolution and underground river flow scenarios; Step S32, labeling the sound wave samples and establishing a corresponding relationship between the acoustic features and the type and degree of soil erosion; Step S33, using principal component analysis PCA and t-distributed random neighbor embedding t-SNE dimensionality reduction technology to extract the distinguishing features of the sound wave sample; The expression of the principal component analysis PCA is: ;in, is the original acoustic feature matrix, is the principal component loading matrix, is the feature matrix after dimensionality reduction; Calculated from the eigenvectors: ;in, is the covariance matrix of the original features, is the eigenvalue diagonal matrix; Use t-distributed random neighbor embedding to calculate conditional probabilities in high-dimensional space: ; Conditional probability in low-dimensional space: ; The optimization goal is to minimize the KL divergence: ; Step S34, using batch gradient descent method to optimize model parameters, the learning rate is initially set to 0.001, and a learning rate decay strategy is used; Cluster analysis uses the K-means algorithm: ;in, For sample Belongs to cluster indicator variable of Cluster the center of is the number of clusters, is the sample size; Step S35, using early stopping method to avoid overfitting, and stopping training when the performance of the validation set no longer improves; The method also includes a sound wave propagation model correction step: establishing a sound wave propagation model in a karst area, taking into account the influencing factors of topography, rock material, vegetation coverage and meteorological conditions, using a known sound source to conduct a test, and obtaining the propagation characteristics of sound waves in a karst environment; based on the test results, correcting the sound wave propagation model parameters, and using the corrected propagation model to improve the accuracy of sound source positioning and signal analysis; The sound wave propagation model is based on the wave equation: ;in, is the sound pressure, is the speed of sound, is the Laplace operator; Under complex terrain conditions, the ray tracing method is used to calculate the sound wave propagation path: ;in, is the sound wave ray path, is the path length parameter, is the refractive index of sound waves, which is related to the sound velocity of the medium. , is the reference sound speed; Sound wave attenuation model considering terrain and atmospheric conditions: ;in, is the sound pressure level at the receiving point; is the sound pressure level of the sound source; is the distance from the sound source to the receiving point; is the atmospheric absorption coefficient; is the attenuation caused by terrain; Attenuation caused by atmospheric conditions.

[0047] The present invention collected acoustic wave samples of five types of soil erosion. The sample data came from 30 monitoring points in three typical karst areas in Liupanshui City, Bijie City and Qiannan Prefecture, Guizhou Province, over a period of two years, covering different seasons and weather conditions. Specifically, there are 12,500 surface runoff samples, each lasting 15 seconds, collected from surface runoff processes after rainfalls of different intensities (light rain, moderate rain, heavy rain, and rainstorms); 10,800 rainfall erosion samples, each lasting 10 seconds, covering various rainfall intensities from drizzle to heavy rainstorms; 3,600 rock collapse samples, each lasting 5 seconds, including small (<0.5m³), medium (0.5-5m³) and large (>5m³) collapse events; 5,200 underground dissolution samples, each lasting 30 seconds, obtained through a combination of specially designed water dissolution experiments and field monitoring; 8,900 underground river flow samples, each lasting 20 seconds, collected at the entrance of underground rivers in different seasons and under different flow conditions. These samples were jointly reviewed and labeled by professional acoustic engineers and soil and water conservation experts to ensure label quality.

[0048] The acoustic sample labeling adopts a three-level labeling system. On-site staff will make preliminary labeling based on the actual situation at the time of collection, recording information such as the type, time, location and visually estimated intensity of soil erosion; acoustic experts will listen to and perform spectral analysis on the acoustic wave signals to verify and refine the preliminary labeling, and accurately mark the starting and ending points of the effective signals on the timeline; soil and water conservation experts will conduct a comprehensive assessment of the samples based on the hydrological and meteorological data collected during the same period and the results of field surveys to determine the precise type of soil erosion and the quantitative severity.

[0049] For example, for a rock collapse acoustic wave sample, the final annotation includes: basic information (collection time: 14:23:45 on July 15, 2022; collection location: northeast corner of Maolan Nature Reserve, Libo County, Qiannan Prefecture; equipment ID: KS-M07; weather conditions: thunderstorm; environmental noise: medium), acoustic characteristics (peak intensity: 82dB; main frequency band: 120-800Hz; duration: 2.3 seconds; signal-to-noise ratio: 18dB), soil erosion information (type: rock collapse; volume: about 2.8m³; collapse material: limestone; associated phenomenon: small landslide; severity coefficient: 0.75). This multi-level and multi-dimensional annotation system ensures the high quality and reliability of training data.

