Micro-seismic event classification method based on deep learning

Abstract

The invention provides a micro-seismic event classification method based on deep learning. The method comprises the following steps: collecting a plurality of historical micro-seismic signals and correspondingly marked type labels of micro-seismic events; filtering each historical microseismic signal, and introducing empirical mode decomposition to obtain a microseismic time domain signal; converting the microseism time domain signal into an equivalent two-dimensional spectrogram; extracting feature vectors Vc representing low-frequency, high-frequency and edge textures corresponding to the two-dimensional spectrogram through a deep learning framework; extracting an image feature descriptor Vh representing the shape and texture information of the two-dimensional spectrogram by adopting a histogram of oriented gradients; performing splicing or weighted fusion on the Vc and the Vh to obtain a high-dimensional feature vector and a category label to train a classifier, and obtaining a classification model of the microseism event; and classifying and discriminating the to-be-processed microseismic signals through the classification model. Therefore, the classification accuracy and robustness can be improved by adopting the classification model constructed by the image features, and real-time micro-seismic event classification is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of microseismic monitoring and identification, and particularly to a method and device for classifying microseismic events based on deep learning. Background Art

[0002] As underground mine engineering continues to develop deeper, dynamic disasters such as rock bursts and rock bumps often occur. Microseismic monitoring technology can provide important early warning information for engineering safety. However, during the process of collecting microseismic signals, it is often affected by various mixed noises such as environmental noise and electromagnetic interference, which increases the difficulty of classifying microseismic events. In related technologies, traditional methods mostly rely on manual experience or only classify based on seismic source parameters, and it is difficult to take into account the interference brought by complex noises. The classification efficiency and accuracy are difficult to meet the monitoring requirements. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems in the related technologies to some extent.

[0004] To this end, the first object of the present invention is to propose a method for classifying microseismic events based on deep learning. The classification model constructed by using image features can improve the classification accuracy and robustness and realize real-time classification of microseismic events.

[0005] The second object of the present invention is to propose a device for classifying microseismic events based on deep learning.

[0006] The third object of the present invention is to propose an electronic device.

[0007] The fourth object of the present invention is to propose a non-transitory computer-readable storage medium storing computer instructions.

[0008] To achieve the above object, the first aspect embodiment of the present invention proposes a method for classifying microseismic events based on deep learning, and the method includes:

[0009] Collect a plurality of historical microseismic signals, wherein each historical microseismic signal is labeled with a category label corresponding to the microseismic event;

[0010] Perform filtering processing on each of the historical microseismic signals in a preset noise frequency band by means of band-pass filtering to obtain a denoised signal;

[0011] Decompose and remove abnormal noise components in the denoised signal by introducing the empirical mode decomposition method to obtain a microseismic time-domain signal;

[0012] Convert the microseismic time-domain signal into an equivalent two-dimensional frequency spectrum diagram by means of the short-time Fourier transform method;

[0013] Input the two-dimensional frequency spectrum diagram into a deep learning framework mainly based on a convolutional neural network for feature extraction to obtain a feature vector V representing the low frequency, high frequency, and edge texture corresponding to the two-dimensional frequency spectrum diagram. c ;

[0014] Perform image feature extraction on the two-dimensional frequency spectrum diagram using the histogram of oriented gradients to obtain an image feature descriptor V characterizing the shape and texture information of the two-dimensional frequency spectrum diagram. h ;

[0015] Concatenate or perform weighted fusion on the V c and V h to obtain a high-dimensional feature vector V = [V c ||V h as the input of the classifier. Each historical microseismic signal is labeled with the type label of the corresponding microseismic event as the output of the classifier, and a classification model for microseismic events is trained.

[0016] Classify and discriminate the high-dimensional feature vector corresponding to the microseismic signal to be processed through the classification model to obtain the target microseismic event corresponding to the microseismic signal to be processed.

[0017] To achieve the above object, an embodiment of the second aspect of the present invention proposes a microseismic event classification device based on deep learning. The device includes:

[0018] An acquisition module for acquiring a plurality of historical microseismic signals, where each historical microseismic signal is labeled with the type label of the corresponding microseismic event.

