CN109034246A - A kind of the determination method and determining system of roadbed saturation state - Google Patents
A kind of the determination method and determining system of roadbed saturation state Download PDFInfo
- Publication number
- CN109034246A CN109034246A CN201810844761.3A CN201810844761A CN109034246A CN 109034246 A CN109034246 A CN 109034246A CN 201810844761 A CN201810844761 A CN 201810844761A CN 109034246 A CN109034246 A CN 109034246A
- Authority
- CN
- China
- Prior art keywords
- saturation state
- error
- roadbed
- unsupervised
- prediction model
- 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.)
- Granted
Links
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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Neurology (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention announces the determination method and determining system of a kind of roadbed saturation state.The method of determination includes: to obtain target roadbed gamma bandpass filter group cepstrum coefficient vector;Target roadbed gamma bandpass filter group cepstrum coefficient vector is input in saturation state prediction model, the saturation state of target roadbed is obtained;The method for building up of saturation state prediction model specifically includes: obtaining training sample, training sample includes multiple training datas, and training data is the gamma bandpass filter group cepstrum coefficient vector of roadbed known to saturation state;Training sample is inputted into deepness belief network model, unsupervised training is carried out to deepness belief network model, obtains saturation state prediction model.Determining method provided by the invention and determining system, participate in without artificial, do not depend on the artificial experience and subjective judgement for the treatment of people, therefore, it is not only high-efficient, but also can effectively avoid due to individual difference bring error, so as to accurately and reliably carry out saturation state detection to existing roadbed.
Description
Technical field
The present invention relates to geophysical probing technique fields, a kind of determination method more particularly to roadbed saturation state and
Determine system.
Background technique
In recent years, with the tremendous expansion of communications and transportation, highdensity vehicle operation and a large amount of underground engineering construction make
The probability increase that diseases occur for all kinds of roadbeds is obtained, but also the degree of original subgrade defect is aggravated, subgrade defect can cause vehicle
Operation smooth degree is deteriorated, and can not only threaten traffic safety, but also threaten the life security and property safety of the people, therefore, and
Shi Faxian subgrade defect simultaneously carry out it is corresponding administer to promoting traffic safety progress to be of great significance to, research quickly,
Reliably, comprehensively existing detection method of roadbed becomes problem in the urgent need to address.
The conventional detection method of subgrade defect with dig spy, pricker visit based on, disadvantage is however that at high cost, long in time limit, information content
It is small, arbitrariness is also big, be easy tested roadbed is damaged.With the progress of research in recent years, the height of roadbed moisture content is universal
It is considered one of an important factor for influencing roadbed quality state, the correct Fast Classification of roadbed saturation state is underground disease
The important evidence of risk assessment.Since the variation of roadbed saturation state can cause the significant change of dielectric permittivity and conductivity
Change, the attributes such as waveform, frequency for receiving signal so as to cause Ground Penetrating Radar change, therefore are satisfied the need using Coherent Noise in GPR Record
It is theoretical feasible scheme that base saturation state, which carries out classification prediction,.Therefore, Ground Penetrating Radar is because having light and fast, anti-interference ability
By force, detection method the advantages that high resolution, quick nondestructive, intuitive section as effective assessment roadbed state.
However, in the existing roadbed saturation state detection technique based on the radar exploration technique, the place of Coherent Noise in GPR Record
Reason relies on the artificial experience for the treatment of people and subjective judgement carries out manual sort, identifies difficulty, low efficiency, there are great individuals
Difference influences the accuracy of roadbed state judgement, drags slow roadbed testing progress.
Therefore, the saturation state for how accurately and efficiently determining roadbed, becomes the skill of those skilled in the art's urgent need to resolve
Art problem.
Summary of the invention
The object of the present invention is to provide a kind of determination method of roadbed saturation state and determine system, it can be accurately and efficiently
Determine the saturation state of roadbed.