[0050] The feature dimension reduction process adopts a two-stage strategy to use principal component analysis (PCA) to reduce the original 68-dimensional acoustic features to 30 dimensions, retaining about 95% of the variance information; then the t-SNE algorithm is applied to further reduce the features to 3 dimensions for visualization analysis and clustering. In the PCA dimension reduction process, it was found that the first 10 principal components were mainly associated with time-frequency features (accounting for 62% of the total variance), the 11th to 20th principal components were mainly associated with statistical features (accounting for 24% of the total variance), and the 21st to 30th principal components were mainly associated with spectral features in acoustic features (accounting for 9% of the total variance). This shows that time-frequency features play a dominant role in the identification of soil erosion types, which is consistent with the physical characteristics of acoustic waves generated by the soil erosion process. Through t-SNE visualization, it is clearly observed that the five types of soil erosion formed relatively separated clusters in the feature space. Among them, the surface runoff type and the underground river flow type have some overlap, which is consistent with the similarity of their physical mechanisms; while the rock collapse type samples are the most dispersed in the feature space, reflecting the huge differences in the acoustic characteristics of collapse events of different scales.

[0051] The batch gradient descent method with momentum was used for model parameter optimization. The batch size was set to 128, the initial learning rate was 0.001, and the momentum coefficient was 0.9. To prevent gradient explosion, the gradient clipping technique was used to limit the gradient norm to less than 5. The learning rate decay strategy adopted cosine annealing, reducing the learning rate to half of the original value every 50 epochs. Multiple regularization techniques were applied during model training: the L2 regularization coefficient was set to 0.0005, and the Dropout rate was set to different values ​​in different layers (0.2 for the convolutional layer, 0.5 for the LSTM layer, and 0.6 for the fully connected layer). In order to deal with the problem of sample imbalance (such as the number of rock collapse samples is significantly less than that of other types), a weighted cross entropy loss function was used, and the weights were inversely proportional to the number of samples of each type. In the K-means clustering analysis, the silhouette coefficient was evaluated to determine the optimal number of clusters K=5, which is consistent with the predefined five types of soil and water loss. The clustering results show that the intra-class compactness and inter-class separation of different types of samples have reached an ideal level, and the clustering accuracy rate has reached 87.3%, which further verifies the effectiveness of the extracted features.

[0052] The specific implementation of the early stopping method is: after every 5 epochs of training, the model performance is evaluated on the validation set. If the F1 score of the model on the validation set does not improve significantly (the improvement is less than 0.1%) after 10 consecutive evaluations, the training is stopped. In this study, the model training usually converges at 120-150 epochs. In addition to the accuracy, other indicators such as precision, recall, F1 score and confusion matrix are calculated to comprehensively evaluate the model performance. The average precision of the five types of soil erosion is 89.6%, the average recall is 86.2%, and the average F1 score is 87.9%. From the confusion matrix analysis, it is found that there is a misclassification rate of about 8% between the surface runoff type and the underground river flow type, which is consistent with the feature overlap observed in the t-SNE visualization; the rock collapse type has the highest recognition accuracy (93.5%), which is attributed to its unique acoustic characteristics.

[0053] The model training of the present invention also considers the improvement of generalization ability and anti-interference ability. In order to enhance the adaptability of the model to different monitoring environments, a variety of background noises are introduced into the training data, including wind noise, biological noise (such as bird calls, insect sounds), man-made noise (such as traffic, construction), etc., and the signal-to-noise ratio ranges from -5dB to 20dB. Through data enhancement technology, the original training sample size has been expanded by 3 times, enhancing the generalization ability of the model. In order to verify the performance of the model in a new environment, test samples were collected in the Puzhehei Karst area of ​​Yunnan Province outside Guizhou Province, and the model recognition accuracy reached 83.7%, proving its good generalization ability. In addition, the present invention also realizes a continuous update mechanism of the model, and collects new sample data every quarter for model fine-tuning, so that the system can adapt to factors such as environmental changes and equipment aging, and maintain high recognition performance. In a two-year practical application, through quarterly updates, the average recognition accuracy of the model at the same monitoring point increased from the initial 89.3% to 94.1%, proving the effectiveness of continuous learning.

[0054] In terms of the hardware implementation of model training, the present invention adopts a distributed computing architecture, using 4 servers equipped with NVIDIA Tesla V100 GPUs for parallel training, and a single complete training takes about 12 hours. In order to meet the needs of edge computing, model pruned and quantized versions have also been developed, which reduce the number of parameters by 78%, increase the inference speed by 3.2 times, and can be deployed on field equipment with limited computing resources. At monitoring points without power supply in the wild, combined with solar power supply and low-power computing modules, the long-term autonomous operation of the system is achieved. The average delay in the model reasoning stage is only 236 milliseconds, which meets the needs of real-time monitoring. Through these technical means, the present invention realizes the optimization of the entire process from training to deployment, and provides an efficient and reliable artificial intelligence solution for soil and water loss monitoring in karst areas.

