CN113705654B - FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium - Google Patents
FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium Download PDFInfo
- Publication number
- CN113705654B CN113705654B CN202110970928.2A CN202110970928A CN113705654B CN 113705654 B CN113705654 B CN 113705654B CN 202110970928 A CN202110970928 A CN 202110970928A CN 113705654 B CN113705654 B CN 113705654B
- Authority
- CN
- China
- Prior art keywords
- model
- arrival
- ffpn
- microseism
- output
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 53
- 230000006870 function Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000015654 memory Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 10
- 238000005457 optimization Methods 0.000 abstract description 5
- 230000008569 process Effects 0.000 abstract description 4
- 230000000644 propagated effect Effects 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000013136 deep learning model Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses an FFPN-based microseism first-arrival intelligent pickup method, an FFPN-based microseism first-arrival intelligent pickup system, FFPN-based microseism first-arrival intelligent pickup equipment and an FFPN-based microseism first-arrival intelligent pickup storage medium, and belongs to the field of exploration geophysics. According to the network model, the output of two scale features is selected as the optimization condition of the model, the finer first-arrival pickup output is calculated through the low-level features with rich detail features, the classification output is calculated through the high-level features with larger receptive fields, the loss of the two parts of output is calculated, and the two parts of output are propagated in the opposite direction in the training process to achieve the optimization of the network model, so that a more accurate classification result (first-arrival pickup result) is obtained. The point-aware loss function provided by the model can calculate the loss of low-level characteristic output, and the effect of optimizing high-level characteristic output is achieved by optimizing low-level characteristics.
Description
Technical Field
The invention belongs to the field of exploration geophysics, and relates to a micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium based on an FFPN model.
Background
First-arrival picking plays an important role in microseismic processing and imaging. With the continuous increase of the received data volume, the manual picking of the first arrival of the microseism is very time-consuming, and the manual interpretation of the first arrival of the microseism depends on the experience of interpreters, so that human errors and interference are easy to generate. During the last decades researchers have struggled to achieve automated/semi-automated microseism first arrival pickup.
STA/LTA is a conventional microseism first arrival pickup method implemented by calculating a moving average energy ratio of a short time window and a long time window. The STA/LTA method is simple in calculation and convenient to operate, and is widely applied to applications such as first arrival interpretation of micro-earthquakes. However, the STA/LTA approach has two major drawbacks: first, the first arriving waveform cannot always be received accurately due to the short window length; second, microseism data often contains strong interference noise, which can affect the accuracy and stability of microseism first arrival pick-up results. Based on this, various automatic/semiautomatic microseism first arrival picking methods are proposed, such as local linear embedding and improved particle swarm optimization clustering, adaptive multi-band selection, kurtosis ratio method, and fast AIC method. The traditional microseism first arrival pickup method is good in performance on microseism data with strong peak amplitude and stable noise, and can accurately explain the microseism first arrival. However, in most practical situations, noise and significant wave energy losses are significant, and these conventional methods often exhibit instability, i.e., the inability to accurately pick up a microseismic first arrival.
Deep learning is a branch of machine learning, and has attracted attention in various fields as computing power evolves. Convolutional neural networks (Convolutional Neural Networks, CNN) proposed by LeCun et al (1989) are one of the most popular and most widely used deep learning algorithms. A typical CNN model is initially used for an image classification task, i.e. converting an image structure into a certain category of feature information, where the classification task concerns the overall features, and finally gives a content description of the whole picture. The last layer of the classification network is generally a fully connected layer for predicting target labels, and the classical CNN network model provides a research basis for other tasks. In addition, expanding to the field of target detection, the performance target needs to pay attention to a specific object target, and the category information and the position information of the target are required to be obtained simultaneously, and the output of the type of detection network is a list, namely each item comprises the category and the position of a detection frame. Some typical networks are Regions with CNN features (RCNN), you Only Look Once (YOLO), feature Pyramid Network (FPN), and the like. With the deep research and the increasing demand, classification tasks are gradually refined to semantic segmentation tasks with pixels as analysis units, namely the image processing field, and network models such as Fully Convolutional Networks (FCN), segNet, deeplab and the like are sequentially proposed to solve the problem.