To achieve the above object, the present invention provides following schemes:
A kind of determination method of roadbed saturation state, the determining method include:
Obtain target roadbed gamma bandpass filter group cepstrum coefficient vector;
The target roadbed gamma bandpass filter group cepstrum coefficient vector is input in saturation state prediction model, is obtained
The saturation state of target roadbed;Wherein, the input of the saturation state prediction model is that target roadbed gamma bandpass filter group is fallen
Pedigree number vector, the output of the saturation state prediction model are the saturation state of target roadbed;The saturation state predicts mould
Type is established according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm;The saturation state
The method for building up of prediction model specifically includes:
Training sample is obtained, the training sample includes multiple training datas, and the training data is known to saturation state
Roadbed gamma bandpass filter group cepstrum coefficient vector;
The training sample is inputted into deepness belief network model, unsupervised instruction is carried out to the deepness belief network model
Practice, obtains the saturation state prediction model.
Optionally, described that the training sample is inputted into deepness belief network model, to the deepness belief network model
Unsupervised training is carried out, the saturation state prediction model is obtained, specifically includes:
The training sample is successively inputted into each limited Boltzmann machine, each limited Boltzmann machine is carried out
Unsupervised training obtains corresponding unsupervised limited Boltzmann machine, and the lower layer of the unsupervised limited Boltzmann machine is can
Depending on layer, the upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Each unsupervised limited Boltzmann machine is stacked, and in the unsupervised limited Boltzmann of top layer
A classification layer is added on the hidden layer of machine, forms the saturation state prediction model.
Optionally, described that the training sample is inputted into deepness belief network model, to the deepness belief network model
Unsupervised training is carried out, after obtaining the saturation state prediction model, further includes:
The training sample is inputted into the saturation state prediction model, obtains unsupervised saturation state prediction result;
According to the unsupervised saturation state prediction result and the saturation state of the corresponding roadbed of each training data
Determine unsupervised prediction error;
Judge whether the unsupervised prediction error is less than prediction error threshold, obtains the first judging result;
When first judging result indicates that the unsupervised prediction error is greater than the prediction error threshold, according to institute
It states unsupervised prediction error and error back propagation adjustment is carried out to the saturation state prediction model, described in after being finely tuned
Saturation state prediction model.
Optionally, described that saturation state prediction model progress error is reversely passed according to the unsupervised prediction error
Adjustment is broadcast, with the saturation state prediction model after being finely tuned, is specifically included:
It obtains verifying sample and test sample, the verifying sample includes multiple verifying samples pair, each verifying sample pair
Including one verifying roadbed saturation state of multiple verifying roadbed gamma bandpass filter group cepstrum coefficient vector sums, the test sample
Including multiple test samples pair, each test sample is to including multiple test roadbed gamma bandpass filter group cepstrum coefficient vector sums
One test roadbed saturation state;
Error back propagation adjustment is carried out to the saturation state prediction model according to the unsupervised prediction error, is obtained
Error after adjustment weight reversely adjusts prediction model;
Prediction model is reversely adjusted to the error using the verifying sample to be trained and cross validation, is verified
Error;
Prediction model is reversely adjusted to the error using the test sample to test, and obtains test error;
Judge whether to meet termination condition, obtain the second judging result, the termination condition are as follows: continuous multiple validation errors
Difference is less than the difference threshold of setting, and continuous multiple validation errors are respectively less than corresponding test error, wherein described to test
Demonstrate,prove the difference that error difference is two adjacent validation errors;
When the second judging result expression meet termination condition when, using the error reversely adjust prediction model as finally
Saturation state prediction model;
When the second judging result sufficient termination condition with thumb down, " by training sample input depth letter described in return
Network model is read, unsupervised training is carried out to the deepness belief network model, obtains the saturation state prediction model ".
Optionally, the target roadbed gamma bandpass filter group cepstrum coefficient includes: to method for determination of amount
Obtain the Coherent Noise in GPR Record of target roadbed;
Down-sampled, the down-sampled spy ground of acquisition is carried out according to the sample rate and sampling number of setting to the Coherent Noise in GPR Record
Radar data;
The down-sampled Coherent Noise in GPR Record is handled using gamma bandpass filter group cepstrum coefficient algorithm, is obtained every
The corresponding gamma bandpass filter group cepstrum coefficient vector of the down-sampled Coherent Noise in GPR Record in road.
