Deep hole ultrasonic stress testing rock-soil body noise reduction conversion method and system

Abstract

This invention belongs to the field of geotechnical engineering and geomechanical testing technology, specifically disclosing a method and system for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses. The method includes: acquiring the original ultrasonic time-domain signal and auxiliary environmental data in a deep-hole environment; extracting the multi-dimensional feature vector of the original ultrasonic time-domain signal to identify the dominant noise type and intensity level in the original ultrasonic time-domain signal; selecting a filtering algorithm to filter the original ultrasonic time-domain signal according to the dominant noise type and intensity level to obtain the filtered signal; and reconstructing and enhancing the filtered signal to obtain a noise-reduced signal, thus completing the noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses. This invention solves the problems of insufficient consideration of noise source characteristics, fixed filtering parameters, poor adaptability, easy distortion or loss of effective signals, and inability to effectively process geological noise in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of geotechnical engineering and geomechanical testing technology, specifically relating to a method and system for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses. Background Technology

[0002] Deep-hole in-situ stress testing is a key technology for assessing the stability of soil and rock masses, providing early warning of geological hazards, and guiding underground engineering construction. Ultrasonic testing, due to its non-destructive nature, high efficiency, and sensitivity to micro-fractures, is widely used in deep-hole stress testing. Its principle is to invert the stress state by measuring changes in parameters such as wave velocity, amplitude, and spectrum of ultrasonic waves propagating in the soil and rock mass. However, the deep-hole testing environment is extremely complex, and the acoustic signals are severely interfered with by various noises during acquisition and transmission, resulting in a low signal-to-noise ratio (SNR) and making it difficult to guarantee testing accuracy and reliability. The main noise sources include: 1. Mechanical noise: Low-frequency vibration noise caused by drilling operations, probe movement friction, borehole wall collapse, etc.

[0003] 2. Electromagnetic noise: Power frequency and harmonic interference induced by the circuit of the testing instrument itself, the cable inside the hole and the ground.

[0004] 3. Geological noise: The inherent joints, fissures, and inhomogeneities within the rock mass cause wave scattering and reflection, forming chaotic echoes.

[0005] 4. Environmental noise: Background noise transmitted through the strata by distant construction, vehicle traffic, water flow, etc.

[0006] 5. Instrument background noise: thermal noise from electronic components such as sensors and amplifiers.

[0007] Existing noise reduction methods mostly employ single conventional filtering techniques (such as bandpass filtering and wavelet threshold denoising), but these methods have obvious drawbacks: 1) insufficient consideration of the characteristics of noise sources, fixed filtering parameters, and poor adaptability; 2) while suppressing noise, they are prone to distortion or loss of effective signals (especially the stress information carried by the first wave and low frequencies); 3) lack of effective means to process geological noise.

[0008] Therefore, developing a noise reduction and conversion method that can intelligently identify noise sources, adaptively select and optimize filtering strategies, and ultimately significantly improve the signal-to-noise ratio of deep hole ultrasonic stress test signals has become a technical challenge that urgently needs to be solved in this field. Summary of the Invention

[0009] The purpose of this invention is to address the problems of insufficient consideration of noise source characteristics, fixed filtering parameters, poor adaptability, easy distortion or loss of effective signals, and inability to effectively process geological noise in existing technologies. This invention proposes a noise reduction and conversion method and system for deep-hole ultrasonic stress testing of soil and rock.

[0010] The technical solution of the present invention is as follows: Firstly, a method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses, comprising the following steps: The original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment are collected, and the multi-dimensional feature vector of the original ultrasonic time-domain signal is extracted. Based on the multi-dimensional feature vector and auxiliary environmental data, the dominant noise type and intensity level in the current original ultrasonic time-domain signal are identified and output using a noise classification and identification model and a multi-source noise feature database. Based on the dominant noise type and intensity level, the corresponding filtering algorithm is dynamically selected or combined from the preset adaptive filter library to filter the original ultrasonic time domain signal and obtain the filtered signal. The filtered signal is reconstructed and enhanced to obtain a noise-reduced signal, thus completing the noise reduction conversion of the rock and soil body for deep hole ultrasonic stress testing.

[0011] As a preferred option, the multi-source noise feature database includes typical feature templates for mechanical noise, electromagnetic noise, geological noise, environmental noise, and mixed noise.