[0019] A filtering module for filtering each of the historical microseismic signals in a preset noise frequency band through a band-pass filtering method to obtain a denoised signal.

[0020] A decomposition module for decomposing and removing abnormal noise components in the denoised signal by the method of introducing empirical mode decomposition to obtain a microseismic time-domain signal.

[0021] A conversion module for converting the microseismic time-domain signal into an equivalent two-dimensional frequency spectrum diagram by the short-time Fourier transform method.

[0022] A first extraction module for inputting the two-dimensional frequency spectrum diagram into a deep learning framework mainly based on a convolutional neural network for feature extraction to obtain a feature vector V representing the low frequency, high frequency, and edge texture corresponding to the two-dimensional frequency spectrum diagram. c ;

[0023] A second extraction module for performing image feature extraction on the two-dimensional frequency spectrum diagram using the histogram of oriented gradients to obtain an image feature descriptor V characterizing the shape and texture information of the two-dimensional frequency spectrum diagram. h ;

[0024] A training module for combining the V c with V h to splice or weighted fuse to obtain a high-dimensional feature vector V = [V c ||V h as the input of the classifier, and each historical microseismic signal is labeled with the type label of the corresponding microseismic event as the output of the classifier, and a classification model of microseismic events is trained;

[0025] A classification module for classifying and discriminating the high-dimensional feature vector corresponding to the microseismic signal to be processed through the classification model to obtain the target microseismic event corresponding to the microseismic signal to be processed.

[0026] To achieve the above object, an embodiment of the third aspect of the present invention provides an electronic device, including: 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, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in the first aspect.

[0027] To achieve the above object, an embodiment of the fourth aspect of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the method described in the first aspect.

[0028] In the method, device, electronic device and storage medium for classifying microseismic events based on deep learning according to the embodiments of the present invention, a plurality of historical microseismic signals and the type labels of the corresponding microseismic events are collected; after filtering each historical microseismic signal, empirical mode decomposition is introduced to obtain a microseismic time-domain signal; the microseismic time-domain signal is converted into an equivalent two-dimensional spectrogram; a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrogram is extracted through a deep learning framework c ; the image feature descriptor V characterizing the shape and texture information of the two-dimensional spectrogram is extracted by using the histogram of oriented gradients h ; the high-dimensional feature vector obtained by splicing or weighted fusing V c with V h and the type label are used to train the classifier to obtain a classification model of microseismic events; the microseismic signal to be processed is classified and discriminated through the classification model. Thus, the classification model constructed by using the image features can improve the classification accuracy and robustness and realize real-time classification of microseismic events.

[0029] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of embodiments in conjunction with the accompanying drawings, wherein:

[0031] Figure 1 It is a schematic flowchart of a microseismic event classification method based on deep learning provided by an embodiment of the present invention;

[0032] Figure 2 It is a technical roadmap of a microseismic event classification method based on deep learning provided by an embodiment of the present invention;

[0033] Figure 3 It is a schematic structural diagram of a microseismic event classification device based on deep learning provided by an embodiment of the present invention. Detailed Embodiments

[0034] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0035] It should be noted that in the technical solution of the present invention, the acquisition, storage, use, processing, etc. of data all comply with the relevant regulations of relevant laws and regulations.

[0036] The microseismic event classification method and device based on deep learning according to the embodiments of the present invention will be described below with reference to the accompanying drawings.

[0037] Figure 1 It is a schematic flowchart of a microseismic event classification method based on deep learning provided by an embodiment of the present invention.

[0038] As Figure 1 shown, the method includes the following steps:

[0039] Step 101, collect a plurality of historical microseismic signals, wherein each historical microseismic signal is labeled with a category label corresponding to a microseismic event.

[0040] In some embodiments, the historical microseismic signals may be a plurality of triaxial velocity sensors arranged on the working face of any mining area, with a sampling frequency of f s (2 ms), and continuously collect microseismic signal data with a total length of T (two months), but not limited thereto.

[0041] Among them, the category labels of microseismic events include blasting event (B) label, rock fracture event (R) label, electromagnetic event (E) label, and pure noise event (N) label. Each historical microseismic signal is labeled with the corresponding category label of the microseismic event, which can be marked manually or semi-automatically and is used to construct the dataset (training set and test set) of historical microseismic signals.

