CN110210296B - Microseism effective signal detection method combining U-net network and DenseNet network - Google Patents
Microseism effective signal detection method combining U-net network and DenseNet network Download PDFInfo
- Publication number
- CN110210296B CN110210296B CN201910336622.4A CN201910336622A CN110210296B CN 110210296 B CN110210296 B CN 110210296B CN 201910336622 A CN201910336622 A CN 201910336622A CN 110210296 B CN110210296 B CN 110210296B
- Authority
- CN
- China
- Prior art keywords
- network
- signal
- data set
- arrival
- learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. analysis, for interpretation, for correction
- G01V1/282—Application of seismic models, synthetic seismograms
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. analysis, for interpretation, for correction
- G01V1/30—Analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention provides a microseism effective signal detection method combining a U-net network and a DenseNet network, which comprises the following steps: (1) generating an original data set; (2) calibrating a data set; (3) constructing an MSNet network; (4) adjusting network parameters; (5) and calibrating the initial point after the signal learning. The invention also provides a microseism effective signal detection system combining the U-net network and the DenseNet network, which comprises: the system comprises an original data set generation module, a data set calibration module, an MSNet network construction module, a network parameter adjustment module and a learning initial arrival point calibration module. The invention is based on the learning and extraction of the signal characteristics of the CNN network by the U-net network, and the purpose of further optimizing the learning and extraction of the signal characteristics is achieved by combining the DenseNet network. The method has the characteristics of deeper feature extraction and more precise signal segmentation in microseism monitoring, and can be widely applied to the field of underground state monitoring.
Description
Technical Field
The invention relates to a geophysical technology for underground state monitoring, in particular to a microseism monitoring technology.
Technical Field
The traditional signal detection technology comprises the steps of carrying out spectrum analysis on signals through fast Fourier transform, carrying out time-frequency conversion on wavelets, curvelets and shear wave transform and the like so as to achieve the purposes of removing noise and retaining effective signals. However, the traditional method cannot obtain satisfactory effect if being directly applied to the microseism data, and the quality and the accuracy of microseism monitoring are directly influenced.
The signal monitoring technology based on deep learning gradually receives wide attention of people in recent years, and the main reason is that the signal monitoring technology has the characteristics of more parameters and large capacity, so that the network has strong processing capacity for mass data. The CNN network has characteristics with strong learning capability, so that the CNN network is widely applied to learning and extracting signal characteristics, but part of important characteristics can be lost when the traditional CNN network is used for extracting the characteristics, and the U-net network, a novel neural network based on an FCN network architecture, can effectively recover some important detail characteristics lost when the CNN is used for learning the characteristics, so that the time and space characteristics of signals can not be lost, but deeper characteristics can not be extracted due to the fact that the number of network layers is small.
Disclosure of Invention
The invention provides a microseism effective signal detection method combining a U-net network and a DenseNet network, which can more efficiently and accurately extract the characteristics of microseism signals.
In a first aspect of the embodiments of the present invention, there is provided a method for detecting a microseism effective signal by combining a U-net network and a DenseNet network, the method including:
step 3, adding two Densebocks in the original U-net network;
and 4, inputting the calibrated signal data set into the MSNet network for learning, performing Softmax function calculation on the output signal after learning, performing cross entropy calculation on the calculation result and the calibrated data set to obtain a loss function, and minimizing the loss function to adjust the network parameters.
And 5, adjusting the output signal after Softmax calculation through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet.
In a second aspect of the embodiments of the present invention, there is provided a microseismic active signal detection system combining a U-net network and a DenseNet network, the system including:
an original data set generation module: the device is used for combining simulation signals generated by finite difference forward modeling under different stratum models and actually acquired actual data to jointly form an original data set;
a dataset calibration module: the system is used for picking up the initial arrival of the original data set, selecting and calibrating signal waveforms at the initial arrival position and the non-initial arrival position respectively;
MSNet network construction module: the method is used for adding two Denseblocks in the original U-net network;
a network parameter adjusting module: the system is used for inputting the calibrated signal data set into the MSNet network for learning, performing Softmax function calculation on the output signal after learning, performing cross entropy calculation on the calculation result and the calibrated data set to obtain a loss function, and minimizing the loss function to adjust the network parameters.