[0012] As a preferred option, the noise classification and recognition model employs an improved dual-stream multi-domain attention fusion network; the improved dual-stream multi-domain attention fusion network includes a dual-stream feature extraction path module, a cross-domain attention fusion module, and an adaptive classification output module connected in sequence; The dual-stream feature extraction path module is used to extract local detail features and long-range dependency features of the original ultrasonic time-domain signal through one-dimensional depthwise separable convolution and dilated convolution; and to extract frequency domain energy distribution features and harmonic features through short-time Fourier transform and two-dimensional convolutional network. The cross-domain attention fusion module is used to receive local detail features, long-range dependency features, frequency domain energy distribution features, and harmonic features, calculate the importance weight of each feature, and adaptively fuse the features based on the importance weight to output the adaptive fused features. The adaptive classification output module includes a global adaptive pooling layer, a fully connected layer, and a Softmax layer connected in sequence. It is used to process the adaptive fusion features to obtain the initial noise type and intensity level. Then, it combines the typical feature templates in the multi-source noise feature database to enhance the initial noise type and intensity level, and outputs the final dominant noise type and intensity level.

[0013] Preferably, the preset adaptive filter library includes: An improved variational mode decomposition filter adaptively decomposes the original ultrasonic time-domain signal into intrinsic mode functions by optimizing the mode number K and penalty factor α of variational mode decomposition. The notch filter adaptive spectral enhancement cascade filter uses an adaptive notch filter to accurately filter out the power frequency and power frequency harmonics in the original ultrasonic time domain signal, and then uses an ALE filter to track and suppress residual narrowband periodic interference. The wavelet packet singular value decomposition combined noise denoiser utilizes wavelet packet transform to achieve a sparse representation of the original ultrasonic time-domain signal under the optimal basis, and combines singular value decomposition to perform rank estimation and dimensionality reduction on the coefficient matrix to separate the noise subspace and the signal subspace.

[0014] As a preferred method, the specific approach of dynamically selecting or combining corresponding filtering algorithms from a preset adaptive filter library based on the dominant noise type and intensity level is as follows: If the noise is identified as mechanical or geological scattering, an improved variational mode decomposition filter is selected, and the parameters of the improved variational mode decomposition filter are adaptively adjusted according to the bandwidth of the original ultrasonic time-domain signal. If electromagnetic noise is identified, the notch adaptive spectral enhancement cascade filter is invoked, and the notch order and step size of the notch adaptive spectral enhancement cascade filter are dynamically optimized. If the noise is identified as environmental noise or instrument background noise, a wavelet packet singular value decomposition joint denoiser is used, and the wavelet packet basis and the number of decomposition layers are adaptively selected. If mixed noise is identified, the improved variational mode decomposition filter, the notch adaptive spectral enhancement cascade filter, and the wavelet packet singular value decomposition joint denoiser are activated in parallel for processing, and the final filtered signal is output through blind source separation and weighted fusion.

[0015] As a preferred method, the specific steps for reconstructing and enhancing the filtered signal to obtain the effective signal are as follows: The filtered signal is reconstructed in reverse to obtain the initial denoised signal; When the initial noise reduction signal From real and effective ultrasonic signals It is formed by linear mixing with residual noise n, that is, the initial noise reduction signal is ,in For the reduced-dimensional observation matrix, Represents the observation matrix. Represents the real number field. Indicates the observation dimension. Indicates the signal dimension; And the real and effective ultrasonic signal x has a certain complete dictionary The following has a sparse representation, that is , where the sparse coefficient vector There are only a few non-zero elements in it; The signal reconstruction problem is transformed into recovering the sparse coefficients a from the initial denoised signal y, and then solving... L The problem of minimizing the L1 norm is used to obtain estimates of the sparse coefficients. The denoised signal calculated based on the sparse coefficient estimate is:

[0016] in, This represents the estimated value of the denoised signal. This represents the estimated value of the sparse coefficients.

[0017] As a preferred option L The problem of minimizing the 1-norm is specifically as follows:

[0018] in, This represents the regularization parameter used to balance reconstruction error with coefficient sparsity. Indicates to make The sparsity coefficient 'a' corresponding to the minimum value is Represents the squared terms of the L2 norm. This represents the L1 norm term.