A kind of determination system of roadbed saturation state, the determining system include:
Coefficient vector obtains module, for obtaining target roadbed gamma bandpass filter group cepstrum coefficient vector;
Saturation state prediction module contains for the target roadbed gamma bandpass filter group cepstrum coefficient vector to be input to
In water state prediction model, the saturation state of target roadbed is obtained;Wherein, the input of the saturation state prediction model is target
Roadbed gamma bandpass filter group cepstrum coefficient vector, the output of the saturation state prediction model are target roadbed containing watery
State;The saturation state prediction model is according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm
It establishes;The subsystem of establishing of the saturation state prediction model specifically includes:
Training sample obtains module, and for obtaining training sample, the training sample includes multiple training datas, the instruction
Practice the gamma bandpass filter group cepstrum coefficient vector that data are roadbed known to saturation state;
Unsupervised training module, for the training sample to be inputted deepness belief network model, to the depth conviction
Network model carries out unsupervised training, obtains the saturation state prediction model.
Optionally, the unsupervised training module specifically includes:
Limited Boltzmann machine training unit, for the training sample successively to be inputted each limited Boltzmann machine,
Unsupervised training is carried out to each limited Boltzmann machine, obtains corresponding unsupervised limited Boltzmann machine, the nothing
The lower layer for supervising limited Boltzmann machine is visual layers, and the upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Prediction model generation unit, for stacking each unsupervised limited Boltzmann machine, and in top layer
A classification layer is added on the hidden layer of the unsupervised limited Boltzmann machine, forms the saturation state prediction model.
Optionally, the determining system further include:
Training sample prediction module is obtained for the training sample to be inputted the saturation state prediction model without prison
Superintend and direct saturation state prediction result;
Unsupervised prediction error determination module, for according to the unsupervised saturation state prediction result and each instruction
The saturation state for practicing the corresponding roadbed of data determines unsupervised prediction error;
First judgment module obtains first for judging whether the unsupervised prediction error is less than prediction error threshold
Judging result;
Error back propagation module, for indicating the unsupervised prediction error greater than described when first judging result
When prediction error threshold, error back propagation tune is carried out to the saturation state prediction model according to the unsupervised prediction error
It is whole, with the saturation state prediction model after being finely tuned.
Optionally, the error back propagation module specifically includes:
Sample acquisition unit includes multiple verifying samples for acquisition verifying sample and test sample, the verifying sample
Right, each verifying sample is to aqueous including multiple verifying roadbed gamma bandpass filter group cepstrum coefficient one verifying roadbed of vector sum
State, the test sample include multiple test samples pair, and each test sample is to including multiple test roadbed gamma pass filters
One test roadbed saturation state of device group cepstrum coefficient vector sum;
Error back propagation unit, for being carried out according to the unsupervised prediction error to the saturation state prediction model
Error back propagation adjustment, the error after obtaining adjustment weight reversely adjust prediction model;
Validation error determination unit is instructed for reversely adjusting prediction model to the error using the verifying sample
Experienced and cross validation obtains validation error;
Test error determination unit is surveyed for reversely adjusting prediction model to the error using the test sample
Examination obtains test error;
Second judgment unit meets termination condition for judging whether, obtains the second judging result, the termination condition
Are as follows: continuous multiple validation error differences are less than the difference threshold of setting, and continuous multiple validation errors are respectively less than corresponding
Test error, wherein the validation error difference is the difference of two adjacent validation errors;
When the second judging result expression meet termination condition when, using the error reversely adjust prediction model as finally
Saturation state prediction model;
When the second judging result sufficient termination condition with thumb down, " by training sample input depth letter described in return
Network model is read, unsupervised training is carried out to the deepness belief network model, obtains the saturation state prediction model ".