[0042] It can be understood that microseismic events are small seismic activities generated in underground engineering; blasting events are phenomena in which rocks suddenly rupture and release energy.

[0043] Step 102: Perform filtering processing on each historical microseismic signal in a preset noise frequency band through a band-pass filtering method to obtain a denoised signal.

[0044] In some embodiments, for problems such as environmental noise and electromagnetic interference that may exist in historical microseismic signals, a band-pass filtering method is first used to preliminarily reduce the preset noise frequency band (ultra-low frequency or ultra-high frequency noise), and the preset noise frequency band interval is selected according to the actual noise frequency band in the mining area.

[0045] Among them, band-pass filtering is a filtering method that only allows signals within a specific frequency range to pass through.

[0046] Step 103: Decompose and remove the abnormal noise components in the denoised signal by introducing the method of empirical mode decomposition to obtain the microseismic time-domain signal.

[0047] Among them, empirical mode decomposition (EMD) is used to decompose and remove the most obvious abnormal noise components to obtain the microseismic time-domain signal x(t) to weaken the influence of non-stationary noise.

[0048] Step 104: Convert the microseismic time-domain signal into an equivalent two-dimensional spectrogram through the short-time Fourier transform method.

[0049] In some embodiments, the short-time Fourier transform (STFT) can be expressed as formula (1):

[0050]

[0051] Among them, ω(t - τ) is a window function with an appropriate length (such as a Hanning window) used to intercept the local time-domain information of the microseismic time-domain signal x(t); τ is the time center position, that is, the center of the sliding window on the time axis; ω is the angular frequency, ω = 2πf, f is the frequency; j represents the imaginary unit.

[0052] By traversing different τ positions, a series of spectral segments are obtained and spliced on the time axis. Finally, a two-dimensional spectrogram can be obtained, which can be written as S(τ, f), where τ is the time step and f is the frequency resolution.

[0053] In addition, when analyzing microseismic signals containing mixed noise, different parameters of STFT need to be used for different frequency band information to enhance the recognition ability of weak signals or high-noise segments.

[0054] Optionally, the two-dimensional spectrogram is the energy distribution diagram of the signal in the time-frequency domain.

[0055] Step 105: Input the two-dimensional spectrogram into a deep learning framework with a convolutional neural network as the main body for feature extraction to obtain a feature vector V representing the low-frequency, high-frequency, and edge texture corresponding to the two-dimensional spectrogram. c 。

[0056] In some embodiments, the deep learning framework (Deep) with a convolutional neural network (Convolutional Neural Network, CNN) as the main body includes a convolutional layer 1, a convolutional layer 2, a pooling layer, a fully connected layer, and an output layer. Input the two-dimensional spectrogram into the deep learning framework with a convolutional neural network as the main body for feature extraction to obtain a feature vector V representing the low-frequency, high-frequency, and edge texture corresponding to the two-dimensional spectrogram. c One implementation can be to input the two-dimensional spectrogram into the convolutional layer 1 of the deep learning framework with a convolutional neural network as the main body to extract the global low-frequency features of the two-dimensional spectrogram; extract the local high-frequency features and edge texture features of the two-dimensional spectrogram through the convolutional layer 1; perform downsampling and feature compression on the low-frequency features, high-frequency features, and edge texture features through the pooling layer to obtain the representative features of the two-dimensional spectrogram; map the representative features to a high-dimensional vector space through the fully connected layer to obtain a feature vector V representing the low-frequency, high-frequency, and edge texture corresponding to the two-dimensional spectrogram. c ; V c It is used to reflect the high-level abstract representation of the two-dimensional spectrogram by the deep learning framework. Among them, the size of the convolutional kernels of the convolutional layer 1 and the convolutional layer 2 and the pooling operation step size of the pooling layer are determined by the frequency band intervals corresponding to the low-frequency features and the high-frequency features. It can still perform detailed recognition on weak signals containing mixed noise.