[0019] Preferably, the method further includes: The quality index of the denoised signal is calculated. If the quality index does not reach the preset threshold, the multi-dimensional feature vectors of the denoised signal and the original ultrasonic time domain signal are fed back to the noise classification and identification model and the filtering algorithm for iterative optimization.

[0020] The beneficial effects of this invention are: 1. Intelligent identification: For the first time in deep hole acoustic wave testing, a multi-source noise feature database was systematically constructed and an intelligent classification model was introduced, realizing the "cause diagnosis" of complex noise environments and laying the foundation for subsequent precise "treatment".

[0021] 2. Adaptive Filtering: It changes the traditional fixed parameter filtering mode and proposes an adaptive mechanism of "diagnostic result-driven filtering strategy". It dynamically calls or fuses the optimal algorithm for the dominant noise type, which is highly targeted and efficient in noise reduction.

[0022] 3. High-fidelity enhancement: Through the blind source separation and sparse reconstruction technology of the post-processing module, while powerfully reducing noise, the effective signal components reflecting the stress state of the soil and rock mass (such as first arrival time and specific frequency band energy) are protected to the maximum extent, thus avoiding signal distortion.

[0023] 4. Closed-loop optimization: A feedback iteration mechanism based on signal quality assessment is introduced to improve the robustness and adaptability of the system under different geological conditions and operating environments.

[0024] 5. Strong engineering applicability: This method can be integrated into existing deep-hole ultrasonic stress testing equipment, and its functionality can be enhanced through software upgrades, significantly improving the quality of raw data and providing a reliable data foundation for high-precision stress inversion.

[0025] Secondly, a noise reduction and conversion system for deep-hole ultrasonic stress testing of soil and rock masses includes: The noise source identification and classification module is used to collect the original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment, and extract the multi-dimensional feature vector of the original ultrasonic time-domain signal. Based on the multi-dimensional feature vector and auxiliary environmental data, it uses a pre-trained classification model and a multi-source noise feature database to identify and output the dominant noise type and intensity level in the current original ultrasonic time-domain signal. The acoustic filtering algorithm module is used to dynamically select or combine corresponding filtering algorithms from a preset adaptive filter library according to the dominant noise type and intensity level, to filter the original ultrasonic time-domain signal and obtain the filtered signal. The post-processing module is used to reconstruct and enhance the filtered signal to obtain a noise-reduced signal, thus completing the noise reduction conversion of the deep-hole ultrasonic stress test soil and rock mass.

[0026] Thirdly, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method as described in the first aspect. Attached Figure Description

[0027] Figure 1 The diagram shows a flowchart of a method for noise reduction and conversion of rock and soil in deep-hole ultrasonic stress testing.

[0028] Figure 2 The diagram shows the structure of the improved noise classification and recognition model (DS-MDAFNet).

[0029] Figure 3 The diagram shows a structural diagram of a deep-hole ultrasonic stress testing noise reduction and conversion system for soil and rock masses. Detailed Implementation

[0030] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0031] Example 1: like Figure 1As shown, a method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses includes the following steps: S1. Collect the original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment, and extract the multi-dimensional feature vector of the original ultrasonic time-domain signal. Based on the multi-dimensional feature vector and auxiliary environmental data, use the noise classification and identification model and multi-source noise feature database to identify and output the dominant noise type and intensity level in the current original ultrasonic time-domain signal. S2. Based on the dominant noise type and intensity level, dynamically select or combine the corresponding filtering algorithms from the preset adaptive filter library to filter the original ultrasonic time-domain signal and obtain the filtered signal. S3. Reconstruct and enhance the filtered signal to obtain a noise-reduced signal, thus completing the noise reduction conversion of the deep-hole ultrasonic stress test soil and rock mass.

[0032] In this embodiment, the multidimensional feature vector includes the time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector of the original ultrasonic time-domain signal; the multi-source noise feature database includes typical feature templates of mechanical noise, electromagnetic noise, geological noise, environmental noise, and mixed noise.