Optionally, stator system includes: the target roadbed gamma bandpass filter group cepstrum coefficient vector really
Coherent Noise in GPR Record obtains module, for obtaining the Coherent Noise in GPR Record of target roadbed;
Down-sampled module is adopted for carrying out drop according to the sample rate and sampling number of setting to the Coherent Noise in GPR Record
Sample obtains down-sampled Coherent Noise in GPR Record;
Coefficient vector determining module, for using gamma bandpass filter group cepstrum coefficient algorithm to the down-sampled spy land mine
It is handled up to data, obtains the corresponding gamma bandpass filter group cepstrum coefficient vector of the down-sampled Coherent Noise in GPR Record of per pass.
The specific embodiment provided according to the present invention, the invention discloses following technical effects:
The present invention provides a kind of determination method of roadbed saturation state and determine system, the method for determination includes: acquisition mesh
Mark roadbed gamma bandpass filter group cepstrum coefficient vector;The target roadbed gamma bandpass filter group cepstrum coefficient vector is inputted
Into saturation state prediction model, the saturation state of target roadbed is obtained;Wherein, the saturation state prediction model is according to gal
Made of horse bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm are established, by the logical filter of the gamma of target roadbed
Wave device group cepstrum coefficient vector inputs saturation state prediction model, can automatically obtain the evaluation knot of the saturation state of target roadbed
Fruit.Compared with traditional manual sort's method, determining method provided by the invention and determining system are participated in without artificial, are disobeyed
Rely the artificial experience and subjective judgement for the treatment of people, it is therefore, not only high-efficient, but also can effectively avoid due to individual difference band
The error come, so as to accurately and reliably carry out saturation state detection to existing roadbed.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without creative efforts, can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the flow chart of the determination method for the roadbed saturation state that the embodiment of the present invention 1 provides;
Fig. 2 is the flow chart of the method for building up for the saturation state prediction model that the embodiment of the present invention 1 provides;
Fig. 3 is the flow chart that the present invention implements the error back propagation adjustment that 1 provides;
Fig. 4 is the structural block diagram of the determination system for the roadbed saturation state that the embodiment of the present invention 2 provides;
Fig. 5 is the flow chart of the determination method for the roadbed saturation state that the embodiment of the present invention 3 provides.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The object of the present invention is to provide a kind of determination method of roadbed saturation state and determine system, it can be accurately and efficiently
Determine the saturation state of roadbed.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
Fig. 1 is the flow chart of the determination method for the roadbed saturation state that the embodiment of the present invention 1 provides.As shown in Figure 1, a kind of
The determination method of roadbed saturation state, the determining method include:
Step 101: obtaining target roadbed gamma bandpass filter group cepstrum coefficient vector;
Step 102: the target roadbed gamma bandpass filter group cepstrum coefficient vector is input to saturation state prediction mould
In type, the saturation state of target roadbed is obtained;Wherein, the input of the saturation state prediction model is the logical filter of target roadbed gamma
Wave device group cepstrum coefficient vector, the output of the saturation state prediction model are the saturation state of target roadbed;It is described containing watery
State prediction model is established according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm.
Fig. 2 is the flow chart of the method for building up for the saturation state prediction model that the embodiment of the present invention 1 provides.As shown in Fig. 2,
The method for building up of the saturation state prediction model specifically includes:
Step 201: obtaining training sample, the training sample includes multiple training datas, and the training data is aqueous
The gamma bandpass filter group cepstrum coefficient vector of roadbed known to state;
Step 202: the training sample being inputted into deepness belief network model, the deepness belief network model is carried out
Unsupervised training obtains the saturation state prediction model;
Step 203: the training sample being inputted into the saturation state prediction model, obtains unsupervised saturation state prediction
As a result;
Step 204: according to the unsupervised saturation state prediction result and the corresponding roadbed of each training data
Saturation state determines unsupervised prediction error;
Step 205: judging whether the unsupervised prediction error is less than prediction error threshold, obtain the first judging result;
If so, executing step 206;
Otherwise, step 207 is executed;
Step 206: exporting the saturation state prediction model;
Step 207: error back propagation is carried out to the saturation state prediction model according to the unsupervised prediction error
Adjustment, with the saturation state prediction model after being finely tuned.
Specifically, step 202: the training sample being inputted into deepness belief network model, to the deepness belief network
Model carries out unsupervised training, obtains the saturation state prediction model, specifically includes:
The training sample is successively inputted into each limited Boltzmann machine, each limited Boltzmann machine is carried out
Unsupervised training obtains corresponding unsupervised limited Boltzmann machine, and the lower layer of the unsupervised limited Boltzmann machine is can
Depending on layer, the upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Each unsupervised limited Boltzmann machine is stacked, and in the unsupervised limited Boltzmann of top layer
A classification layer is added on the hidden layer of machine, forms the saturation state prediction model.
Fig. 3 is the flow chart that the present invention implements the error back propagation adjustment that 1 provides.As shown in figure 3, step 206: according to
The unsupervised prediction error carries out error back propagation adjustment to the saturation state prediction model, with the institute after being finely tuned
Saturation state prediction model is stated, is specifically included:
Step 301: obtaining verifying sample and test sample, the verifying sample include multiple verifying samples pair, each test
Demonstrate,prove sample to include one verifying roadbed saturation state of multiple verifying roadbed gamma bandpass filter group cepstrum coefficient vector sums, it is described
Test sample includes multiple test samples pair, and each test sample is to including multiple test roadbed gamma bandpass filter group cepstrum systems
Number vector and a test roadbed saturation state;
Step 302: error back propagation is carried out to the saturation state prediction model according to the unsupervised prediction error
Adjustment, the error after obtaining adjustment weight reversely adjust prediction model;
Step 303: prediction model is reversely adjusted to the error using the verifying sample and is trained and cross validation,
Obtain validation error;
Step 304: prediction model reversely being adjusted to the error using the test sample and is tested, test is obtained and misses
Difference;
Step 305: judging whether to meet termination condition, obtain the second judging result, the termination condition are as follows: continuous multiple
Validation error difference is less than the difference threshold of setting, and continuous multiple validation errors are respectively less than corresponding test error,
In, the validation error difference is the difference of two adjacent validation errors;
If so, executing step 306: the error reversely being adjusted prediction model as final saturation state and predicts mould
Type;
Otherwise, it returns to the step 202: the training sample being inputted into deepness belief network model, the depth is believed
It reads network model and carries out unsupervised training, obtain the saturation state prediction model.
Embodiment 2:
Fig. 4 is the structural block diagram of the determination system for the roadbed saturation state that the embodiment of the present invention 2 provides.As shown in figure 4, one
The determination system of kind roadbed saturation state, the determining system include:
Coefficient vector obtains module 401, for obtaining target roadbed gamma bandpass filter group cepstrum coefficient vector;
Saturation state prediction module 402, for inputting the target roadbed gamma bandpass filter group cepstrum coefficient vector
Into saturation state prediction model, the saturation state of target roadbed is obtained;Wherein, the input of the saturation state prediction model is
Target roadbed gamma bandpass filter group cepstrum coefficient vector, the output of the saturation state prediction model are the aqueous of target roadbed
State.The saturation state prediction model is calculated according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model
Method is established;The subsystem of establishing of the saturation state prediction model specifically includes:
Training sample obtains module 403, and for obtaining training sample, the training sample includes multiple training datas, institute
State the gamma bandpass filter group cepstrum coefficient vector that training data is roadbed known to saturation state;
Unsupervised training module 404 believes the depth for the training sample to be inputted deepness belief network model
It reads network model and carries out unsupervised training, obtain the saturation state prediction model;
Training sample prediction module 405 obtains nothing for the training sample to be inputted the saturation state prediction model
Supervise saturation state prediction result;
Unsupervised prediction error determination module 406, for according to the unsupervised saturation state prediction result and each institute
The saturation state for stating the corresponding roadbed of training data determines unsupervised prediction error;
First judgment module 407 obtains for judging whether the unsupervised prediction error is less than prediction error threshold
One judging result;
Error back propagation module 408, for indicating that the unsupervised prediction error is greater than when first judging result
When the prediction error threshold, error is carried out to the saturation state prediction model according to the unsupervised prediction error and is reversely passed
Adjustment is broadcast, with the saturation state prediction model after being finely tuned.
Specifically, the unsupervised training module 404 specifically includes:
Limited Boltzmann machine training unit, for the training sample successively to be inputted each limited Boltzmann machine,
Unsupervised training is carried out to each limited Boltzmann machine, obtains corresponding unsupervised limited Boltzmann machine, the nothing
The lower layer for supervising limited Boltzmann machine is visual layers, and the upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Prediction model generation unit, for stacking each unsupervised limited Boltzmann machine, and in top layer
A classification layer is added on the hidden layer of the unsupervised limited Boltzmann machine, forms the saturation state prediction model.
Specifically, the error back propagation module 401 specifically includes:
Sample acquisition unit includes multiple verifying samples for acquisition verifying sample and test sample, the verifying sample
Right, each verifying sample is to aqueous including multiple verifying roadbed gamma bandpass filter group cepstrum coefficient one verifying roadbed of vector sum
State, the test sample include multiple test samples pair, and each test sample is to including multiple test roadbed gamma pass filters
One test roadbed saturation state of device group cepstrum coefficient vector sum;
Error back propagation unit, for being carried out according to the unsupervised prediction error to the saturation state prediction model
Error back propagation adjustment, the error after obtaining adjustment weight reversely adjust prediction model;
Validation error determination unit is instructed for reversely adjusting prediction model to the error using the verifying sample
Experienced and cross validation obtains validation error;
Test error determination unit is surveyed for reversely adjusting prediction model to the error using the test sample
Examination obtains test error;
Second judgment unit meets termination condition for judging whether, obtains the second judging result, the termination condition
Are as follows: continuous multiple validation error differences are less than the difference threshold of setting, and continuous multiple validation errors are respectively less than corresponding
Test error, wherein the validation error difference is the difference of two adjacent validation errors;
When the second judging result expression meet termination condition when, using the error reversely adjust prediction model as finally
Saturation state prediction model;
When the second judging result sufficient termination condition with thumb down, " by training sample input depth letter described in return
Network model is read, unsupervised training is carried out to the deepness belief network model, obtains the saturation state prediction model ".
In the present embodiment, stator system includes: the target roadbed gamma bandpass filter group cepstrum coefficient vector really
Coherent Noise in GPR Record obtains module, for obtaining the Coherent Noise in GPR Record of target roadbed;
Down-sampled module is adopted for carrying out drop according to the sample rate and sampling number of setting to the Coherent Noise in GPR Record
Sample obtains down-sampled Coherent Noise in GPR Record;
Coefficient vector determining module, for using gamma bandpass filter group cepstrum coefficient algorithm to the down-sampled spy land mine
It is handled up to data, obtains the corresponding gamma bandpass filter group cepstrum coefficient vector of the down-sampled Coherent Noise in GPR Record of per pass.
Embodiment 3:
Fig. 5 is the flow chart of the determination method for the roadbed saturation state that the embodiment of the present invention 3 provides.As shown in figure 5, roadbed
The specific implementation process of the determination method of saturation state the following steps are included:
Step 1: the original Coherent Noise in GPR Record of roadbed known to saturation state is obtained, and to original Coherent Noise in GPR Record
It is carried out according to fixed sample rate and sampling number down-sampled.
Step 2: according to known roadbed saturation state when detection, treated that Coherent Noise in GPR Record is divided to down-sampled
Class label information supplement, according to moisture content less than 15%, greater than 15% less than 25%, greater than 25% less than 35% and greater than 35%
It is divided into tetra- kinds of classifications of I, II, III, IV.
Step 3: lead to (γ-Tone) filter group cepstrum coefficient (GFCC) algorithm to down-sampled spy land mine using gamma
It is handled up to data, obtains the corresponding GFCC coefficient vector of the down-sampled Coherent Noise in GPR Record of per pass.
Single track Coherent Noise in GPR Record after will be down-sampled is expressed as column vector Expression one is single-row
Matrix, n indicate down-sampled after single track points,In element pnSingle track Coherent Noise in GPR Record after indicating down-sampled it is every
The value of a sampled point asks the process of γ-Tone filter group cepstrum coefficient as follows every track data:
γ-Tone filter is expressed as form shown in formula (1) first:
G (f, t)=kta-1e-2πbtcos(2πft+φ) (1)
Wherein, g (f, t) indicates the time domain response of single γ-Tone filter, and t >=0, t indicate sampling time sequence, f table
Show the centre frequency of filter, k indicates that filter gain, a indicate that filter order, φ indicate phase, and e indicates natural constant, b
Indicate that decay factor, decay factor determine the bandwidth of corresponding filter, it is shown with the relationship such as formula (2) of centre frequency f:
The response of γ-Tone filter group is formed by stacking by the response of multichannel γ-Tone filter.To apply multichannel
γ-Tone filter needs first to carry out framing to data, i.e., by single track Coherent Noise in GPR RecordIt is cut into several shorter single frames
DataThe framing quantity of M expression data.
Shown in response such as formula (3) of each γ-Tone filter to each frame data in filter group:
Wherein, symbol " * " indicates that convolution algorithm symbol, i=0,1 ..., N-1, N indicate the number of channels of filter group, Gm
(i) response of i-th of γ-Tone filter to m-th of frame data in filter group, t are indicatedmIndicate m-th of frame data
Corresponding sampling time sequence, tm>=0,Indicate i-th of frame data, fiIt indicates in filter group in i-th of filter
Frequency of heart, expression formula are as follows:
Wherein, fHIndicate the cutoff frequency of filter;viIndicate filter overlap factor, for specify adjacent filter it
Between overlapping percentages, and i=0,1 ..., N-1.
Thus calculated each γ-Tone filter constitutes a matrix to the response of each frame data, indicates
The Coherent Noise in GPR Record of input is right in the changes in distribution of defined specific frequency domain, referred to as γ-Tone filter (GF) characteristic coefficient
Each GF characteristic coefficient carries out discrete cosine transform (DCT) and is overlapped the response of the filter in each channel:
Wherein, j=1,2 ..., N-1, CjIndicate the cepstrum coefficient (GFCC) of j-th of γ-Tone filter.
Strictly speaking this characteristic coefficient CjIt is not cepstrum coefficient, because the generation of cepstrum coefficient will take logarithmic energy,
But energy is not taken herein.However this by GFCC as cepstrum coefficient, be GFCC and Meier due in above-mentioned calculating process
Frequency cepstral coefficient (MFCC) has a similar characteristic parameter, and the extraction conversion of characteristic parameter has functionally similar
Property.
Step 4: data set is divided into training sample, verifying sample and test sample.
The training sample includes multiple training samples pair, and each training sample is to including the logical filter of multiple trained roadbed gammas
Vector sum one trained roadbed saturation state of wave device group cepstrum coefficient.The verifying sample includes multiple verifying samples pair, each
Sample is verified to including one verifying roadbed saturation state of multiple verifying roadbed gamma bandpass filter group cepstrum coefficient vector sums.Institute
Stating test sample includes multiple test samples pair, and each test sample is to including multiple test roadbed gamma bandpass filter group cepstrums
Coefficient vector and a test roadbed saturation state.
Step 5: multiple trained roadbed gamma bandpass filter group cepstrum coefficient vectors are successively inputted into each limited Bohr hereby
Graceful machine carries out unsupervised training to each limited Boltzmann machine, obtains corresponding unsupervised limited Boltzmann machine, institute
The lower layer for stating unsupervised limited Boltzmann machine is visual layers, and the upper layer of the unsupervised limited Boltzmann machine is hidden layer.
Each unsupervised limited Boltzmann machine is stacked, and hiding in the unsupervised limited Boltzmann machine of top layer
A classification layer is added on layer, regard the deepness belief network model (DBN) of formation as saturation state prediction model.Wherein, by
It limits and is connected entirely between the visual node layer and hiding node layer of Boltzmann machine, but visual layers and hiding neuron section interior layer by layer
Point is connectionless.
The classification layer corresponds to the saturation state of roadbed, the hidden layer (the last one hidden layer) adjacent with the classification layer
A feedforward neural network is formed with the classification layer.The visual layers of undermost unsupervised limited Boltzmann machine are containing watery
The neuron node number of the input layer of state prediction model, input layer is consistent with the dimension of each sample data.First layer is hidden
The neuron node number for hiding layer is less than the neuron node number of input layer, and hidden layer neuron number of nodes backward successively compares
Preceding layer is few, and one added on the last layer hidden layer layer is the classification layer for characterizing four characteristic of division.