In recent years, deep learning has achieved good performance in many fields, and as researchers continue to explore and practice, many algorithms and models based on deep learning are used for seismic processing and interpretation of microseismic data, such as Long Short-Term Memory (LSTM) and the like. With the development of these deep learning models, many methods for microseism first arrival pickup have also emerged, such as Atomic Energy Network (AEnet), encoder-decoder Network (Encoder-decoder Network), transfer learning (Transfer Learning), and Human-computer interaction learning (Human-Computer Interaction Learning). The deep learning models and the successful application thereof in the first-arrival picking of the micro-earthquake show the application prospect of the deep learning models in the first-arrival picking of the micro-earthquake. However, while these deep learning models show good results in microseismic first-arrival pickup applications, their generalization and computational efficiency still require further improvements and enhancements.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a micro-seismic first-arrival intelligent pickup method, a system, equipment and a storage medium based on an FFPN model, which aim to solve the technical problems of generalization of micro-seismic first-arrival pickup and low calculation efficiency in the prior art.
The invention provides a micro-seismic first-arrival intelligent pickup method of an FFPN model, which comprises the following steps:
s1, preprocessing an actual microseism data set, establishing a sample data set and a label data set, randomly selecting sample data and label data corresponding to the sample data in a certain proportion as a training set, and taking the rest sample data and label data corresponding to the sample data as a test set;
s2, reserving a basic structure of an original FPN model, replacing the two-dimensional convolution with one-dimensional convolution, constructing an initial one-dimensional feature pyramid first-arrival pick-up network model FPN, and training the one-dimensional feature pyramid initial pick-up network model FPN to obtain a one-dimensional feature pyramid initial pick-up network training model;
s3, performing microseism first-arrival prediction according to the one-dimensional feature pyramid initial quality pickup network training model obtained in the S2, building an improved feature pyramid model FFPN with double output and double loss, introducing a specific loss function into the improved pyramid model FFPN, and training the improved pyramid model FFPN by adopting the training set obtained in the S1 to obtain actual microseism data first-arrival intelligent pickup.
Preferably, in S3, the FFPN network model uses one-dimensional convolution entirely, canceling the second layer output of the FPN network model, modifying the first layer output to calculate the first-arrival pick-up output, and calculating the classification output by the high-level features with receptive fields.
Preferably, in S3, the training data selected randomly in S1 is input into the improved pyramid model FFPN, and the improved pyramid model FFPN is supervised by using the two classification results corresponding to the first arrival pickup tag, and training is performed to obtain the improved pyramid training model.
Preferably, the classification output loss calculation is as shown in formula (1):
l cla ＝CrossEntropy(m pred ,m g ) (1)
wherein CrossEntropy is the cross entropy loss, m pred Is classified output of network, m g Is a classification truth.
Preferably, the specific loss function is added as point-aware loss, and the loss function is used for calculating the output error of the first arrival pickup point of the microseism.
Preferably, the method for calculating the loss function as point-aware loss is as shown in formula (2):
l p ＝CrossEntropy(p pred ,p g ) (2)
wherein CrossEntropy is cross entropy loss, p pred Is the first arrival pick-up output of the network, p g Is a first arrival pick-up truth value.
Preferably, the calculation method of the total loss function of the improved feature pyramid model FFPN is as shown in formula (3):
l＝l cla +λl p (3)
where lambda is used as a super parameter to equalize the gradients of the two loss functions.
The invention also discloses a first arrival picking system of the micro-seismic first arrival intelligent picking method based on the FFPN model, which comprises the following steps:
the data acquisition module is used for preprocessing the actual microseism data to acquire a sample data set and a tag data set, randomly selecting a certain proportion of the actual microseism data and the corresponding tag data thereof as a training set, and taking the rest as a test set;
the model acquisition module is used for building and training a one-dimensional characteristic pyramid initial quality pickup network model to acquire a one-dimensional characteristic pyramid initial quality pickup network training model;
the first arrival picking module is used for carrying out microseism first arrival prediction on the one-dimensional feature pyramid initial quality picking network training model, constructing and improving the feature pyramid model and training the feature pyramid model to obtain first arrival intelligent picking of actual microseism data.
The invention provides computer equipment, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of a micro-seismic first-arrival intelligent pickup method based on an FFPN model when executing the computer program.