[0057] Step 106: Use the histogram of oriented gradients to extract the image features of the two-dimensional spectrogram to obtain an image feature descriptor V characterizing the shape and texture information of the two-dimensional spectrogram. h 。

[0058] In some embodiments, Histogram of Oriented Gradients (HOG) is used to extract image features from the two-dimensional spectrogram, so as to obtain an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrogram. h One implementation can be as follows: the two-dimensional spectrogram is divided into multiple non-overlapping cells by using the histogram of oriented gradients, and the image gradients in the horizontal and vertical directions within each non-overlapping cell, namely the horizontal gradient and the vertical gradient, are extracted; based on the horizontal gradient and the vertical gradient corresponding to each pixel point in the two-dimensional spectrogram, the local gradient intensity and the gradient direction of each pixel point are calculated; the gradient directions within each non-overlapping cell are statistically analyzed or quantified, divided into multiple direction intervals, and then a direction histogram is constructed based on the local gradient intensities within the multiple direction intervals (usually divided into 9 to 10 direction intervals), forming an h-dimensional local vector with a length of H; the h-dimensional local vectors of adjacent non-overlapping cells are combined into blocks and normalized to obtain an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrogram. h Thus, through the operation of the histogram of oriented gradients, the structured information of the useful waveform can be enhanced while suppressing noise interference.

[0059] Specifically, the calculation formulas for the local gradient intensity G and the gradient direction θ of each pixel point (x, y) are as shown in Equations (2) and (3):

[0060]

[0061] where I(x, y) represents the intensity value of the two-dimensional spectrogram at the pixel point (x, y), is the horizontal gradient, is the vertical gradient.

[0062] Step 107, the high-dimensional feature vector V = [V c ||V h obtained by splicing or weighted fusion of V c ||V h is used as the input of the classifier, and each historical microseismic signal is labeled with the type label of the corresponding microseismic event as the output of the classifier to train a classification model for microseismic events.

[0063] In some embodiments, the classifier is constructed by using a support vector machine or a fully connected layer plus a softmax function. The kernel function of the support vector machine (Support Vector Machine, SVM) uses a radial basis function (Radial Basis Function, RBF), and the grid search method is used to determine the kernel function parameters.

[0064] Specifically, V cConcatenating or weighted fusion with V h is a vector-level fusion method (weighted) to obtain a high-dimensional vector containing artificial prior and depth features, i.e., formula (4).

[0065] f HOOG-Deep = α·f CNN = β·f HOOG (4)

[0066] where α and β are adjustable weights.

[0067] Optionally, considering the cases of mixed noise and small samples, the high-dimensional feature V = [V c ||V h extracted by the HOOG-Deep network is input into the traditional SVM classifier, and the radial basis function RBF is as shown in formula (5)

[0068] K(X i , X j ) = exp(-γ||X i - X j || 2 ) (5)

[0069] where X i , X j represent the feature vectors of two training samples (V c and V h ) respectively; γ represents the width parameter of the kernel function, and the larger the γ value, the more complex the decision boundary;

[0070] In the dual problem of SVM, the penalty coefficient C is used to balance the trade-off between maximizing the margin and the classification error rate. Thus, when determining the kernel function parameters (C, γ) by the grid search method, cross-validation is performed on the training set and the validation set to find the optimal parameter combination (C, γ) and hyperparameters such as the convolutional kernel size, learning rate, and number of iterations in the CNN network.

[0071] where the kernel function is the function used for feature space mapping in SVM.

[0072] Step 108, classify and discriminate the high-dimensional feature vector corresponding to the to-be-processed microseismic signal through the classification model to obtain the target microseismic event corresponding to the to-be-processed microseismic signal.

[0073] In some embodiments, by selecting a two-dimensional spectrogram and fusing HOOG features or other image feature extraction methods, effective spectral feature information of microseismic events can still be obtained under mixed complex noise, enabling the accurate construction of the classification model, which has strong generality and can be used in various forms of underground engineering, mines, and other high-noise environments that require real-time monitoring.