The module is marked at the first arrival point after learning: and adjusting the output signal after Softmax calculation through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet.
According to the technical scheme, the embodiment of the invention has the following advantages:
the invention aims to overcome the defects of the background technology, and a Denseblock is added on the basis of the original U-net network to construct a novel network named MSNet so as to realize deeper feature extraction and be suitable for more finely segmenting signals.
The microseism effective signal detection method combining the U-net network and the DenseNet network has the following main benefits by adding a DenseLock layer in the U-net network:
1. the method can extract the deep level characteristics of signal data distribution so as to conveniently realize the accurate extraction of the signals.
2. Due to the characteristics of the U-net network, spatial information lost in the convolution process can be recovered to a greater extent.
3. The native U-net network has a small number of layers and cannot fully extract the characteristics of signal data distribution, so that the added block layer is very suitable for deep characteristics of signal data distribution.
Drawings
FIG. 1 is a general technical process diagram of the microseismic active signal detection method of the present invention combining U-net network and DenseNet network.
FIG. 2 is a schematic diagram of the calibration of the data set in step 2 in the microseismic active signal detection method of the present invention combining the U-net network and the DenseNet network.
FIG. 3 is a flow chart of the MSNet network constructed in step 4 of the microseism effective signal detection method combining the U-net network and the DenseNet network.
FIG. 4 is a schematic diagram of the probability of adjusting the network parameters of the microseismic active signal detection method of the present invention combining the U-net network and the DenseNet network.
Detailed Description
The present invention will be further illustrated by the following specific examples, which are to be construed as merely illustrative, and not limitative of the remainder of the disclosure in any way whatsoever.
Fig. 1 is a schematic general technical flow diagram of a microseism effective signal detection method combining a U-net network and a DenseNet network according to an embodiment of the present invention, including the following steps:
And 2, carrying out first arrival picking on the original data set by using an algorithm, and selecting and calibrating signal waveforms at first arrival positions and non-first arrival positions respectively.
The first arrival picking is to calculate signals by using an STA algorithm, select a point corresponding to the peak of a picking curve as a first arrival point, make a one-hot label, mark the first arrival point as 1, and represent the probability of the first arrival point as 1.
And 3, adding two DenseLock in the original U-net network.
The U-net network can effectively recover some important detail characteristics lost by the CNN network when the characteristics are learned, so that the time and space characteristics of signals cannot be lost, but the characteristics of signal data distribution cannot be fully extracted due to the fact that the number of network layers is small. The two DenseLock modules are added to solve the problem that the feature extraction amount of the signal is small when one DenseLock is added, and overfitting is possible when three Denselocks are added. The method can enable the U-net network to extract the signal characteristics more deeply so as to be suitable for more finely dividing the signals.
And 4, inputting the calibrated signal data set into the MSNet network for learning, performing Softmax function calculation on the output signal after learning, performing cross entropy calculation on the calculation result and the calibrated data set to obtain a loss function, and minimizing the loss function to adjust the network parameters.
Optionally, the step 4 may specifically include the following steps:
and 4.1, inputting the signal data set into the MSNet network for learning.
And 4.2, performing Softmax function calculation on the output signal after learning, calculating output probability by using the Softmax function, obtaining two classification probabilities of each point, wherein 0 is classified as a non-first arrival, 1 is classified as a first arrival, and only the probability peak value of the first arrival classification is selected as a first arrival point.
And 4.3, taking the output signal after Softmax calculation as a prediction label of the function, taking a one-hot label made by first arrival picking as a real label, and performing cross entropy calculation on the prediction label and the real label to obtain a loss function.
Step 4.4 minimizes the loss function with Adam's algorithm to adjust the network parameters.
The formula of the Softmax function for calculating the probability distribution in the step 4.2 is as follows:
wherein k is 1 to6001 denotes the number of sample points, qk(x) Representing the probability of each point being a first arrival.
It should be understood that the value of the above formula k is not limited to 1 to 6001, and depends on the number of sampling points.