The invention provides a computer readable storage medium, which stores a computer program, and the computer program realizes the steps of a micro-seismic first-arrival intelligent pickup method based on an FFPN model when being executed by a processor.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a micro-seismic first-arrival intelligent pickup method based on an FFPN model, which combines the characteristics of actual micro-seismic data to carry out structure adjustment and improvement on an initial FPN network model and provides a micro-seismic first-arrival intelligent pickup method based on an improved feature pyramid (FFPN) model. The model reserves the multi-scale feature fusion capability of the original FPN network model, is suitable for the first arrival intelligent pickup problem, and achieves a finer first arrival output result. The original one-dimensional FPN network has the core advantage of multi-scale feature fusion capability, and the model can fully utilize detail information rich in low-level features and semantic information rich in high-level networks, so that multi-scale feature extraction and learning can be performed. The FPN model also uses the output of multiple scale features to directly calculate model loss, thereby optimizing the network model to achieve more accurate feature extraction and learning. The first arrival intelligent picking method based on the improved characteristic pyramid model with double output and double loss is obtained by combining actual microseism data characteristics and first arrival characteristics to deform on the basis of an original one-dimensional FPN network, and the core adjustment thought is to modify multi-layer output and add specific loss functions of the model structure of the original FPN model for specific tasks such as target detection and the like, so that the method is more suitable for the first arrival intelligent picking task. The invention finally obtains the first-arrival intelligent pickup FFPN model with double output and double loss, and can realize accurate and efficient first-arrival intelligent pickup work of micro-earthquakes by utilizing the FFPN model trained by the invention.
Further, the invention solves the problem that the original FPN network model is applied to the first arrival picking problem and shows classification ambiguity at the first arrival boundary, the FFPN network model provided by the invention selects the output of two scale characteristics as the optimization condition (o 1, o 3) of the model, calculates the finer first arrival picking output (o 1) only through the low-level characteristics with abundant detail characteristics, calculates the classification output (o 3) through the high-level characteristics with larger receptive field, calculates the loss of the two parts of outputs, and performs back propagation in the training process to achieve the optimization of the network model, thereby obtaining more accurate classification results (first arrival picking results).
Furthermore, the invention provides a point-aware loss function, which only calculates the output error of the first arrival pick-up point, specifically calculates the loss (o 1) of the low-level characteristic output, and optimizes the high-level characteristic output by optimizing the low-level characteristic.
The invention also provides a first-arrival picking system of the micro-seismic first-arrival intelligent picking method based on the FFPN model, which adopts the modularized thought to realize the first-arrival picking, so that each module is mutually independent, and unified management of each module is convenient to realize.
Drawings
FIG. 1 is actual microseismic data and its tags used in the present invention;
FIG. 2 shows training data and labels (red part is training set) used in the present invention;
FIG. 3 is a schematic diagram of the original FPN deep neural network structure;
FIG. 4 is a schematic diagram of the FFPN deep neural network structure of the present invention;
FIG. 5 shows the first arrival results of the present invention using the present model (red dots), the encoding-decoding model (blue dots) and the STA/LTA method (green dots) in actual microseism data prediction;
FIG. 6 shows the first arrival results of the present invention predicted from microseism data with a signal-to-noise ratio of 0.50 using FFPN (red dots), encoding-decoding model (blue dots) and STA/LTA method (green dots);
FIG. 7 shows the first arrival results of the present invention predicted from microseism data with a signal-to-noise ratio of 0.30 using FFPN (red dots), encoding-decoding model (blue dots) and STA/LTA method (green dots);
FIG. 8 is a first arrival result of microseism data prediction with a signal-to-noise ratio of 0.15 using FFPN (red dots), encoding-decoding model (blue dots) and STA/LTA method (green dots);
fig. 9 is a first arrival result of actual microseism data predictions of other work areas using a trained FFPN model in accordance with the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The invention is described in further detail below with reference to the attached drawing figures:
the invention provides an intelligent first arrival picking method of microseism based on an FFPN model, and referring to FIG. 1, actual microseism data and a first arrival picking label thereof used in the invention are shown; randomly selecting 80 channels of data from the left rectangular frame as a training set, and using the rest as a test set; referring to fig. 2, which is a schematic diagram of microseism data in a left rectangular frame in fig. 1, wherein a red part is randomly selected part data, and is used as a training set for training a model, the microseism first arrival intelligent pickup method comprises the following steps:
s1, performing manual first-arrival interpretation according to an actual microseism data set, establishing a sample data set and a label data set for subsequent microseism first-arrival intelligent pickup, randomly selecting part of actual microseism data and label data corresponding to the actual microseism data as a training set, and taking the rest sample data and label data corresponding to the sample data as a test set;
s2, reserving a basic structure of an original FPN model, replacing two-dimensional convolution with one-dimensional convolution, building an initial one-dimensional feature pyramid (Feature Pyramid Network, FPN) first-arrival pick-up network model, and training the one-dimensional feature pyramid model FPN to obtain a one-dimensional feature pyramid training model;
s3, optimizing and adjusting an initial one-dimensional feature pyramid network model FPN, building an improved feature pyramid model (Fine-tuning Feature Pyramid Network, FFPN) with double output and double loss, carrying out improved feature pyramid model training again by utilizing a part of randomly selected actual microseism data sets and label data sets, and carrying out actual microseism data first-arrival intelligent pickup by utilizing the trained FFPN model.