[0055] In step S4, based on the output of the acoustic pattern recognition model, a soil erosion risk prediction result is generated, and a soil erosion risk prediction score is generated. It is expressed as: ; For the Weight coefficient of soil erosion class; For the The probability of soil erosion occurring is For the the severity of soil erosion; Set multi-level warning thresholds based on risk scores, divided into four warning levels: minor, moderate, severe and catastrophic; When the risk score exceeds the warning threshold, the system automatically triggers the corresponding level of warning signal; According to different warning levels, corresponding emergency response measures are formulated, including increasing monitoring frequency, notifying relevant departments and activating emergency plans.

[0056] In the present invention, the types of soil erosion are classified in detail according to the acoustic characteristics, and are divided into five categories: surface runoff type, rainfall erosion type, rock collapse type, underground dissolution type and underground river flow type. The characteristics of surface runoff type soil erosion are low-frequency (20-200Hz) continuous sound wave signals with uniform spectral energy distribution, and the amplitude increases with the increase of runoff intensity; the characteristics of rainfall erosion type soil erosion are medium-frequency (200-1000Hz) irregular pulse sound wave signals with obvious raindrop impact sound spectrum characteristics, and the spectral energy varies with rainfall intensity; the characteristics of rock collapse type soil erosion are broadband (100-5000Hz) sudden high-energy sound wave signals with obvious impact and attenuation processes, and the spectral energy is mainly concentrated in the low-frequency band; the characteristics of underground dissolution type soil erosion are low-frequency (50-500Hz) continuous weak signals with special spectral characteristics of water flow and rock interaction; the characteristics of underground river flow type soil erosion are medium-low frequency (100-800Hz) continuous sound wave signals with typical water turbulence acoustic characteristics, and the spectral centroid moves to high frequency with increasing flow velocity.

[0057] Based on the above soil erosion types, the present invention sets detailed risk scoring standards and warning thresholds. Specifically, the severity , The quantitative standards are as follows: For surface runoff type, when the flow velocity is less than 0.5m / s, The value is 0.3; when the flow velocity is between 0.5-1.5m / s, The value is 0.6; when the flow velocity is greater than 1.5m / s, The value is 0.9. Rainfall erosion type, when the rainfall intensity is less than 30mm / h, The value is 0.4; when the rainfall intensity is between 30-60mm / h, The value is 0.7; when the rainfall intensity is greater than 60mm / h, The value is 0.95. Rock collapse type, according to the collapse volume, when it is less than 1m³, The value is 0.5; when the volume is between 1-10m³, Take the value as 0.8; when the volume is greater than 10m³, The value is 1.0. Underground dissolution type, according to the dissolution rate, when the rate is low (less than 5mm / year), The value is 0.3; at medium rate (5-20mm / year), The value is 0.6; at high rates (greater than 20 mm / year), The value is 0.8. Underground river flow type, according to the flow change rate, when the increase is less than 50%, The value is 0.4; when the growth is between 50-100%, The value is 0.7; when the growth is greater than 100%, The value is 0.9.

[0058] Weight coefficient The determination is based on the historical data statistics of soil and water loss in the karst area and the evaluation of expert experience, taking into account the degree of harm and frequency of occurrence of different types of soil and water loss. The specific value is: surface runoff type =0.30, rainfall wash type =0.25, rock fall type =0.20, underground dissolution type =0.10, underground river flow type = 0.15. The sum of these weighting coefficients equals 1, ensuring the normalization of the risk score.

[0059] The specific division of warning thresholds and the corresponding emergency response measures are as follows: when the risk score is between 0.0-0.3, it is judged as a minor warning level, and the emergency response measures include maintaining regular monitoring frequency, recording monitoring data, and no special intervention is required; when the risk score is between 0.3-0.5, it is judged as a moderate warning level, and the emergency response measures include increasing the monitoring frequency to once every hour, notifying local soil and water conservation departments to keep attention, and preparing necessary protective materials; when the risk score is between 0.5-0.7, it is judged as a severe warning level, and the emergency response measures include increasing the monitoring frequency to once every 30 minutes, dispatching technical personnel to high-risk areas for on-site inspections, notifying relevant communities and enterprises to make protective preparations, and activating the first-level emergency plan; when the risk score is greater than 0.7, it is judged as a catastrophic warning level, and the emergency response measures include increasing the monitoring frequency to once every 10 minutes, immediately notifying local governments and soil and water conservation, emergency management and other departments, considering personnel evacuation and traffic control measures, and fully activating emergency plans and rescue systems.