The formula of the step 4.3 of adjusting the network parameter loss function is expressed as:
wherein H (p, q) represents a loss function, pi (x) represents a true tag (a one-hot tag made by first arrival picking), qi (x) represents a predicted tag (a predicted tag distribution of a trained model), an optimal network parameter is obtained by minimizing the loss function, the true tag is a signal obtained by calibrating in step 2, and the predicted tag is a first arrival position identified by network learning in step 4.
And 5, adjusting the output signal after Softmax calculation through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet.
In practice, the position of the first-arrival point serving as the label is calibrated by a traditional algorithm, the algorithm is mature at present, the first-arrival point can be basically seen by naked eyes and is convenient to check, but the calibration speed of the algorithm is very slow. In the embodiment of the invention, the calibration speed is accelerated by using the U-net neural network so as to realize batch calibration. Because the initial weight and bias of the U-net neural network are randomly set, the predicted value of the forward propagation output is often greatly different from the value of the real label before a large amount of iteration is not carried out, and the U-net neural network replaces the original weight by the weight which can minimize the loss through a method of calculating the loss and reducing the gradient so as to improve the output precision of the network, thereby achieving the aim of adjusting the network parameters.
Fig. 2 is a schematic structural diagram of a method for detecting a microseism effective signal by combining a U-net network and a DenseNet network according to an embodiment of the present invention, which specifically introduces a specific implementation process of step 3, and details are as follows:
adding two DenseLock modules in an original U-net network, wherein the MSNet network comprises the following steps:
1) shrink path through U-net: inputting a signal, wherein a signal data set passes through a contraction path of a U-net network and is used for acquiring context information of the signal;
2) by Denseblock 1: the signal data set passes through a first Dense block module added in the U-net network, the process does not affect the quantity of the characteristics of the signal data set, only enhances the utilization of the characteristics, and can realize the recovery of the original signal characteristic vector through an up-sampling structure originally possessed by the U-net network, and finally outputs the vector which is the same as the input signal vector, thereby realizing end-to-end identification;
3) by convolution: the signal data set passes through the convolution layer, and the convolution kernel slides on the signal to obtain characteristic graphs of different channels, namely the characteristics of the learning signal;
4) through pooling: the signal data set passes through a pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce unnecessary network parameters;
5) by Denseblock 2: the method is the same as that of DenseBlock1 through a second DenseBlock module added in a U-net network, so that the defect that one DenseBlock may extract the features of the data set can be avoided, and the feature extraction amount of the data set signals is increased;
6) by convolution: the signal data set passes through the convolution layer, and the convolution kernel slides on the signal to obtain characteristic graphs of different channels, namely the characteristics of the learning signal;
7) through pooling: the signal data set passes through a pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce unnecessary network parameters;
8) the sequence of expanding the path through U-net proceeds: the signal data set accurately positions the part needing to be segmented in the data set through an expansion path of the U-net network, and signals are output;
fig. 3 is a schematic diagram of data set calibration of a microseism effective signal detection method combining a U-net network and a DenseNet network according to an embodiment of the present invention, which is specifically implemented as follows:
the method comprises the steps of utilizing an STA algorithm to carry out first arrival picking on an original data set signal, selecting a point corresponding to a peak value of a picking curve as a first arrival point through output information, manufacturing a one-hot label, marking the first arrival point as 1, and using a vertical thick line in a graph 2 as the first arrival point.