[0033] In this embodiment, the noise classification and recognition model employs an improved Dual-Stream Multi-Domain Attention Fusion Network (DS-MDAFNet), specifically designed for noise recognition of ultrasonic signals from deep-hole tunnel drilling. This improved network, based on the traditional multi-scale feature fusion residual network, incorporates targeted structural improvements and functional enhancements by combining the time-domain, frequency-domain, and time-frequency-domain features of the ultrasonic signal. This improved model innovatively introduces a dual-stream structure and cross-domain attention mechanism, better aligning with the multi-domain characteristics of ultrasonic signals from tunnel drilling, thus improving the accuracy and robustness of noise recognition and demonstrating significant structural innovation and technological advancement. During model training, a training set is first constructed using field data labeled with noise types. After preprocessing and multi-domain feature extraction, the data is input into the network. Hierarchical features are extracted through multi-scale convolution and residual connections, and key information is focused using an attention mechanism. Finally, cross-entropy loss and gradient descent optimization are used to achieve a stable high classification accuracy on the validation set. Figure 2As shown, the improved dual-stream multi-domain attention fusion network extracts local detail features and long-range dependency features of the original ultrasonic time-domain signal through one-dimensional depthwise separable convolution and dilated convolution; it extracts frequency domain energy distribution features and harmonic features through short-time Fourier transform and two-dimensional convolutional network; the local detail features, long-range dependency features, frequency domain energy distribution features, and harmonic features are input into the cross-domain attention fusion module for interactive fusion to obtain adaptive fused features; the fused features are input into the adaptive classification output module for classification and enhancement processing to obtain the final dominant noise type and intensity level. The cross-domain attention fusion module interactively fuses local detail features, long-range dependency features, frequency domain energy distribution features, and harmonic features to obtain adaptive fused features. Specifically, it processes local detail features, long-range dependency features, frequency domain energy distribution features, and harmonic features through a lightweight attention mechanism, calculates the importance weight of each feature, and adaptively fuses each feature based on the importance weight to output the adaptive fused features. The adaptive classification output module consists of a global adaptive pooling layer, a fully connected layer, and a Softmax layer connected in sequence. The adaptive fusion features are processed by the global adaptive pooling layer and the fully connected layer to obtain the initial noise type and intensity level. The initial noise type and intensity level are enhanced by the Softmax layer in combination with typical feature templates from the multi-source noise feature database, and the final dominant noise type and intensity level are output.

[0034] In this embodiment, the preset adaptive filter library includes: An improved variational mode decomposition filter adaptively decomposes the original ultrasonic time-domain signal into intrinsic mode functions (IMFs) by optimizing the mode number K and penalty factor α. The adaptive parameter determination strategy is as follows: with the goal of maximizing the overall kurtosis of each IMF component after decomposition, grid search or gradient descent optimization is performed within a preset range to obtain the optimal parameter combination that matches the current signal noise characteristics. Then, the noise-dominant modes are accurately eliminated based on their correlation coefficient and center frequency characteristics. The notch filter adaptive spectral enhancement cascade filter uses an adaptive notch filter to accurately filter out the power frequency and power frequency harmonics in the original ultrasonic time domain signal, and then uses an ALE filter to track and suppress residual narrowband periodic interference. The wavelet packet singular value decomposition (WPSVD) combined noise denoiser utilizes wavelet packet transform to achieve a sparse representation of the original ultrasonic time-domain signal under the optimal basis, and combines singular value decomposition to perform rank estimation and dimensionality reduction on the coefficient matrix, effectively separating the noise subspace and the signal subspace.

[0035] In this embodiment, the method for dynamically selecting or combining corresponding filtering algorithms from a preset adaptive filter library based on the dominant noise type and intensity level is as follows: If the noise is identified as high-intensity mechanical or geological scattering, an improved variational mode decomposition (VMD) filter is selected, and the parameters of the improved variational mode decomposition filter are adaptively adjusted according to the bandwidth of the original ultrasonic time-domain signal. If electromagnetic noise (such as power frequency harmonics) is identified, the Notch Adaptive Line Enhancement (NFALE) cascade filter is invoked, and the notch order and step size of the Notch Adaptive Line Enhancement cascade filter are dynamically optimized. If the noise is identified as broadband environmental noise or instrument background noise, a wavelet packet singular value decomposition joint denoiser is used, and the wavelet packet basis and the number of decomposition layers are adaptively selected. If mixed noise is identified, the improved variational mode decomposition filter, the notch adaptive spectral enhancement cascade filter, and the wavelet packet singular value decomposition joint denoiser are activated in parallel for processing, and the final filtered signal is output through blind source separation and weighted fusion.