[0074] The micro-seismic event classification method based on deep learning according to the embodiments of the present invention collects a plurality of historical micro-seismic signals and the category labels of the corresponding labeled micro-seismic events; after filtering each historical micro-seismic signal, empirical mode decomposition is introduced to obtain the micro-seismic time-domain signal; the micro-seismic time-domain signal is converted into an equivalent two-dimensional spectrogram; the feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrogram is extracted through a deep learning framework c ; the image feature descriptor V representing the shape and texture information of the two-dimensional spectrogram is extracted by using the histogram of oriented gradients h ; V c and V h are spliced or weighted and fused to obtain a high-dimensional feature vector and the category label to train the classifier, and a classification model of micro-seismic events is obtained; the micro-seismic signal to be processed is classified and discriminated through the classification model. Thus, the classification model constructed by using image features can improve the classification accuracy and robustness, and realize real-time micro-seismic event classification

[0075] In addition, for better understanding of the present invention, the present invention also proposes a technical roadmap of another micro-seismic event classification method based on deep learning. Specifically, the technical roadmap includes collecting raw data (collecting a plurality of historical micro-seismic signals and the category labels of the corresponding labeled micro-seismic events), band-pass filtering (filtering each historical micro-seismic signal in a preset noise frequency band by a band-pass filtering method to obtain a denoised signal), empirical mode decomposition (introducing the method of empirical mode decomposition to decompose and remove the abnormal noise components in the denoised signal to obtain the micro-seismic time-domain signal), spectral analysis (converting the micro-seismic time-domain signal into an equivalent two-dimensional time-frequency spectrogram), extracting HOOG features (extracting image features of the two-dimensional time-frequency spectrogram by using the histogram of oriented gradients to obtain the image feature descriptor V h ), inputting into CNN (inputting the two-dimensional time-frequency spectrogram into a deep learning framework mainly composed of a convolutional neural network for feature extraction to obtain the feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional time-frequency spectrogram c ), weighted splicing (splicing or weighted fusing V c and V h to obtain a high-dimensional feature vector V = [V c ||V h ), inputting into SVM (inputting the high-dimensional feature vector V = [V c ||V hAs the input of the classifier, each historical microseismic signal is labeled with the category label of the corresponding microseismic event. As the output of the classifier, a classification model of microseismic events is trained and the final classification is completed (by classifying and discriminating the high-dimensional feature vector corresponding to the microseismic signal to be processed through the classification model to obtain the target microseismic event corresponding to the microseismic signal to be processed). Thus, by fusing the convolutional neural network (CNN) with machine vision features, it is used for the automatic classification of microseismic events. This method can be applied to high-noise scenarios such as mine exploitation to provide efficient early warning and safety guarantee for dynamic disasters such as rock bursts and rock bumps.

[0076] To implement the above embodiments, the present invention also proposes a microseismic event classification device based on deep learning.

[0077] Figure 3 It is a schematic structural diagram of a microseismic event classification device based on deep learning provided by an embodiment of the present invention.

[0078] As Figure 3 shown, the microseismic event classification device 30 based on deep learning includes: a collection module 31, a filtering module 32, a decomposition module 33, a transformation module 34, a first extraction module 35, a second extraction module 36, a training module 37, and a classification module 38.

[0079] The collection module 31 is used to collect a plurality of historical microseismic signals, wherein each historical microseismic signal is labeled with the category label of the corresponding microseismic event;

[0080] The filtering module 32 is used to perform filtering processing on each of the historical microseismic signals in a preset noise frequency band by means of band-pass filtering to obtain a denoised signal;

[0081] The decomposition module 33 is used to decompose and remove the abnormal noise components in the denoised signal by introducing the empirical mode decomposition method to obtain a microseismic time-domain signal;

[0082] The transformation module 34 is used to transform the microseismic time-domain signal into an equivalent two-dimensional spectrogram by means of the short-time Fourier transform method;

[0083] The first extraction module 35 is used to input the two-dimensional spectrogram into a deep learning framework with a convolutional neural network as the main body for feature extraction to obtain a feature vector V representing the low frequency, high frequency, and edge texture corresponding to the two-dimensional spectrogram c ;

[0084] The second extraction module 36 is used to perform image feature extraction on the two-dimensional spectrogram by using the histogram of oriented gradients to obtain an image feature descriptor V representing the shape and texture information of the two-dimensional spectrogram h ;

[0085] The training module 37 is used to combine the V c with V h to splice or weighted fuse the obtained high-dimensional feature vector V = [V c ||V h as the input of the classifier, and each historical microseismic signal is labeled with the type label of the corresponding microseismic event as the output of the classifier, and a classification model of microseismic events is trained;

[0086] The classification module 38 is used to classify and discriminate the high-dimensional feature vector corresponding to the microseismic signal to be processed through the classification model, so as to obtain the target microseismic event corresponding to the microseismic signal to be processed.