[0060] In the long-term monitoring practice, the present invention also establishes an adaptive adjustment mechanism for the warning threshold. By comparing the consistency between the warning results and the actual soil and water loss situation, the warning threshold parameters are regularly updated. In the rainy season, the warning threshold is appropriately lowered to enhance the sensitivity of the system; in the dry season, the warning threshold is appropriately increased to reduce the false alarm rate. In addition, according to the geological characteristics and soil and water loss history of different karst areas, the warning threshold is customized for each monitoring area to further improve the accuracy and practicality of the warning.

[0061] For example, for areas with loose geological structure and severe soil erosion in history, the threshold of the severe warning level can be appropriately lowered to 0.45-0.65; while for areas with stable geology and complete soil and water conservation measures, the threshold can be appropriately raised to 0.55-0.75.

[0062] To verify the effectiveness of the early warning system, the present invention was tested and applied for one year in three typical karst small watersheds in a county in a province.

[0063] The results show that the matching degree between the warning information generated by the system and the actual soil and water loss events is 82%, of which the matching degree of severe and catastrophic warning levels is as high as 91%. The warning information can be issued on average 40 minutes in advance, providing valuable decision-making time for local soil and water conservation management and disaster prevention and mitigation. Through this year's application, the system has issued 182 warnings, including 98 minor warnings, 56 moderate warnings, 22 severe warnings, and 6 catastrophic warnings, which effectively guided the local soil and water conservation work and avoided many soil and water loss disasters that could have caused serious consequences.

[0064] The method also includes the step of calculating the amount of soil loss based on acoustic characteristics. The method also includes the step of calculating the amount of soil loss based on acoustic characteristics. By analyzing the intensity, frequency distribution and time domain variation characteristics of the sound wave signal, a quantitative relationship model between the acoustic characteristics and the amount of soil loss is established to achieve an accurate assessment of the degree of soil erosion in the karst area.

[0065] Calculate soil loss using acoustic signal characteristics monitored by acoustic sensors , whose expression is: ;in, is the amount of soil loss, in tons / hectare; \kappa is the calibration coefficient related to soil type; and is the start and end time of monitoring; is the frequency-dependent soil loss transfer function; is the sound wave signal intensity in the time-frequency domain.

[0066] The conversion function Determined according to different soil types and particle size distribution:

[0067] in, is the base conversion factor; For the Weight coefficient of soil particle size; For the The characteristic frequency of soil particle size; is the frequency distribution width; M is the number of soil particle size classifications.

[0068] The acoustic signal strength It is positively correlated with the scouring intensity during soil erosion and is calculated by the power spectral density of short-time Fourier transform: ;in, For the On the time frame The short-time Fourier transform coefficients of the frequency points are Corresponding to the frequency index , Corresponds to the timeframe index ; The system calculates the amount of soil loss With preset threshold The comparison results are used to adjust the soil and water loss risk prediction score. The risk score increases ,in is the adjustment factor, This is the largest amount of soil loss in history.

[0069] The present invention also provides a soil loss assessment method based on acoustic characteristics. The method uses the physical correlation between the characteristics of the acoustic wave signal and the soil scouring process to establish a quantitative relationship model between acoustic parameters and soil loss. Studies have shown that the sound waves generated by soil particles of different particle sizes during the scouring process have different frequency characteristics: fine sand mainly produces 800-1200Hz sound waves, coarse sand mainly produces 400-800Hz sound waves, small-particle soil aggregates mainly produce 200-400Hz sound waves, and large-particle soil aggregates mainly produce 100-200Hz sound waves. Based on this characteristic, the present invention can identify the loss characteristics of soils of different components by analyzing the spectral distribution of the sound wave signal.

[0070] In practical applications, the system first establishes a standard soil sample library, which contains the acoustic characteristic data of typical soil types in the study area. Through controlled experiments, the acoustic wave signals generated when water flows of different intensities flush different types of soil are measured, and the corresponding soil loss is recorded to establish a calibration curve. Calibration coefficient Determined by factors such as soil type, moisture content, and compactness, usually between 0.5-2.0. Frequency conversion function It reflects the corresponding relationship between acoustic wave signals of different frequencies and soil loss. The parameters in this function are obtained by fitting experimental data.