[0036] In this embodiment, the method for reconstructing and enhancing the filtered signal to obtain an effective signal is as follows: Signal reconstruction: The filtered signal is reconstructed in reverse to obtain the initial noise-reduced signal; Residual suppression and signal enhancement: Blind source separation (BSS) or independent component analysis (ICA) is performed on the initially denoised signal to further separate potentially coupled residual noise. Subsequently, a sparse reconstruction algorithm based on compressed sensing is used to recover the effective signal from the noisy observation with high fidelity, taking advantage of the sparsity of the ultrasonic signal in a specific transform domain (Gabor dictionary). This includes the following steps: Constructing an observation model: Assuming the initial denoised signal From real and effective ultrasonic signals It is formed by linear mixing with residual noise n, that is, the initial noise reduction signal is ,in This is a reduced-dimensional observation matrix; where, Represents the observation matrix. Represents the real number field. Indicates the observation dimension. Indicates the signal dimension; Assuming a real, valid ultrasonic signal x exists in a certain overcomplete dictionary (Gabor dictionary) has sparse representation, that is , where the sparse coefficient vector There are only a few non-zero elements in it; The signal reconstruction problem is transformed into recovering the sparse coefficients a from the initial denoised signal y, and then solving the following... L1-norm minimization problem:

[0037] in, This represents the regularization parameter used to balance reconstruction error with coefficient sparsity. Indicates to make The sparsity coefficient 'a' corresponding to the minimum value is Represents the squared terms of the L2 norm. Represent the L1 norm term; solve the above using convex optimization algorithms such as iterative soft thresholding. L The problem of minimizing the 1-norm yields estimates of the sparse coefficients. ; The denoised signal calculated based on the sparse coefficient estimate is:

[0038] in, This represents the estimated value of the noise-reduced signal.

[0039] Quality assessment and feedback: Calculate the signal-to-noise ratio (SNR), impulse response sharpness, and other indicators of the denoised signal. If the indicators do not reach the preset threshold, the denoising results and the characteristics of the original ultrasonic time-domain signal are fed back to the noise classification and identification model in step S1 and the filter parameter selection logic in step S2 for iterative optimization.

[0040] Example 2: Based on Example 1, this embodiment of the invention uses an ultrasonic transducer with a center frequency of 50kHz to conduct stress tests in an exploration borehole at a depth of 150 meters in a hydropower station. The original acoustic time-domain signal x(t) is acquired simultaneously, and vibration and electromagnetic environment data within the borehole are recorded using a triaxial accelerometer and magnetometer. Features such as Mel-frequency cepstral coefficients (MFCC), short-time zero-crossing rate, and wavelet packet energy entropy of x(t) are extracted and combined with the dominant frequency characteristics of the vibration data, then input into a noise classification and recognition model trained with a large amount of field data. The model determines that the current dominant noise category is mechanical vibration (60%), geological scattering (30%), and the noise intensity is strong to medium.

[0041] Based on the identification results, the filter selection logic prioritizes the use of the improved VMD filter. With kurtosis maximization as the objective function, K=7 and α=2000 are adaptively determined. After decomposition, the correlation coefficient and center frequency of each IMF component with the original signal are calculated. IMF12 is determined to be mainly high-frequency mechanical vibration and scattering noise, and is therefore removed; IMF37 is an effective signal component and low-frequency noise, and is retained.

[0042] The retained IMF components are reconstructed to obtain y'(t).

[0043] FastICA processing is performed on y'(t) to separate two independent components. The component that is closer to the effective acoustic pulse is manually selected based on the waveform and spectral characteristics.

[0044] The component is then reconstructed using OMP (Orthogonal Matching Pursuit) sparse reconstruction with the Gabor dictionary to obtain the final enhanced signal.

[0045] The assessment showed that the SNR improved from the original 5.2dB to 18.7dB, and the clarity of the first arrival wave apex was improved by 300%.

[0046] Key features (such as VMD parameters and filtered SNR) from this noise reduction process are recorded and updated to the feature database and classification model to optimize processing performance under similar conditions in the future.

[0047] Example 3: Building upon Example 1, this embodiment of the invention, in a deep-hole test in a mine with strong interference from the construction site's power grid, identifies electromagnetic interference (50Hz power frequency and its 150Hz harmonics) as the dominant noise using a noise classification model. The system automatically invokes an NFALE cascaded filter. First, an adaptive notch filter based on the LMS algorithm precisely filters out the 50Hz and 150Hz components; subsequently, an ALE filter, using a delay Δ=1 / 100Hz as a reference, further suppresses residual narrowband interference. The post-processing stage primarily employs wavelet soft thresholding denoising to handle broadband background noise. Ultimately, the signal waveform is smoothed, and harmonic interference is completely eliminated, ensuring accurate wave velocity calculation in the subsequent process.