Fig. 4 is a schematic diagram of adjusting network parameters of a microseism effective signal detection method combining a U-net network and a DenseNet network according to an embodiment of the present invention, which is specifically implemented as follows:
and adjusting the output signal calculated by utilizing Softmax through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet. In fig. 4, the abscissa represents sampling points, and the ordinate represents probability, where a sampling point with a probability of 1 is the probability peak, and the sampling point is the initial point.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. A microseism effective signal detection method combining a U-net network and a DenseNet network is characterized by comprising the following steps:
step 1, generating analog signals under different stratum models by utilizing finite difference forward modeling, and forming an original data set together with actual stratum data actually acquired by utilizing a detector;
step 2, carrying out first arrival picking on the original data set by using an algorithm, selecting signal waveforms at first arrival positions and non-first arrival positions and respectively calibrating the signal waveforms;
step 3, adding two Densebocks in the original U-net network;
step 4, inputting the calibrated signal data set into an MSNet network for learning, performing Softmax function calculation on the output signal after learning, performing cross entropy calculation on the calculation result and the calibrated data set to obtain a loss function, and minimizing the loss function to adjust network parameters;
the specific process of the step 4 is as follows:
step 4.1: inputting the signal data set into an MSNet network for learning, wherein the learning step specifically comprises the following steps:
1) shrink path through U-net: inputting a signal, wherein a signal data set passes through a contraction path of a U-net network and is used for acquiring context information of the signal;
2) by Denseblock 1: the signal data set realizes the recovery of the original signal characteristic vector through a first Dense block module added in the U-net network and an up-sampling structure originally possessed by the U-net network, so that a vector which is the same as an input signal vector is finally output, and end-to-end identification is realized;
3) by convolution: the signal data set passes through the convolution layer, and the convolution kernel slides on the signal to obtain a characteristic diagram of each channel, namely the characteristics of the learning signal;
4) through pooling: the signal data set passes through a pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce network parameters;
5) by Denseblock 2: the signal data set passes through a second Dense block module added in the U-net network, and the method is the same as that of DenseLock 1;
6) by convolution: the signal data set passes through the convolution layer, and the convolution kernel slides on the signal to obtain a characteristic diagram of each channel, namely the characteristics of the learning signal;
7) through pooling: the signal data set passes through a pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce network parameters;
8) the sequence of expanding the path through U-net proceeds: the signal data set accurately positions the part needing to be segmented in the data set through an expansion path of the U-net network, and signals are output;
step 4.2: performing Softmax function calculation on the output signals after learning, calculating output probability by using the Softmax function to obtain the two classification probabilities of each point, classifying 0 as a non-first-arrival, classifying 1 as a first-arrival, and selecting the probability peak value of the first-arrival classification as a first-arrival point;
step 4.3: taking the output signal after Softmax calculation as a prediction label of a function, taking a one-hot label made by first arrival picking as a real label, and performing cross entropy calculation on the prediction label and the real label to obtain a loss function;
step 4.4: minimizing a loss function with an Adam algorithm to adjust network parameters;
and 5, adjusting the output signal after Softmax calculation through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet.
2. The method for detecting the microseism effective signal by combining the U-net network and the DenseNet network according to claim 1, wherein the specific process in the step 2 is as follows:
and (3) performing first arrival picking on the signal by using an STA algorithm, selecting a point corresponding to the peak value of a picking curve as a first arrival point, making a one-hot label, and marking the first arrival point as 1.
3. The method for detecting a microseismic effect signal according to claim 1 and combining a U-net network and a DenseNet network, wherein in the step 4.2 and the step 4.3, the formula of Softmax function for calculating probability distribution is as follows:
wherein k is 1-6001 and q represents the number of sampling pointsk(x) Representing the probability of each point being a first arrival.
4. The method for detecting a microseismic effect signal combining the U-net network and the DenseNet network as claimed in claim 1, wherein the formula for adjusting the loss function of the network parameter in step 4.3 and step 4.4 is as follows:
wherein, H (p, q) represents a loss function, pi (x) represents a real label, qi (x) represents a predicted label, and the optimal network parameters are obtained by minimizing the loss function, the real label is a one-hot label produced by the first arrival picking in step 2, and the predicted label is the first arrival position identified by the network learning in step 4.