[0071] Taking a certain study area as an example, when monitoring a 4-hour heavy rainfall process, the acoustic signal intensity of a karst depression monitoring point increased significantly in the 400-800Hz frequency band, and the average power spectrum density reached 8 times the normal level. The system calculated the soil loss at the monitoring point based on the pre-established model and concluded that it reached 4.3 tons / hectare, exceeding the warning threshold of 3.0 tons / hectare. Based on this, the risk assessment unit raised the risk score by 0.15, and the final risk score reached 0.65, triggering a severe level of warning signal.

[0072] In the two years since the system was deployed, 32 soil loss assessment results were compared with field measurements, with an assessment accuracy of 88.4% and an average relative error of 12.3%. This result shows that the soil loss assessment method based on acoustic characteristics has good accuracy and practicality, and can provide an important quantitative basis for soil and water conservation work in karst areas.

[0073] Example 2 like Figure 2 As shown, it is a schematic diagram of the composition of a karst area soil and water loss monitoring system of the present invention, which is used to execute the monitoring method of Example 1. The monitoring system includes: a signal acquisition unit, a signal processing unit, a pattern recognition unit and a risk assessment unit connected in sequence.

[0074] The signal acquisition unit is used to receive the sound wave signal collected by the acoustic sensor array and convert the sound wave signal into frequency domain feature data.

[0075] The acoustic sensor array includes multiple acoustic sensors for collecting sound wave signals in the target area. The acoustic sensor array is set at surface monitoring points and underground monitoring points. The surface monitoring points include at least karst depressions and dissolution grooves, and the underground monitoring points include karst caves, cave entrances, and underground river entrances.

[0076] The signal processing unit is used to perform preprocessing, time-frequency analysis and feature extraction on the frequency domain feature data to obtain a multidimensional feature vector including time domain features, frequency domain features, acoustic features and statistical features.

[0077] The pattern recognition unit includes a pre-trained acoustic pattern recognition model, which is used to receive the multi-dimensional feature vector, output the recognition results of the type, degree and development trend of soil and water loss, and calculate the amount of soil loss.

[0078] The risk assessment unit is used to calculate the soil and water loss risk score based on the output result of the acoustic pattern recognition model, and to determine the risk level according to a preset early warning threshold value in combination with the soil loss amount assessment result to generate a soil and water loss risk prediction result.

[0079] The risk assessment unit also includes: an early warning release module, which is used to automatically generate and release early warning information according to the risk score, and transmit the soil and water loss risk prediction result to a monitoring center.

[0080] The signal acquisition unit consists of an acoustic sensor array deployed on site and a data preprocessing module. The acoustic sensor array uses three types of acoustic sensors: omnidirectional microphones for surface monitoring (frequency range: 20Hz-20kHz, sensitivity: -38dB, signal-to-noise ratio: >68dB), with a sampling rate set to 48kHz; low-frequency sensors for cave interior monitoring (frequency range: 5Hz-10kHz, sensitivity: -35dB, signal-to-noise ratio: >72dB), with a sampling rate set to 44.1kHz; hydrophones for underwater environmental monitoring (frequency range: 10Hz-15kHz, sensitivity: -160dB re 1V / μPa, maximum operating depth: 50m), with a sampling rate set to 32kHz. All sensors are designed with IP67 waterproof housings and built-in dust and moisture-proof filters to adapt to the high humidity environment of the karst area.

[0081] The data transmission adopts a multi-level architecture: the internal sensors of the monitoring point are connected to the on-site edge computing unit via low-power Bluetooth (BLE 5.0); the edge computing unit transmits the data to the cloud server via 4G / 5G mobile network or LoRa long-distance IoT technology. In remote areas with poor network coverage, a satellite communication backup link is equipped to ensure the reliability of data transmission. To reduce the transmission bandwidth requirements, the edge computing unit performs preliminary signal compression and feature extraction, and only transmits key data, which can reduce the data transmission volume by 90% on average. The edge computing unit is designed based on the ARM Cortex-A72 processor, equipped with 2GB RAM and 32GB flash memory, supports the TensorFlow Lite framework, and can perform lightweight model reasoning locally. The unit adopts a low-power design, equipped with a 100W solar panel and a 26800mAh lithium battery pack, and can work continuously for 14 days without an external power supply.

[0082] The signal processing unit is deployed on a cloud server, using a dual-core Intel Xeon processor and 32GB memory configuration, responsible for high-computational signal processing tasks. The unit first performs denoising on the received sound wave signal, using a combined filtering strategy: using a Wiener filter for environmental stable noise; using a median filter for impulsive noise; and using an adaptive wavelet threshold filter for random noise. The signal-to-noise ratio of the denoised signal is increased by an average of 12dB. The time-frequency analysis module uses a combination of short-time Fourier transform and wavelet packet transform to capture the short-time and frequency characteristics of the signal at the same time. The feature extraction module uses a parallel computing architecture to process multiple signal fragments at the same time, and the computing speed is 8 times faster than that of a single-thread implementation.