[0048] Example 4: Based on Example 1, this embodiment of the invention provides a noise reduction and conversion system for deep-hole ultrasonic stress testing of soil and rock masses, which can be used to implement the noise reduction and conversion method for deep-hole ultrasonic stress testing of soil and rock masses as described in the previous embodiments. Figure 3 As shown, the system includes: The noise source identification and classification module is used to collect the original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment, and extract the multi-dimensional feature vector of the original ultrasonic time-domain signal. Based on the multi-dimensional feature vector and auxiliary environmental data, it uses a pre-trained classification model and a multi-source noise feature database to identify and output the dominant noise type and intensity level in the current original ultrasonic time-domain signal. The acoustic filtering algorithm module is used to dynamically select or combine corresponding filtering algorithms from a preset adaptive filter library according to the dominant noise type and intensity level, to filter the original ultrasonic time-domain signal and obtain the filtered signal. The post-processing module is used to reconstruct and enhance the filtered signal to obtain a noise-reduced signal, thus completing the noise reduction conversion of the deep-hole ultrasonic stress test soil and rock mass.

[0049] The system also includes a central processing and control unit, which coordinates the work of the noise source identification and classification module, the acoustic filtering algorithm module, and the post-processing module, and executes an iterative optimization process.

[0050] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0051] In an exemplary embodiment, the electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the deep-hole ultrasonic stress testing noise reduction conversion method for soil and rock masses as described in Embodiment 1 above.

[0052] In an exemplary embodiment, the readable storage medium may be a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the deep-hole ultrasonic stress testing noise reduction and conversion method for soil and rock masses according to Embodiment 1 above.

[0053] In an exemplary embodiment, the computer program product includes a computer program that, when executed by a processor, implements the deep-hole ultrasonic stress testing noise reduction and conversion method for soil and rock masses as described in Embodiment 1 above.

[0054] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0055] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0056] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0057] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0058] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0059] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses, characterized in that, Includes the following steps: The original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment are collected, and the multi-dimensional feature vector of the original ultrasonic time-domain signal is extracted. Based on the multi-dimensional feature vector and auxiliary environmental data, the dominant noise type and intensity level in the current original ultrasonic time-domain signal are identified and output using a noise classification and identification model and a multi-source noise feature database. Based on the dominant noise type and intensity level, the corresponding filtering algorithm is dynamically selected or combined from the preset adaptive filter library to filter the original ultrasonic time domain signal and obtain the filtered signal. The filtered signal is reconstructed and enhanced to obtain a noise-reduced signal, thus completing the noise reduction conversion of the rock and soil body for deep hole ultrasonic stress testing.

2. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 1, characterized in that, The multi-source noise feature database includes typical feature templates for mechanical noise, electromagnetic noise, geological noise, environmental noise, and mixed noise.

3. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 1, characterized in that, The noise classification and recognition model employs an improved dual-stream multi-domain attention fusion network; the improved dual-stream multi-domain attention fusion network includes a dual-stream feature extraction path module, a cross-domain attention fusion module, and an adaptive classification output module connected in sequence; The dual-stream feature extraction path module is used to extract local detail features and long-range dependency features of the original ultrasonic time-domain signal through one-dimensional depthwise separable convolution and dilated convolution; and to extract frequency domain energy distribution features and harmonic features through short-time Fourier transform and two-dimensional convolutional network. The cross-domain attention fusion module is used to receive local detail features, long-range dependency features, frequency domain energy distribution features, and harmonic features, calculate the importance weight of each feature, and adaptively fuse the features based on the importance weight to output the adaptive fused features. The adaptive classification output module includes a global adaptive pooling layer, a fully connected layer, and a Softmax layer connected in sequence. It is used to process the adaptive fusion features to obtain the initial noise type and intensity level. Then, it combines the typical feature templates in the multi-source noise feature database to enhance the initial noise type and intensity level, and outputs the final dominant noise type and intensity level.

4. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 1, characterized in that, The pre-defined adaptive filter library includes: An improved variational mode decomposition filter adaptively decomposes the original ultrasonic time-domain signal into intrinsic mode functions by optimizing the mode number K and penalty factor α of variational mode decomposition. The notch filter adaptive spectral enhancement cascade filter uses an adaptive notch filter to accurately filter out the power frequency and power frequency harmonics in the original ultrasonic time domain signal, and then uses an ALE filter to track and suppress residual narrowband periodic interference. The wavelet packet singular value decomposition combined noise denoiser utilizes wavelet packet transform to achieve a sparse representation of the original ultrasonic time-domain signal under the optimal basis, and combines singular value decomposition to perform rank estimation and dimensionality reduction on the coefficient matrix to separate the noise subspace and the signal subspace.

5. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 4, characterized in that, The method for dynamically selecting or combining corresponding filtering algorithms from a preset adaptive filter library based on the dominant noise type and intensity level is as follows: If the noise is identified as mechanical or geological scattering, an improved variational mode decomposition filter is selected, and the parameters of the improved variational mode decomposition filter are adaptively adjusted according to the bandwidth of the original ultrasonic time-domain signal. If electromagnetic noise is identified, the notch adaptive spectral enhancement cascade filter is invoked, and the notch order and step size of the notch adaptive spectral enhancement cascade filter are dynamically optimized. If the noise is identified as environmental noise or instrument background noise, a wavelet packet singular value decomposition joint denoiser is used, and the wavelet packet basis and the number of decomposition layers are adaptively selected. If mixed noise is identified, the improved variational mode decomposition filter, the notch adaptive spectral enhancement cascade filter, and the wavelet packet singular value decomposition joint denoiser are activated in parallel for processing, and the final filtered signal is output through blind source separation and weighted fusion.

6. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 1, characterized in that, The specific method for reconstructing and enhancing the filtered signal to obtain the effective signal is as follows: The filtered signal is reconstructed in reverse to obtain the initial denoised signal; When the initial noise reduction signal From real and effective ultrasonic signals It is formed by linear mixing with residual noise n, that is, the initial noise reduction signal is ,in For the reduced-dimensional observation matrix, Represents the observation matrix. Represents the real number field. Indicates the observation dimension. Indicates the signal dimension; And the real and effective ultrasonic signal x has a certain complete dictionary The following has a sparse representation, that is , where the sparse coefficient vector There are only a few non-zero elements in it; The signal reconstruction problem is transformed into recovering the sparse coefficients a from the initial denoised signal y, and then solving... L The problem of minimizing the 1-norm yields estimates of the sparse coefficients. The denoised signal calculated based on the sparse coefficient estimate is: in, This represents the estimated value of the denoised signal. This represents the estimated value of the sparse coefficients.

7. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 6, characterized in that, L The problem of minimizing the 1-norm is specifically as follows: in, This represents the regularization parameter used to balance reconstruction error with coefficient sparsity. Indicates to make The sparsity coefficient 'a' corresponding to the minimum value is Represents the squared terms of the L2 norm. This represents the L1 norm term.

8. The method for noise reduction and conversion of deep-hole ultrasonic stress testing in soil and rock masses according to claim 1, characterized in that, The method further includes: The quality index of the denoised signal is calculated. If the quality index does not reach the preset threshold, the multi-dimensional feature vectors of the denoised signal and the original ultrasonic time domain signal are fed back to the noise classification and identification model and the filtering algorithm for iterative optimization.

9. A noise reduction and conversion system for deep-hole ultrasonic stress testing of soil and rock masses, characterized in that, include: The noise source identification and classification module is used to collect the original ultrasonic time-domain signal and auxiliary environmental data in the deep hole environment, and extract the multi-dimensional feature vector of the original ultrasonic time-domain signal. Based on the multi-dimensional feature vector and auxiliary environmental data, the module uses the noise classification and identification model and the multi-source noise feature database to identify and output the dominant noise type and intensity level in the current original ultrasonic time-domain signal. The acoustic filtering algorithm module is used to dynamically select or combine corresponding filtering algorithms from a preset adaptive filter library according to the dominant noise type and intensity level, to filter the original ultrasonic time-domain signal and obtain the filtered signal. The post-processing module is used to reconstruct and enhance the filtered signal to obtain a noise-reduced signal, thus completing the noise reduction conversion of the deep-hole ultrasonic stress test soil and rock mass.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8.

Images