5. A microseismic active signal detection system combining a U-net network and a DenseNet network, comprising:
an original data set generation module: the device is used for combining simulation signals generated by finite difference forward modeling under different stratum models and actually acquired actual data to jointly form an original data set;
a dataset calibration module: the system is used for picking up the initial arrival of the original data set, selecting signal waveforms at the initial arrival position and the non-initial arrival position and calibrating the signal waveforms respectively;
MSNet network construction module: the method is used for adding two Denseblocks in the original U-net network;
a network parameter adjusting module: the system comprises a calibration module, a data acquisition module, a data processing module and a data processing module, wherein the calibration module is used for inputting a calibrated signal data set into an MSNet network for learning, performing Softmax function calculation on a signal output after learning, performing cross entropy calculation on a calculation result and the calibrated data set to obtain a loss function, and minimizing the loss function to adjust network parameters;
the module is marked at the first arrival point after learning: adjusting the output signal after Softmax calculation through network parameters to obtain the position of the probability peak point, namely the first arrival point calibrated after the input signal is learned through MSNet;
the MSNet network construction module specifically comprises:
u-net shrink path Unit: the method comprises the steps that a signal is input, a signal data set passes through a contraction path of a U-net network, and context information of the signal is obtained;
denseblock1 unit: the signal data set realizes the recovery of the original signal characteristic vector through a first Dense block module added in the U-net network and an up-sampling structure originally possessed by the U-net network, and finally outputs a vector which is the same as the input signal vector, thereby realizing end-to-end identification;
a convolution unit: the method is used for enabling a signal data set to pass through a convolution layer, enabling convolution kernels to slide on signals, and obtaining feature graphs of all channels, namely features of learning signals;
a pooling unit: the device is used for the signal data set to pass through the pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce network parameters;
denseblock2 unit: a second Dense block module added to the signal data set through the U-net network, the same as the Denseblock1 method;
a convolution unit: the method is used for enabling a signal data set to pass through a convolution layer, enabling convolution kernels to slide on signals, and obtaining feature graphs of all channels, namely features of learning signals;
a pooling unit: the device is used for the signal data set to pass through the pooling layer, down-sampling is carried out on signals, and the dimensionality of features is reduced so as to reduce network parameters;
u-net expansion path unit: the method is used for accurately positioning the part needing to be segmented in the data set through the expansion path of the U-net network by the signal data set and outputting signals.
6. The microseismic active signal detection system combining the U-net network and the DenseNet network as claimed in claim 5, wherein the dataset calibration module specifically comprises:
and (3) performing first arrival picking on the signal by using an STA algorithm, selecting a point corresponding to the peak value of a picking curve as a first arrival point, making a one-hot label, and marking the first arrival point as 1.
7. The system for detecting the effective microseism signals of claim 5, wherein the network parameter adjusting module specifically comprises:
a learning unit: inputting the signal data set into an MSNet network for learning;
a probability distribution calculation unit: according to the formula
Performing Softmax function calculation on the output signal after learning, wherein k is 1-6001 and represents the number of sampling points, and q isk(x) Representing the probability that each point is a first arrival, calculating the output probability by using a Softmax function to obtain the two classification probabilities of each point, wherein 0 is classified as a non-first arrival, 1 is classified as a first arrival, and only the probability peak value of the first arrival classification is selected as a first arrival point;
a cross entropy calculation unit: taking the output signal after Softmax calculation as a prediction label of a function, taking a one-hot label made by first arrival picking as a real label, and according to a formula
Performing cross entropy calculation on the predicted label and the real label to obtain a loss function, wherein H (p, q) represents the loss function, pi (x) represents the real label, namely a one-hot label made by first arrival picking, and qi (x) represents the predicted label;