[0083] The pattern recognition unit implements a two-level recognition architecture: the first level is soil erosion type recognition, using a CNN-LSTM hybrid model to identify five basic soil erosion types; the second level is severity assessment, using a regression model to output a severity coefficient between 0 and 1. To improve the robustness of the system, the unit also implements a model integration mechanism, running three independently trained models (CNN-LSTM, pure CNN, and gradient boosting tree) at the same time, and using a weighted voting strategy to determine the final recognition result. The model weights are dynamically adjusted based on the historical performance of each model under different environmental conditions. The unit also implements an incremental learning function, updating the model with newly collected labeled data every month to ensure continuous optimization of system performance. In actual applications, the model accuracy has gradually increased from 88% at the initial deployment to 95% after one year.

[0084] The pattern recognition unit also includes a soil loss assessment module, which is used to calculate the soil loss according to the acoustic feature data, and the calculation formula is:

[0085] A mapping relationship between acoustic characteristics of different frequencies and the soil loss process is established, and the multi-channel acoustic parameter analysis method is used to identify the erosion characteristic frequencies of soil particles of different particle sizes. The soil erosion area and impact range are estimated through the spatial distribution analysis of acoustic energy combined with the topographic and geomorphological characteristics of the karst area. The spatiotemporal distribution map of soil loss is generated to provide a decision-making basis for the risk assessment unit.

[0086] The risk assessment unit is the decision-making center of the system, responsible for calculating the soil erosion risk score based on the identification results and generating early warning information. The unit adopts a combination of rule-based expert systems and statistical models, taking into account acoustic identification results, historical data, terrain features and meteorological conditions. The risk assessment process introduces spatiotemporal correlation analysis: if multiple adjacent monitoring points detect soil erosion signals at the same time, the risk score will automatically increase; if the current identification results are significantly inconsistent with the historical pattern, the system will start a cross-validation mechanism to prevent false alarms. The early warning release module distributes early warning information through multiple channels: sending detailed technical reports to the administrator of the monitoring center; sending mobile phone text messages and application push notifications to field staff; sending standard format early warning notifications to local government departments; and publishing early warning information suitable for ordinary people to understand on public platforms. The early warning information contains elements such as risk level, impact range, development trend, recommended measures and validity period. The content is automatically adjusted according to the recipient to ensure the effectiveness and operability of information communication.

[0087] In terms of system operation and maintenance, this embodiment has designed a complete self-diagnosis and remote maintenance mechanism. Each hardware component is equipped with a health monitoring module to regularly check the battery power, storage space, network connection and sensor status. When an abnormality is detected, the system automatically records the diagnostic log and sends an alarm to the maintenance personnel. For common faults, the system can execute automatic recovery procedures, such as restarting the communication module, switching the backup power supply, or starting the backup sensor. The remote maintenance interface allows technicians to access on-site devices through an encrypted connection to perform firmware updates, parameter adjustments and troubleshooting, greatly reducing the need for on-site maintenance. Statistics show that the system's mean trouble-free operation time (MTBF) reaches 8760 hours (one year), the mean time to recover from a fault (MTTR) is less than 24 hours, and the annual availability exceeds 99.7%.

[0088] In actual application, this system deployed 25 monitoring points in a county in a certain province in 2021, covering a typical karst area of ​​about 50 square kilometers. During the 18-month operation period, the system issued a total of 253 warnings, including 126 minor levels, 84 moderate levels, 37 severe levels, and 6 catastrophic levels. Compared with the results of manual on-site surveys, the warning accuracy rate was 92.1%. In particular, in a heavy rainstorm in June 2022, the system warned of an impending large-scale rock collapse 45 minutes in advance, enabling local management departments to close the dangerous area in time and avoid potential casualties. The system also discovered three previously unknown active areas of underground cave development, providing valuable data for karst geological research. Comprehensive evaluation shows that after the system is deployed, the timeliness of responding to soil erosion incidents in the monitoring area has increased by 300%, the efficiency of emergency disposal has increased by 150%, and the related economic losses have decreased by about 65%, which fully proves the practical application value and social benefits of the present invention.

[0089] The system's scalability design allows for the flexible addition of new monitoring points and sensor types without changing the core architecture. The cloud platform supports parallel data processing of up to 1,000 monitoring points, and the single-point data processing delay is controlled within 100 milliseconds. The system also reserves external data interfaces that can be integrated with other monitoring systems such as meteorological stations, hydrological stations, and earthquake monitoring networks to achieve data interoperability and comprehensive analysis. This open architecture lays the foundation for future system function expansion and application scenario expansion, making the system not only suitable for soil and water loss monitoring, but also scalable to geological disaster early warning, ecological environmental protection, and smart city construction.