Referring to fig. 4, a schematic diagram of an FFPN deep neural network structure according to the present invention is shown, wherein the left image is a seismic shot set, each time the network inputs a piece of data shown in a rectangular frame, and outputs a row of two classified output images; the rightmost image represents the auxiliary output of the network, the bars in the figure represent the characteristic diagrams of the outputs of the layers of the network, the arrows represent the convolution operations, and the detailed operation steps of S1 to S3 are as follows:
s1, establishing a sample data set and a label data set for a first arrival intelligent pickup model of a microseism, randomly selecting a certain proportion of actual microseism data and a corresponding label data set as a training set, and taking the rest sample data and label data corresponding to the sample data as a test set; wherein, training set and test set proportion is 1:5.4.
the supervised deep learning model is utilized to carry out intelligent pickup of the first arrival of the microseism, a sample data set and a label data set are needed, manual first arrival interpretation is carried out according to the actual microseism data set, and the sample data set and the label data set are established for the follow-up intelligent pickup model of the first arrival of the microseism.
S2, reserving a basic structure of an original FPN model, replacing two-dimensional convolution with one-dimensional convolution, building an initial one-dimensional feature pyramid first-arrival pickup network model (Feature Pyramid Network, FPN), training the initial one-dimensional feature pyramid first-arrival pickup network model FPN, and analyzing the network structure and the actual microseism data first-arrival pickup effect;
the original FPN network proposed was initially designed to solve the conventional target detection problem, and the model has a multi-scale and multi-level output (multiscale and multiscale output) network structure. Because the target scale in the target detection task has a large difference, the conventional network model often shows poor effect on the detection of small objects; the high-level features have large receptive fields, so that more semantic information can be captured, but the original image area corresponding to the unit pixels is large, so that the mapping of small objects in the high-level features is inaccurate. The microseism first arrival has very fine structural characteristics and is difficult to accurately position in high-level features, so that the invention builds an initial one-dimensional FPN first arrival pickup model on the basis of referencing the multi-scale feature fusion of an original FPN network.
The FPN network model is initialized through training, the network structure characteristics and the first arrival picking effect are analyzed in detail, and the network model is found to have poor classification effect near the first arrival. The reason for this network model performing poorly is as follows: the original FPN model aims at solving the problem of target detection of conventional images, and the target detection involves feature analysis of each region of the whole image, so that the feature extraction is not targeted, and therefore, features near the first arrival extracted by the model are not unique and targeted. The first arrival is very fine in microseismic data, so that the network model needs to be forced to strengthen the extraction and learning of data features nearby the first arrival so as to achieve better classification effect (namely, first arrival pickup effect).
For the original FPN network model, although the network model uses multi-scale feature fusion, the detail information of low-level features is fully utilized, data features around the first arrival are not extracted pertinently, and the first arrival features learned by the network model are insufficient to better optimize classification results near the first arrival.
S3, optimizing and adjusting an initial one-dimensional feature pyramid first arrival picking network model FPN, constructing an improved feature pyramid model (Fine-tuning Feature Pyramid Network, FFPN) with double output and double loss, introducing a specific loss function into the improved pyramid model FFPN, carrying out model training again by utilizing randomly selected partial microseism data and label data, and carrying out actual microseism data first arrival intelligent picking by utilizing the trained FFPN model.
Aiming at the characteristics of actual microseism data, first arrival data and initial FPN network, the invention readjusts the topological structure of the initial FPN network, and provides an improved FPN network structure.
The method and the device have the core of solving the problems that an original FPN model has no pertinence in feature extraction, the data features near the first arrival of the microseism cannot be accurately extracted, and finally the classification effect near the first arrival is poor. Firstly, the FFPN network model provided by the invention cancels the second layer output (o 2) of an original FPN network, modifies the first layer output into a first arrival pickup output (o 1) with more refined calculation, and calculates a classification output (o 3) through the high-level characteristics with larger receptive field; inputting the training data randomly selected in the S1 into an improved pyramid model FFPN, supervising the improved pyramid model FFPN by utilizing a two-classification result corresponding to the first arrival pickup tag, and training to obtain an improved pyramid training model; the output loss calculation of the second class is shown in the formula (1):
l cla ＝CrossEntropy(m pred ,m g ) (1)
wherein CrossEntropy is the cross entropy loss, m pred Is classified output of network, m g Is a classification truth.
Secondly, a point-aware loss function is provided, the loss function only calculates the output error of the first arrival pick-up point, specifically, calculates the loss (o 1) of the low-level characteristic output, and the effect of optimizing the high-level characteristic output is achieved by optimizing the low-level characteristic, wherein the calculation method of the loss function is shown in a formula (2):
l p ＝CrossEntropy(p pred ,p g ) (2)
wherein CrossEntropy is cross entropy loss, p pred Is the first arrival pick-up output of the network, p g Is a first arrival pick-up truth value.
The calculation method of the total loss function of the improved feature pyramid model FFPN is shown in a formula (3):
l＝l cla +λl p (3)
where lambda is used as a super parameter to equalize the gradients of the two loss functions.
And then training the network provided by the invention by using actual microseism data and label data to obtain an FFPN model.
The invention also discloses a first arrival picking system of the micro-seismic first arrival intelligent picking method based on the FFPN model, which comprises the following steps:
the data acquisition module is used for preprocessing the actual microseism data to acquire a sample data set and a tag data set, randomly selecting a certain proportion of the actual microseism data and the corresponding tag data thereof as a training set, and taking the rest as a test set;
the model acquisition module is used for building and training a one-dimensional characteristic pyramid initial quality pickup network model to acquire a one-dimensional characteristic pyramid initial quality pickup network training model;
the first arrival picking module is used for carrying out microseism first arrival prediction on the one-dimensional feature pyramid initial quality picking network training model, constructing and improving the feature pyramid model and training the feature pyramid model to obtain first arrival intelligent picking of actual microseism data.
Referring to fig. 3, which is a top-down network corresponding to a feature pyramid diagram of the feature pyramid diagram network, if targets with different sizes exist in the original image in fig. 3, and the different targets have different features, simple targets can be distinguished by using the features of the shallow layer; complex objects can be distinguished by deep features. From top to bottom in fig. 3, layer 1 outputs the instance segmentation result of the larger target, layer 2 outputs the instance detection result of the next largest target, and layer 3 outputs the instance segmentation result of the smaller target. The same is true of detection, with simple targets being output at layer 1, more complex targets being output at layer 2, and complex targets being output at layer 3. The enlarged area in fig. 3 is the cross-connect, and the main effect of using 1*1 convolution kernels is to reduce the number of convolution kernels.