a loss function calculation unit: the loss function is minimized with the Adam algorithm to adjust the network parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910336622.4A CN110210296B (en) | 2019-04-25 | 2019-04-25 | Microseism effective signal detection method combining U-net network and DenseNet network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910336622.4A CN110210296B (en) | 2019-04-25 | 2019-04-25 | Microseism effective signal detection method combining U-net network and DenseNet network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110210296A CN110210296A (en) | 2019-09-06 |
| CN110210296B true CN110210296B (en) | 2021-09-17 |
Family
ID=67786287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910336622.4A Active CN110210296B (en) | 2019-04-25 | 2019-04-25 | Microseism effective signal detection method combining U-net network and DenseNet network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210296B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11947061B2 (en) * | 2019-10-18 | 2024-04-02 | Korea University Research And Business Foundation | Earthquake event classification method using attention-based convolutional neural network, recording medium and device for performing the method |
| CN112711604B (en) * | 2019-10-25 | 2023-10-31 | 中国石油天然气股份有限公司 | Geophysical prospecting training data set construction method and device |
| CN111523258B (en) * | 2020-01-17 | 2023-06-13 | 长江大学 | Microseism effective signal first arrival pickup method and system based on MS-Net network |
| CN111505705B (en) * | 2020-01-19 | 2022-08-02 | 长江大学 | Microseism P wave first arrival pickup method and system based on capsule neural network |
| CN111505706B (en) * | 2020-04-28 | 2023-04-18 | 长江大学 | Microseism P wave first arrival pickup method and device based on deep T-Net network |
| CN113960657A (en) * | 2020-07-20 | 2022-01-21 | 中国石油天然气股份有限公司 | Seismic data feature acquisition method and system |
| CN113960654A (en) * | 2020-07-20 | 2022-01-21 | 中国石油天然气股份有限公司 | Seismic data processing method and system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2797056B1 (en) * | 1999-07-28 | 2001-09-07 | Inst Francais Du Petrole | METHOD OF ANALYSIS OF ACQUIRED SIGNALS TO AUTOMATICALLY POINT ON THEM AT LEAST ONE SIGNIFICANT MOMENT |
| CN103837891B (en) * | 2014-03-26 | 2017-01-18 | 中国石油大学(华东) | High-precision pickup method of first arrival of microseism |
| KR101738445B1 (en) * | 2016-11-16 | 2017-05-22 | 한국지질자원연구원 | A method of first arrival detection for multi-channel seismic data considering source-receiver geometry |
| CN106896407B (en) * | 2017-03-28 | 2018-07-13 | 吉林大学 | A kind of microseism signal first break pickup method based on approximate negentropy |
-
2019
- 2019-04-25 CN CN201910336622.4A patent/CN110210296B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110210296A (en) | 2019-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210296B (en) | Microseism effective signal detection method combining U-net network and DenseNet network | |
| CN108229550B (en) | Cloud picture classification method based on multi-granularity cascade forest network | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN113516012B (en) | Pedestrian re-identification method and system based on multi-level feature fusion | |
| CN114564982B (en) | Automatic identification method for radar signal modulation type | |
| CN113159048A (en) | Weak supervision semantic segmentation method based on deep learning | |
| CN111050315B (en) | Wireless transmitter identification method based on multi-core two-way network | |
| CN110619264B (en) | Method and device for identifying microseism effective signals based on UNet++, and method and device for identifying microseism effective signals based on UNet++ | |
| CN116206185A (en) | Lightweight small target detection method based on improved YOLOv7 | |
| CN112017178A (en) | Remote sensing image region change detection method based on double-temporal difference image method | |
| CN111695413A (en) | Signal first arrival pickup method and device combining U-Net and Temporal encoding | |
| CN111815526B (en) | Rain image rainstrip removing method and system based on image filtering and CNN | |
| CN114978842B (en) | Timing synchronization method of two-stage OFDM system based on neural network | |
| CN104504391B (en) | A kind of hyperspectral image classification method based on sparse features and markov random file | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| CN112818777B (en) | Remote sensing image target detection method based on dense connection and feature enhancement | |
| CN116363536B (en) | Unmanned aerial vehicle inspection data-based power grid infrastructure equipment defect archiving method | |
| CN113869289B (en) | Multi-channel ship radiation noise feature extraction method based on entropy | |
| CN114359786A (en) | Lip language identification method based on improved space-time convolutional network | |
| CN115035408A (en) | Unmanned aerial vehicle image tree species classification method based on transfer learning and attention mechanism | |
| CN114049519A (en) | Optical remote sensing image scene classification method | |
| CN113537573A (en) | Wind power operation trend prediction method based on dual space-time feature extraction | |
| CN113609970A (en) | Underwater target identification method based on grouping convolution depth U _ Net | |
| CN113111774A (en) | Radar signal modulation mode identification method based on active incremental fine adjustment | |
| CN112434716A (en) | Underwater target data amplification method and system based on conditional adversarial neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |