CN112686315A - Deep learning-based unnatural earthquake classification method - Google Patents
Deep learning-based unnatural earthquake classification method Download PDFInfo
- Publication number
- CN112686315A CN112686315A CN202011624427.0A CN202011624427A CN112686315A CN 112686315 A CN112686315 A CN 112686315A CN 202011624427 A CN202011624427 A CN 202011624427A CN 112686315 A CN112686315 A CN 112686315A
- Authority
- CN
- China
- Prior art keywords
- data
- unnatural
- earthquake
- model
- seismic
- 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.)
- Pending
Links
Images
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention belongs to the technical field of artificial intelligence, and particularly relates to a deep learning-based unnatural earthquake classification method, which comprises the following steps: reading data; segmenting data; data annotation: marking different types of unnatural earthquakes for later network training; data denoising: the unnatural earthquake often contains a large amount of random noise, and excessive noise can greatly influence the data quality, so that the model identification effect is poor, and therefore, the data needs to be subjected to noise reduction processing, and the signal-to-noise ratio of the data is improved; data enhancement: expanding the data volume to avoid poor classification effect caused by model under-fitting; normalization; identifying a model; and (5) training a model. The invention uses an artificial intelligence method to automatically and intelligently classify the unnatural earthquake, has high classification accuracy and high speed, and does not need artificial participation in the whole process. The method is used for classifying the unnatural earthquake.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, and particularly relates to a deep learning-based unnatural earthquake classification method.
Background
The frequency difference between the natural earthquake and the unnatural earthquake is large, the discrimination between the natural earthquake and the unnatural earthquake is easy, the frequencies of the unnatural earthquakes of different categories are similar, and the manual direct discrimination is difficult to be large.
Problems or disadvantages of the prior art: at present, classification of earthquake motion caused by unnatural earthquakes, namely blasting, collapse and the like, mainly depends on manual identification, and manual classification has high requirements on the professional degree of related personnel and low manual classification efficiency.
Disclosure of Invention
Aiming at the technical problem that the existing earthquake classification mainly depends on manual identification, the invention provides the deep learning-based unnatural earthquake classification method which is high in efficiency, high in accuracy and high in speed.
In order to solve the technical problems, the invention adopts the technical scheme that:
a deep learning based unnatural earthquake classification method comprises the following steps:
s1, data reading: the method comprises the following steps that an original earthquake is SEED format data, and the format data cannot be directly used, so that format conversion is carried out on the original data firstly, and then the converted data are processed;
s2, data segmentation: the data recorded by the seismograph are continuous waveform data, the data are too long in length and contain a lot of seismic information, so that different earthquakes need to be intercepted;
s3, data annotation: marking different types of unnatural earthquakes for later network training;
s4, data denoising: the unnatural earthquake often contains a large amount of random noise, and excessive noise can greatly influence the data quality, so that the model identification effect is poor, and therefore, the data needs to be subjected to noise reduction processing, and the signal-to-noise ratio of the data is improved;
s5, data enhancement: expanding the data volume to avoid poor classification effect caused by model under-fitting;
s6, normalization: for the classification of the unnatural earthquake, only the waveform and frequency characteristics of the unnatural earthquake need to be learned, the classification effect can be influenced by the existence of amplitude characteristics, and meanwhile, the model efficiency can be influenced by data with large magnitude difference, so that the data are subjected to normalization processing;
s7, identifying the model: the model is carried out in a mode of combining CNN and RNN, the CNN network at the front 2 layers is used for extracting data characteristics, shortening time step and expanding data dimension, the RNN layer is used for analyzing and extracting data time domain characteristics, the CNN layer is used for further extracting characteristics and reducing dimension of the characteristics, and the full connection layer is used for classifying the extracted characteristics;
s8, model training: inputting the data obtained after the preprocessing into a network, performing iterative training on the network, and storing the model when the model effect is not improved.
The data reading method in S1 includes: reading unnatural seismic SEED format data recorded by a seismograph by using SAC, converting the data into miniSEED format and storing the data, and then reading the miniSEED format data by using obspy, converting the data into npy format.
The data segmentation method in S2 includes: and intercepting data containing the seismic phase characteristics by a time window with the length of 30s, wherein the data sampling rate is 100Hz, and each segment of data contains 3000 time steps.
The data annotation method in S3 includes: and marking the segmented data segment by using labels, wherein the marking types comprise 0-blasting, 1-collapse, 2-reservoir earthquake and 3-mine earthquake, and the labels adopt One-hot form.
The method for reducing the noise of the data in the S4 includes: the unnatural seismic signal data contain a large amount of random noise, a [0,10HZ ] band-pass filter is used for denoising the data, and then the data are subjected to linear elimination processing.
The data enhancement method in the S5 comprises the following steps: the method for expanding the unnatural seismic signal data comprises the following steps of noise adding and segmentation, wherein the noise adding method comprises the following steps: the maximum amplitude of the piece of data is accumulated for 5% and 10% for each time step of the data. I.e. si′＝si+ α max (S), said siIs the amplitude of the ith time step, the alpha is the noise adding amplitude, the sum of the alpha is 5 percent and 10 percent, and the S is the amplitudeAmplitude of all time steps of the strip data;
and after the data is subjected to noise addition, segmenting the data again, and dividing the data with the length of 30s into small segments of data with the length of 10 s.
The normalization method in S6 is as follows: min-max normalization of data
Preventing the model from learning wrong features to overfit and speeding up model training.
The S7 further includes: using sigmoid function for classification results
And (6) performing calculation.
Compared with the prior art, the invention has the following beneficial effects:
the method comprises the steps of firstly using a band-pass filter to reduce noise, then using a deep learning data amplification mode to expand data, and then using a fusion type deep learning model to classify the unnatural earthquake. The invention uses an artificial intelligence method to automatically and intelligently classify the unnatural earthquake, has high classification accuracy and high speed, and does not need artificial participation in the whole process.
Drawings
FIG. 1 is a system flow diagram of the present invention;
FIG. 2 is a diagram of a deep learning model according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A method for classifying unnatural seismic based on deep learning, as shown in fig. 1, comprising the following steps:
step 4, data denoising: the unnatural earthquake often contains a large amount of random noise, and excessive noise can greatly influence the data quality, so that the model identification effect is poor, and therefore, the data needs to be subjected to noise reduction processing, and the signal-to-noise ratio of the data is improved;
step 7, as shown in fig. 2, identifying the model: the model is carried out in a mode of combining CNN and RNN, the CNN network at the front 2 layers is used for extracting data characteristics, shortening time step and expanding data dimension, the RNN layer is used for analyzing and extracting data time domain characteristics, the CNN layer is used for further extracting characteristics and reducing dimension of the characteristics, and the full connection layer is used for classifying the extracted characteristics;
step 8, model training: inputting the data obtained after the preprocessing into a network, performing iterative training on the network, and storing the model when the model effect is not improved.
Further, the data reading method in step 1 is as follows: reading unnatural seismic SEED format data recorded by a seismograph by using SAC, converting the data into miniSEED format and storing the data, and then reading the miniSEED format data by using obspy, converting the data into npy format.
Further, the data segmentation method in step 2 is as follows: and intercepting data containing the seismic phase characteristics by a time window with the length of 30s, wherein the data sampling rate is 100Hz, and each segment of data contains 3000 time steps.
Further, the data annotation method in step 3 is as follows: and marking the segmented data segment by using labels, wherein the marking types comprise 0-blasting, 1-collapse, 2-reservoir earthquake and 3-mine earthquake, and the labels adopt One-hot form.
Further, the method for reducing noise of data in step 4 comprises: the unnatural seismic signal data contain a large amount of random noise, a [0,10HZ ] band-pass filter is used for denoising the data, and then the data are subjected to linear elimination processing.
Further, the data enhancement method in step 5 is as follows: the method for expanding the unnatural seismic signal data comprises the following steps of noise adding and segmentation, wherein the noise adding method comprises the following steps: the maximum amplitude of the piece of data is accumulated for 5% and 10% for each time step of the data. I.e. si′＝si+αmax(S)，siTaking the amplitude of the ith time step, taking alpha as the noise amplitude, taking 5% and 10%, and taking S as the amplitude of all time steps of the data;
and after the data is subjected to noise addition, segmenting the data again, and dividing the data with the length of 30s into small segments of data with the length of 10 s.
Further, the normalization method in step 6 is as follows: min-max normalization of data
Preventing the model from learning wrong features to overfit and speeding up model training.
Further, step 7 further comprises: using sigmoid function for classification results
And (6) performing calculation.
Although only the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and all changes are encompassed in the scope of the present invention.
Claims (8)
1. A deep learning-based unnatural seismic classification method is characterized by comprising the following steps: comprises the following steps:
s1, data reading: the method comprises the following steps that an original earthquake is SEED format data, and the format data cannot be directly used, so that format conversion is carried out on the original data firstly, and then the converted data are processed;
s2, data segmentation: the data recorded by the seismograph are continuous waveform data, the data are too long in length and contain a lot of seismic information, so that different earthquakes need to be intercepted;
s3, data annotation: marking different types of unnatural earthquakes for later network training;
s4, data denoising: the unnatural earthquake often contains a large amount of random noise, and excessive noise can greatly influence the data quality, so that the model identification effect is poor, and therefore, the data needs to be subjected to noise reduction processing, and the signal-to-noise ratio of the data is improved;
s5, data enhancement: expanding the data volume to avoid poor classification effect caused by model under-fitting;
s6, normalization: for the classification of the unnatural earthquake, only the waveform and frequency characteristics of the unnatural earthquake need to be learned, the classification effect can be influenced by the existence of amplitude characteristics, and meanwhile, the model efficiency can be influenced by data with large magnitude difference, so that the data are subjected to normalization processing;
s7, identifying the model: the model is carried out in a mode of combining CNN and RNN, the CNN network at the front 2 layers is used for extracting data characteristics, shortening time step and expanding data dimension, the RNN layer is used for analyzing and extracting data time domain characteristics, the CNN layer is used for further extracting characteristics and reducing dimension of the characteristics, and the full connection layer is used for classifying the extracted characteristics;
s8, model training: inputting the data obtained after the preprocessing into a network, performing iterative training on the network, and storing the model when the model effect is not improved.
2. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the data reading method in S1 includes: reading unnatural seismic SEED format data recorded by a seismograph by using SAC, converting the data into miniSEED format and storing the data, and then reading the miniSEED format data by using obspy, converting the data into npy format.
3. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the data segmentation method in S2 includes: and intercepting data containing the seismic phase characteristics by a time window with the length of 30s, wherein the data sampling rate is 100Hz, and each segment of data contains 3000 time steps.
4. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the data annotation method in S3 includes: and marking the segmented data segment by using labels, wherein the marking types comprise 0-blasting, 1-collapse, 2-reservoir earthquake and 3-mine earthquake, and the labels adopt One-hot form.
5. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the method for reducing the noise of the data in the S4 includes: the unnatural seismic signal data contain a large amount of random noise, a [0,10HZ ] band-pass filter is used for denoising the data, and then the data are subjected to linear elimination processing.
6. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the data enhancement method in the S5 comprises the following steps: the method for expanding the unnatural seismic signal data comprises the following steps of noise adding and segmentation, wherein the noise adding method comprises the following steps: the maximum amplitude of the piece of data is accumulated for 5% and 10% for each time step of the data. I.e. s'i＝si+ α max (S), said siIs the amplitude magnitude of the ith time step, theAlpha is the noise amplitude, 5 percent and 10 percent are taken, and S is the amplitude of all time steps of the data;
and after the data is subjected to noise addition, segmenting the data again, and dividing the data with the length of 30s into small segments of data with the length of 10 s.
7. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the normalization method in S6 is as follows: min-max normalization of data
Preventing the model from learning wrong features to overfit and speeding up model training.
8. The method of deep learning based classification of unnatural seismic as claimed in claim 1, wherein: the S7 further includes: using sigmoid function for classification results
And (6) performing calculation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011624427.0A CN112686315A (en) | 2020-12-31 | 2020-12-31 | Deep learning-based unnatural earthquake classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011624427.0A CN112686315A (en) | 2020-12-31 | 2020-12-31 | Deep learning-based unnatural earthquake classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112686315A true CN112686315A (en) | 2021-04-20 |
Family
ID=75454010
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011624427.0A Pending CN112686315A (en) | 2020-12-31 | 2020-12-31 | Deep learning-based unnatural earthquake classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112686315A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086872A (en) * | 2018-07-30 | 2018-12-25 | 东北大学 | Seismic wave recognizer based on convolutional neural networks |
| CN110309304A (en) * | 2019-06-04 | 2019-10-08 | 平安科技（深圳）有限公司 | A kind of file classification method, device, equipment and storage medium |
| CN110488351A (en) * | 2019-08-15 | 2019-11-22 | 东北大学 | Seismic wave based on machine learning shakes property recognition methods |
| US20200088897A1 (en) * | 2018-09-14 | 2020-03-19 | Bp Corporation North America Inc. | Machine Learning-Based Analysis of Seismic Attributes |
| CN110927791A (en) * | 2018-09-20 | 2020-03-27 | 中国石油化工股份有限公司 | Method and device for predicting fluid by utilizing seismic data based on deep learning |
-
2020
- 2020-12-31 CN CN202011624427.0A patent/CN112686315A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086872A (en) * | 2018-07-30 | 2018-12-25 | 东北大学 | Seismic wave recognizer based on convolutional neural networks |
| US20200088897A1 (en) * | 2018-09-14 | 2020-03-19 | Bp Corporation North America Inc. | Machine Learning-Based Analysis of Seismic Attributes |
| CN110927791A (en) * | 2018-09-20 | 2020-03-27 | 中国石油化工股份有限公司 | Method and device for predicting fluid by utilizing seismic data based on deep learning |
| CN110309304A (en) * | 2019-06-04 | 2019-10-08 | 平安科技（深圳）有限公司 | A kind of file classification method, device, equipment and storage medium |
| CN110488351A (en) * | 2019-08-15 | 2019-11-22 | 东北大学 | Seismic wave based on machine learning shakes property recognition methods |
Non-Patent Citations (3)
| Title |
|---|
| 唐霞等: "基于深度学习的音乐情感识别", 《电脑知识与技术》 * |
| 李健等: "基于深度卷积神经网络的地震震相拾取方法研究", 《地球物理学报》 * |
| 江金生等: "基于卷积自编码器的地震数据处理", 《浙江大学学报(工学版)》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019169895A1 (en) | Fast side-face interference resistant face detection method | |
| CN110097075B (en) | Deep learning-based marine mesoscale vortex classification identification method | |
| CN103856689B (en) | Character dialogue subtitle extraction method oriented to news video | |
| CN110059694A (en) | The intelligent identification Method of lteral data under power industry complex scene | |
| CN111815602B (en) | Building PDF drawing wall identification device and method based on deep learning and morphology | |
| CN108805076A (en) | The extracting method and system of environmental impact assessment report table word | |
| CN114333070A (en) | Examinee abnormal behavior detection method based on deep learning | |
| CN112100435B (en) | Automatic labeling method based on edge traffic audio and video synchronization samples | |
| CN110879985B (en) | Anti-noise data face recognition model training method | |
| CN116030396B (en) | Accurate segmentation method for video structured extraction | |
| CN114170666A (en) | Facial expression recognition method based on multi-region convolutional neural network | |
| CN111178133A (en) | Natural scene image text recognition method based on pruning depth model | |
| CN107832753B (en) | Face feature extraction method based on four-value weight and multiple classification | |
| CN113052243A (en) | Target detection method based on CycleGAN and condition distribution self-adaption | |
| WO2023035558A1 (en) | Anchor point cut-based image processing method and apparatus, device, and medium | |
| CN110503092B (en) | Improved SSD monitoring video target detection method based on field adaptation | |
| CN117576632B (en) | Multi-mode AI large model-based power grid monitoring fire early warning system and method | |
| CN112151067B (en) | Digital audio tampering passive detection method based on convolutional neural network | |
| CN112686315A (en) | Deep learning-based unnatural earthquake classification method | |
| CN112507881A (en) | sEMG signal classification method and system based on time convolution neural network | |
| CN111144280A (en) | Monitoring video leaf occlusion detection method | |
| CN109117841B (en) | Scene text detection method based on stroke width transformation and convolutional neural network | |
| CN107403192B (en) | Multi-classifier-based rapid target detection method and system | |
| CN116597438A (en) | Improved fruit identification method and system based on Yolov5 | |
| CN112069995A (en) | Method, system and storage medium for extracting dense features of face in depth domain |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210420 |
|
| RJ01 | Rejection of invention patent application after publication |