[0090] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring soil and water loss in karst areas, wherein the method obtains soil and water loss risk prediction results of a target area based on a deployed acoustic sensor array, and is characterized in that: The monitoring method comprises the following steps: Step S1, deploying an acoustic sensor array at key monitoring points in a target area to be monitored, wherein the acoustic sensor array includes a plurality of acoustic sensors for simultaneously collecting acoustic wave signals of the surface and underground environments; The key monitoring points in the target area are divided into surface monitoring points and underground monitoring points. The surface monitoring points include at least karst depressions and karst grooves; the underground monitoring points include karst caves, cave entrances and underground river entrances. Step S2, collecting the sound wave signals detected by the acoustic sensor array, and converting the sound wave signals into frequency domain feature data; Step S3, inputting the frequency domain feature data into a pre-trained acoustic pattern recognition model, wherein the model is based on a deep learning algorithm and is used to identify acoustic features related to soil erosion; Step S4, based on the output results of the acoustic pattern recognition model, determine the type, degree and development trend of soil erosion in the monitored area, and generate soil erosion risk prediction results.

2. The monitoring method according to claim 1, characterized in that: The process of extracting the frequency domain feature data includes: Preprocessing the original sound wave signal, wherein the preprocessing includes denoising, filtering and signal enhancement; performing time-frequency analysis on the preprocessed signal to extract time domain features and frequency domain features; The acoustic features of the sound wave signal are calculated, and the acoustic features include: Mel frequency cepstral coefficients MFCC, spectral flux, spectral centroid and spectral bandwidth; the statistical features of the sound wave signal are extracted, and the statistical features include: mean, variance, skewness, kurtosis and energy distribution.

3. The monitoring method according to claim 2, characterized in that: The acoustic features are calculated using short-time Fourier transform (STFT): ;in, is the original sound wave signal, Indicated in In the analysis window, the position in the window is The sample value of It is the sample index in the window, i.e. the number of FFT points, ranging from 0 to N-1; is the time frame index, indicating the analysis window; is the frame shift, which indicates the sample number offset between adjacent windows; k is the frequency index, which indicates the frequency point in the short-time Fourier transform STFT; is a window function; Represents the short-time Fourier transform coefficient of the kth frequency point on the mth time frame; The calculation formula of Mel frequency cepstral coefficient MFCC is: ;in, represents the number of Mel filter banks, represents the cepstral coefficient index; The calculation formula of spectral centroid SC is: ; The calculation formula of spectral flux SF is: ; Sound wave energy distribution The calculation formula is: ;in, represents the frequency band index, and Indicates The lower and upper frequency indices of the frequency band.

4. The monitoring method according to claim 3, characterized in that: The acoustic pattern recognition model is a hybrid model of a deep convolutional neural network and a long short-term memory network, and its structure includes: an input layer, a convolution layer, a pooling layer, a long short-term memory network layer, a fully connected layer and a softmax layer; The input layer receives an acoustic feature vector, which is a multidimensional array including time domain features, frequency domain features, acoustic features and statistical features; the time domain features include the time series data of the original signal after preprocessing; the frequency domain features are the spectrum data obtained by short-time Fourier transform; the acoustic features include Mel frequency cepstral coefficients MFCC, spectral flux SF, spectral centroid SC and spectral bandwidth SBW; the statistical features include the mean, variance, skewness, kurtosis and energy distribution of each frequency band of the signal; The convolution layer extracts the local pattern of acoustic features, the pooling layer reduces the feature dimension and improves the model robustness, the long short-term memory network layer captures the temporal dependency of acoustic features, and the fully connected layer and softmax layer output the probability distribution of soil erosion type and degree; The mathematical expression of the convolutional layer operation is: ,in: is the output of the previous layer; is the convolution kernel weight; is the bias term; is the activation function of the rectified linear unit; * indicates the convolution operation; Input gate of long short-term memory network LSTM: ; Forget Gate: , output gate: ; Candidate memory cells: ; Memory unit update: ; Hide status updates: ; in: is the input of the current time step, is the hidden state of the previous time step, is the memory cell state at the previous time step, , , , is the weight matrix, , , , is the bias vector, is the hyperbolic tangent activation function, is the Hadamard product, using element-by-element multiplication; Represents the sigmoid activation function; The expression of the output layer is: ;in, is the output of the LSTM layer; is the output layer weight matrix; is the output layer bias vector; is the normalized exponential function.

5. The monitoring method according to claim 4, characterized in that: The acoustic pattern recognition model is trained using the cross entropy loss function: ;in, is the total number of soil erosion types, is the true label, Predict probabilities for the model.

6. The monitoring method according to claim 5, characterized in that: The training steps of the acoustic pattern recognition model are: Step S31, collecting acoustic wave samples generated by different types of soil erosion processes, including acoustic wave samples under surface runoff, rainfall erosion, rock collapse, underground dissolution and underground river flow scenarios; Step S32, labeling the sound wave samples and establishing a corresponding relationship between the acoustic features and the type and degree of soil erosion; Step S33, using principal component analysis PCA and t-distributed random neighbor embedding t-SNE dimensionality reduction technology to extract the distinguishing features of the sound wave sample; The expression of the principal component analysis PCA is: ;in, is the original acoustic feature matrix, is the principal component loading matrix, is the feature matrix after dimensionality reduction; Calculated from the eigenvector: ;in, is the covariance matrix of the original features, is the eigenvalue diagonal matrix; Use t-distributed random neighbor embedding to calculate conditional probabilities in high-dimensional space: ; Conditional probability in low-dimensional space: ; The optimization goal is to minimize the KL divergence: ; Step S34, using batch gradient descent method to optimize model parameters, the learning rate is initially set to 0.001, and a learning rate decay strategy is used; Cluster analysis uses the K-means algorithm: ;in, For sample Belongs to cluster indicator variable of Cluster the center of is the number of clusters, is the sample size; Step S35, use the early stopping method to avoid overfitting, and stop training when the performance of the validation set no longer improves.

7. The monitoring method according to claim 6, characterized in that: The method also includes a sound wave propagation model correction step: establishing a sound wave propagation model in a karst area, taking into account the influencing factors of topography, rock material, vegetation coverage and meteorological conditions, using a known sound source to conduct a test, and obtaining the propagation characteristics of sound waves in a karst environment; based on the test results, correcting the sound wave propagation model parameters, and using the corrected propagation model to improve the accuracy of sound source positioning and signal analysis; The sound wave propagation model is based on the wave equation: ;in, is the sound pressure, is the speed of sound, is the Laplace operator; Under complex terrain conditions, the ray tracing method is used to calculate the sound wave propagation path: ;in, is the sound wave ray path, is the path length parameter, is the refractive index of sound waves, which is related to the sound velocity of the medium. , is the reference sound speed; Sound wave attenuation model considering terrain and atmospheric conditions: ;in, is the sound pressure level at the receiving point; is the sound pressure level of the sound source; is the distance from the sound source to the receiving point; is the atmospheric absorption coefficient; Attenuation caused by terrain; Attenuation caused by atmospheric conditions.

8. The monitoring method according to claim 7, characterized in that: In step S4, based on the output of the acoustic pattern recognition model, a soil erosion risk prediction result is generated, and a soil erosion risk prediction score is generated. It is expressed as: ; For the Weight coefficient of soil erosion class; For the The probability of soil erosion occurring is For the the severity of soil erosion; According to the risk score, there are four warning levels: mild, moderate, severe and catastrophic; When the risk score exceeds the warning threshold, the system automatically triggers a warning signal of the corresponding level.

9. A karst area soil erosion monitoring system, used to implement the monitoring method according to any one of claims 1 to 8, characterized in that: The monitoring system comprises: a signal acquisition unit, a signal processing unit, a pattern recognition unit and a risk assessment unit connected in sequence; The signal acquisition unit is used to receive the sound wave signal collected by the acoustic sensor array and convert the sound wave signal into frequency domain feature data; The acoustic sensor array includes a plurality of acoustic sensors for collecting acoustic wave signals in a target area. The acoustic sensor array is arranged at surface monitoring points and underground monitoring points. The surface monitoring points include at least karst depressions and karst grooves. The underground monitoring points include karst caves, cave entrances, and underground river entrances. The signal processing unit is used to perform preprocessing, time-frequency analysis and feature extraction on the frequency domain feature data to obtain a multidimensional feature vector including time domain features, frequency domain features, acoustic features and statistical features; The pattern recognition unit includes a pre-trained acoustic pattern recognition model for receiving the multi-dimensional feature vector and outputting recognition results of the type, degree and development trend of soil and water loss; The risk assessment unit is used to calculate the soil and water loss risk score based on the output result of the acoustic pattern recognition model, determine the risk level according to a preset early warning threshold, and generate a soil and water loss risk prediction result.

10. The monitoring system according to claim 9, characterized in that: The risk assessment unit also includes: an early warning release module, which is used to automatically generate and release early warning information according to the risk score, and transmit the soil and water loss risk prediction result to a monitoring center.

Images