The trained FFPN Network model is used for first-arrival intelligent pickup of actual microseismic data and is compared with an Encoder-decoder Network (Encoder-decoder Network) model and STA/LTA algorithm. From the initial microseism data first-arrival interpretation results shown in fig. 5, it can be found that the model (red dots) provided by the invention can obtain more accurate microseism first-arrival interpretation results, and the encoder/decoder network model (blue dots) and the STA/LTA algorithm (green dots) cannot obtain accurate first-arrival interpretation results at certain positions, as shown by yellow ellipses. In order to further verify the stability of the model provided by the invention, random noise with the signal-to-noise ratio of 0.50 is added into original microseism data, and as can be found from the result of fig. 6, the model (red dots) provided by the invention can still accurately explain the first arrival of the microseism under the condition of noise, the encoder decoder network model (blue dots) and the STA/LTA algorithm (green dots) have poor stability, and an accurate first arrival interpretation result cannot be obtained, as shown by yellow ellipses; as can be found from the result in FIG. 7, the model (red dot) provided by the invention can accurately explain the first arrival of the microseism under the condition of noise, the encoder decoder network model (blue dot) and the STA/LTA algorithm (green dot) have poor stability, and an accurate first arrival interpretation result cannot be obtained, as shown by a yellow ellipse; as can be seen from the result in FIG. 8, the model (red dot) provided by the invention can accurately explain the first arrival of the microseism under the condition of noise, and the encoder decoder network model (blue dot) and the STA/LTA algorithm (green dot) have poor stability, so that an accurate first arrival interpretation result cannot be obtained, as shown by a yellow ellipse. In order to further verify the generalization of the model provided by the invention, the model is used for performing first-arrival intelligent interpretation on microseism data of other work areas, and the result of the graph 9 shows that the model can accurately interpret the first-arrival of microseism of different work areas, so that the generalization of the model is verified.
The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (5)
1. An intelligent pickup method for first arrival of microseism based on FFPN model is characterized by comprising the following steps:
s1, preprocessing actual microseism data, establishing a sample data set and a label data set, randomly selecting a certain proportion of sample data and label data corresponding to the sample data from the actual microseism data as a training set, and taking the rest of sample data and label data corresponding to the sample data as a test set;
s2, reserving a basic structure of an original FPN model, replacing the two-dimensional convolution with one-dimensional convolution, constructing a one-dimensional feature pyramid initial quality picking network model FPN, and training the one-dimensional feature pyramid initial quality picking network model FPN to obtain a one-dimensional feature pyramid initial quality picking network training model;
s3, performing microseism first-arrival prediction according to the one-dimensional feature pyramid initial quality pickup network training model obtained in the S2, building an improved feature pyramid model FFPN with double output and double loss, introducing a specific loss function to the improved pyramid model FFPN, and training the improved pyramid model FFPN by adopting the training set obtained in the S1 to obtain actual microseism data first-arrival intelligent pickup;
in S3, the FFPN network model totally uses one-dimensional convolution, the second-layer output of the FPN network model is cancelled, the first-layer output is modified into calculation first-arrival pickup output, and the classification output is calculated through the high-layer characteristics with receptive fields;
inputting the training data randomly selected in the S1 into an improved pyramid model FFPN, supervising the improved pyramid model FFPN by utilizing a two-classification result corresponding to the first arrival pickup tag, and training to obtain an improved pyramid training model;
the two-class output loss calculation is shown in formula (1):
wherein CrossEntropy is the cross entropy loss, m pred Is classified output of network, m g Is a classification truth value;
adding a specific loss function which is used for calculating the output error of the first arrival pick-up point of the microseism into the point-aware loss;
the calculation method for the loss function of point-aware is shown in the formula (2):
wherein CrossEntropy is cross entropy loss, p pred Is the first arrival pick-up output of the network, p g Is a first arrival pick-up truth value.
2. The FFPN model-based microseism first arrival intelligent pickup method according to claim 1, wherein a total loss function calculation method of the improved feature pyramid model FFPN is as shown in formula (3):
where lambda is used as a super parameter to equalize the gradients of the two loss functions.
3. A system for implementing the FFPN model-based micro-seismic first arrival intelligent pickup method of any one of claims 1 to 2, comprising:
the data acquisition module is used for preprocessing the actual microseism data to acquire a sample data set and a tag data set, randomly selecting a certain proportion of the actual microseism data and the corresponding tag data thereof as a training set, and taking the rest as a test set;
the model acquisition module is used for building and training a one-dimensional characteristic pyramid initial quality pickup network model to acquire a one-dimensional characteristic pyramid initial quality pickup network training model;
the first arrival picking module is used for carrying out microseism first arrival prediction on the one-dimensional feature pyramid initial quality picking network training model, constructing and training the improved feature pyramid model FFPN, and acquiring the first arrival intelligent picking of actual microseism data.
4. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the FFPN model-based micro-seismic first-arrival intelligent pickup method according to any one of claims 1 to 2.
5. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the FFPN model-based microseism first arrival intelligent pickup method of any one of claims 1 to 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970928.2A CN113705654B (en) | 2021-08-23 | 2021-08-23 | FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970928.2A CN113705654B (en) | 2021-08-23 | 2021-08-23 | FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705654A CN113705654A (en) | 2021-11-26 |
| CN113705654B true CN113705654B (en) | 2024-04-05 |
Family
ID=78654276
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110970928.2A Active CN113705654B (en) | 2021-08-23 | 2021-08-23 | FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705654B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027452A (en) * | 2019-12-05 | 2020-04-17 | 中国矿业大学(北京) | Microseismic signal arrival time and seismic phase identification method and system based on deep neural network |
| CN111695413A (en) * | 2020-04-28 | 2020-09-22 | 长江大学 | Signal first arrival pickup method and device combining U-Net and Temporal encoding |
| WO2021139069A1 (en) * | 2020-01-09 | 2021-07-15 | 南京信息工程大学 | General target detection method for adaptive attention guidance mechanism |
-
2021
- 2021-08-23 CN CN202110970928.2A patent/CN113705654B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027452A (en) * | 2019-12-05 | 2020-04-17 | 中国矿业大学(北京) | Microseismic signal arrival time and seismic phase identification method and system based on deep neural network |
| WO2021139069A1 (en) * | 2020-01-09 | 2021-07-15 | 南京信息工程大学 | General target detection method for adaptive attention guidance mechanism |
| CN111695413A (en) * | 2020-04-28 | 2020-09-22 | 长江大学 | Signal first arrival pickup method and device combining U-Net and Temporal encoding |
Non-Patent Citations (2)
| Title |
|---|
| 基于深度信念神经网络的微震波到时拾取方法;郑晶;曹子原;姜天琪;沈帅帅;;矿业科学学报(第06期);全文 * |
| 基于深度卷积生成对抗网络的地震初至拾取;周创;居兴国;李子昂;刘小民;;石油物探(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705654A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111583263B (en) | Point cloud segmentation method based on joint dynamic graph convolution | |
| CN109271960A (en) | A kind of demographic method based on convolutional neural networks | |
| CN109117793B (en) | Direct-push type radar high-resolution range profile identification method based on deep migration learning | |
| CN108985268B (en) | Inductive radar high-resolution range profile identification method based on deep migration learning | |
| CN107944410B (en) | Cross-domain facial feature analysis method based on convolutional neural network | |
| CN109635744A (en) | A kind of method for detecting lane lines based on depth segmentation network | |
| CN112541508A (en) | Fruit segmentation and recognition method and system and fruit picking robot | |
| Wang et al. | Distilling knowledge from an ensemble of convolutional neural networks for seismic fault detection | |
| Chu et al. | Method of image segmentation based on fuzzy C-means clustering algorithm and artificial fish swarm algorithm | |
| Lv et al. | A visual identification method for the apple growth forms in the orchard | |
| CN110263855B (en) | Method for classifying images by utilizing common-basis capsule projection | |
| CN108805102A (en) | A kind of video caption detection and recognition methods and system based on deep learning | |
| CN114764577A (en) | Lightweight modulation recognition model based on deep neural network and method thereof | |
| CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
| CN117079098A (en) | Space small target detection method based on position coding | |
| CN111882057A (en) | Multi-stage deep learning method oriented to space-time sequence data modeling and application | |
| CN110674845B (en) | Dish identification method combining multi-receptive-field attention and characteristic recalibration | |
| CN113297964A (en) | Video target recognition model and method based on deep migration learning | |
| CN113705654B (en) | FFPN model-based micro-seismic first-arrival intelligent pickup method, system, equipment and storage medium | |
| Li et al. | Detecting Plant Leaves Based on Vision Transformer Enhanced YOLOv5 | |
| CN113837039A (en) | Fruit growth form visual identification method based on convolutional neural network | |
| CN114358058A (en) | Wireless communication signal open set identification method and system based on deep neural network | |
| CN112487909A (en) | Fruit variety identification method based on parallel convolutional neural network | |
| Hu et al. | Image Recognition of Marine Organisms Based on Convolutional Neural Networks | |
| CN113313091B (en) | Density estimation method based on multiple attention and topological constraints under warehouse logistics |
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 |