[0087] Further, in a possible implementation manner of the embodiment of the present invention, the type labels of the microseismic events include blasting event labels, rock fracture event labels, electromagnetic event labels, and pure noise event labels.

[0088] Further, in a possible implementation manner of the embodiment of the present invention, the deep learning framework with a convolutional neural network as the main body includes a convolutional layer 1, a convolutional layer 2, a pooling layer, a fully connected layer, and an output layer. The first extraction module 35 is specifically used for:

[0089] Input the two-dimensional spectrogram into the convolutional layer 1 of the deep learning framework with a convolutional neural network as the main body to extract the low-frequency features of the global two-dimensional spectrogram;

[0090] Extract the high-frequency features and edge texture features of the local two-dimensional spectrogram through the convolutional layer 1;

[0091] Perform downsampling and feature compression on the low-frequency features, high-frequency features, and edge texture features through the pooling layer to obtain the representation features of the two-dimensional spectrogram;

[0092] Map the representation features to a high-dimensional vector space through the fully connected layer to obtain the feature vector V c ;

[0093] Among them, the size of the convolution kernels of the convolutional layer 1 and the convolutional layer 2 and the pooling operation step of the pooling layer are determined by the frequency band intervals corresponding to the low-frequency features and the high-frequency features.

[0094] Further, in a possible implementation manner of the embodiment of the present invention, the second extraction module 36 is specifically used for:

[0095] Use the histogram of oriented gradients to divide the two-dimensional spectrogram into multiple non-overlapping cells, and extract the image gradients in the horizontal and vertical directions within each non-overlapping cell, that is, the horizontal gradient and the vertical gradient;

[0096] Based on the horizontal gradient and vertical gradient corresponding to each pixel point in the two-dimensional spectrogram, calculate the local gradient intensity and gradient direction of each pixel point;

[0097] Statistically analyze or quantify the gradient directions within each non-overlapping unit, divide them into multiple direction intervals, and then construct a direction histogram based on the local gradient intensities within the multiple direction intervals to form an h-dimensional local vector of length H;

[0098] Combine the h-dimensional local vectors of adjacent non-overlapping units into blocks and perform normalization processing to obtain an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrogram h 。

[0099] Furthermore, in a possible implementation manner of the embodiment of the present invention, the classifier is constructed by using a support vector machine or a fully connected layer plus a softmax function. The kernel function of the support vector machine uses a radial basis function, and a grid search method is used to determine the kernel function parameters.

[0100] The microseismic event classification device based on deep learning in the embodiment of the present invention collects multiple historical microseismic signals and the type labels of the corresponding labeled microseismic events; after filtering each historical microseismic signal, introduce empirical mode decomposition to obtain the microseismic time-domain signal; convert the microseismic time-domain signal into an equivalent two-dimensional spectrogram; extract a feature vector V representing the low frequency, high frequency, and edge texture corresponding to the two-dimensional spectrogram through a deep learning framework c ; use the histogram of oriented gradients to extract an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrogram h ; use V c and V h to splice or weighted fuse the obtained high-dimensional feature vector and the type label to train the classifier to obtain a microseismic event classification model; classify and discriminate the microseismic signal to be processed through the classification model. Thus, the classification model constructed by using image features can improve the classification accuracy and robustness and realize real-time microseismic event classification.

[0101] To implement the above embodiments, the present invention also proposes an electronic device, including:

[0102] At least one processor; and

[0103] A memory communicatively connected to the at least one processor; wherein,

[0104] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the foregoing method.

[0105] To implement the above embodiments, the present invention further provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the foregoing method.

[0106] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0107] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of these features. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0108] Any process or method description in a flowchart or described in other ways herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions may be executed in a manner not shown or discussed, including substantially simultaneously according to the involved functions or in a reverse order, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0109] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definable sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0110] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0111] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0112] In addition, each functional unit in various embodiments of the present invention may be integrated into one processing module, may exist separately as individual physical units, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0113] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A microseismic event classification method based on deep learning, characterized in that: The method comprises: Collecting multiple historical microseismic signals, wherein each historical microseismic signal is labeled with a type label of a corresponding microseismic event; Performing filtering processing on each of the historical microseismic signals in a preset noise frequency band by means of bandpass filtering to obtain a denoised signal; Decomposing and removing abnormal noise components in the denoised signal by the method of introducing empirical mode decomposition to obtain a microseismic time domain signal; The microseismic time domain signal is converted into an equivalent two-dimensional spectrum diagram by short-time Fourier transform method; The two-dimensional spectrum graph is input into a deep learning framework based on a convolutional neural network for feature extraction to obtain a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrum graph. c ; The image feature extraction of the two-dimensional spectrum graph is performed using the directional gradient histogram to obtain an image feature descriptor V representing the shape and texture information of the two-dimensional spectrum graph. h ; The V c With V h The high-dimensional feature vector V = [V c ||V h ] as the input of the classifier, each historical microseismic signal is annotated with the type label of the corresponding microseismic event as the output of the classifier, and a classification model for microseismic events is trained; The high-dimensional feature vector corresponding to the microseismic signal to be processed is classified and judged through the classification model to obtain the target microseismic event corresponding to the microseismic signal to be processed.

2. The microseismic event classification method based on deep learning according to claim 1, characterized in that: in, The type labels of the microseismic events include blasting event labels, rock fracture event labels, electromagnetic event labels, and pure noise event labels.

3. The microseismic event classification method based on deep learning according to claim 1, characterized in that: The deep learning framework with the convolutional neural network as the main body includes a convolutional layer 1, a convolutional layer 2, a pooling layer, a fully connected layer and an output layer. The two-dimensional spectrum graph is input into the deep learning framework with the convolutional neural network as the main body for feature extraction to obtain a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrum graph. c ,include: Inputting the two-dimensional spectrum graph into the convolution layer 1 of the deep learning framework based on the convolutional neural network to extract the global low-frequency features of the two-dimensional spectrum graph; The convolution layer 1 is used to extract the local high-frequency features and edge texture features of the two-dimensional spectrum graph; Downsampling and feature compression are performed on the low-frequency features, high-frequency features, and edge texture features through a pooling layer to obtain representation features of a two-dimensional spectrum graph; The representation feature is mapped to a high-dimensional vector space through a fully connected layer to obtain a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrum graph c ; Among them, the size of the convolution kernel of convolution layer 1 and convolution layer 2 and the step size of the pooling operation of the pooling layer are determined by the frequency band interval corresponding to the low-frequency features and the high-frequency features.

4. The microseismic event classification method based on deep learning according to claim 1, characterized in that: The image feature extraction of the two-dimensional spectrum graph using the directional gradient histogram is performed to obtain an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrum graph. h ,include: The two-dimensional spectrum graph is divided into a plurality of non-overlapping units by using a directional gradient histogram, and image gradients in the horizontal direction and the vertical direction, i.e., the horizontal gradient and the vertical gradient, are extracted in each non-overlapping unit; Based on the horizontal gradient and vertical gradient corresponding to each pixel in the two-dimensional spectrum graph, the local gradient intensity and gradient direction of each pixel are calculated; The gradient direction in each non-overlapping unit is counted or quantized and divided into multiple direction intervals, and then a direction histogram is constructed according to the local gradient strength in the multiple direction intervals to form an h-dimensional local vector with a length of H; The h-dimensional local vectors of each adjacent non-overlapping unit are combined into a block and normalized to obtain the image feature descriptor V that represents the shape and texture information of the two-dimensional spectrum graph. h .

5. The microseismic event classification method based on deep learning according to claim 1, characterized in that: The classifier is constructed by using a support vector machine or a fully connected layer plus a normalized exponential function. The kernel function of the support vector machine uses a radial basis function, and a grid search method is used to determine the kernel function parameters.

6. A microseismic event classification device based on deep learning, characterized in that: The device comprises: An acquisition module, used for acquiring a plurality of historical microseismic signals, wherein each historical microseismic signal is annotated with a type label of a corresponding microseismic event; A filtering module, used to filter each of the historical microseismic signals in a preset noise frequency band by bandpass filtering to obtain a denoised signal; A decomposition module, used for decomposing and removing abnormal noise components in the denoised signal by introducing the empirical mode decomposition method to obtain a microseismic time domain signal; A conversion module, used to convert the microseismic time domain signal into an equivalent two-dimensional spectrum diagram by using a short-time Fourier transform method; The first extraction module is used to input the two-dimensional spectrum graph into a deep learning framework based on a convolutional neural network to perform feature extraction, so as to obtain a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrum graph. c ; The second extraction module is used to extract image features from the two-dimensional spectrum using a directional gradient histogram to obtain an image feature descriptor V that characterizes the shape and texture information of the two-dimensional spectrum. h ; A training module is used to convert the V c With V h The high-dimensional feature vector V = [V c ||V h ] as the input of the classifier, each historical microseismic signal is annotated with the type label of the corresponding microseismic event as the output of the classifier, and a classification model for microseismic events is trained; The classification module is used to classify and distinguish the high-dimensional feature vectors corresponding to the microseismic signal to be processed through the classification model to obtain the target microseismic event corresponding to the microseismic signal to be processed.

7. The microseismic event classification device based on deep learning according to claim 6, characterized in that: in, The type labels of the microseismic events include blasting event labels, rock fracture event labels, electromagnetic event labels, and pure noise event labels.

8. The microseismic event classification device based on deep learning according to claim 6, characterized in that: The deep learning framework with the convolutional neural network as the main body includes convolution layer 1, convolution layer 2, pooling layer, fully connected layer and output layer. The first extraction module is specifically used for: Inputting the two-dimensional spectrum graph into the convolution layer 1 of the deep learning framework based on the convolutional neural network to extract the global low-frequency features of the two-dimensional spectrum graph; The convolution layer 1 is used to extract the local high-frequency features and edge texture features of the two-dimensional spectrum graph; Downsampling and feature compression are performed on the low-frequency features, high-frequency features, and edge texture features through a pooling layer to obtain representation features of a two-dimensional spectrum graph; The representation feature is mapped to a high-dimensional vector space through a fully connected layer to obtain a feature vector V representing the low frequency, high frequency and edge texture corresponding to the two-dimensional spectrum graph c ; Among them, the size of the convolution kernel of convolution layer 1 and convolution layer 2 and the step size of the pooling operation of the pooling layer are determined by the frequency band interval corresponding to the low-frequency features and the high-frequency features.

9. The microseismic event classification device based on deep learning according to claim 6, characterized in that: The second extraction module is specifically used for: The two-dimensional spectrum graph is divided into a plurality of non-overlapping units by using a directional gradient histogram, and image gradients in the horizontal direction and the vertical direction, i.e., the horizontal gradient and the vertical gradient, are extracted in each non-overlapping unit; Based on the horizontal gradient and vertical gradient corresponding to each pixel in the two-dimensional spectrum graph, the local gradient intensity and gradient direction of each pixel are calculated; The gradient direction in each non-overlapping unit is counted or quantized and divided into multiple direction intervals, and then a direction histogram is constructed according to the local gradient strength in the multiple direction intervals to form an h-dimensional local vector with a length of H; The h-dimensional local vectors of each adjacent non-overlapping unit are combined into a block and normalized to obtain the image feature descriptor V that represents the shape and texture information of the two-dimensional spectrum graph. h .

10. The microseismic event classification device based on deep learning according to claim 6, characterized in that: The classifier is constructed by using a support vector machine or a fully connected layer plus a normalized exponential function. The kernel function of the support vector machine uses a radial basis function, and a grid search method is used to determine the kernel function parameters.

11. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

12. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-5.

Images