Step 6: establishing the objective function of deepness belief network, and net is optimized and updated by stochastic gradient descent
The parameter of network.
Multiple trained roadbed gamma bandpass filter group cepstrum coefficient vectors are inputted into the saturation state prediction model, are obtained
Unsupervised saturation state prediction result.By unsupervised saturation state prediction result compared with known trained roadbed saturation state,
The error between unsupervised learning output and desired output is calculated, i.e. saturation state determines unsupervised prediction error.
When saturation state determines that unsupervised prediction error is greater than prediction error threshold, according to the unsupervised prediction error
Error back propagation adjustment is carried out to the saturation state prediction model, i.e., is missed using the feedforward neural network of the last layer
The weight of poor backpropagation fine tuning whole network, addition verifying sample and test sample in trim process, when this complete batch of training
A cross validation and test are carried out after amount data.It remains unchanged for a long period of time when validation error tends to certain value, and validation error is lower than test
When error, stop network training.I.e. during loop iteration training network model, continuous multiple validation error differences are less than
The difference threshold of setting, and when continuous multiple validation errors are respectively less than corresponding test error, illustrate to meet termination condition,
It can stop network training, obtain final saturation state prediction model.Wherein, the validation error difference be it is adjacent twice
The difference of two validation errors obtained in training process.
Step 7: carrying out the original Coherent Noise in GPR Record of target roadbed to be downsampled to particular sample rate and sampling number,
GFCC coefficient vector, which is calculated, by γ-Tone filter group cepstrum coefficient algorithm and is input to final saturation state predicts mould
In type, the output of saturation state prediction model is the saturation state classification results of target roadbed.
Saturation state identification is carried out using one group data of the DBN model provided by the invention to a certain existing roadbed, is utilized
The accuracy rate that DBN predicts roadbed Gpr Data saturation state is more than 90%, shows that method provided by the invention can
With the prediction for roadbed radar cross-section saturation state, the time-consuming for carrying out classification prediction using DBN is very short, shows offer of the present invention
Method can be applied to detection it is real-time handle in.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For system disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention;At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (10)
1. a kind of determination method of roadbed saturation state, which is characterized in that the determining method includes:
Obtain target roadbed gamma bandpass filter group cepstrum coefficient vector;
The target roadbed gamma bandpass filter group cepstrum coefficient vector is input in saturation state prediction model, target is obtained
The saturation state of roadbed;Wherein, the input of the saturation state prediction model is target roadbed gamma bandpass filter group cepstrum system
Number vector, the output of the saturation state prediction model are the saturation state of target roadbed;The saturation state prediction model is
It is established according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm;The saturation state prediction
The method for building up of model specifically includes:
Training sample is obtained, the training sample includes multiple training datas, and the training data is road known to saturation state
The gamma bandpass filter group cepstrum coefficient vector of base;
The training sample is inputted into deepness belief network model, unsupervised training is carried out to the deepness belief network model,
Obtain the saturation state prediction model.
2. determining method according to claim 1, which is characterized in that described that the training sample is inputted depth conviction net
Network model carries out unsupervised training to the deepness belief network model, obtains the saturation state prediction model, specific to wrap
It includes:
The training sample is successively inputted into each limited Boltzmann machine, each limited Boltzmann machine is carried out without prison
Supervising and instructing white silk, obtains corresponding unsupervised limited Boltzmann machine, the lower layer of the unsupervised limited Boltzmann machine is visual layers,
The upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Each unsupervised limited Boltzmann machine is stacked, and in the unsupervised limited Boltzmann machine of top layer
A classification layer is added on hidden layer, forms the saturation state prediction model.
3. determining method according to claim 1, which is characterized in that described that the training sample is inputted depth conviction net
Network model carries out unsupervised training to the deepness belief network model and also wraps after obtaining the saturation state prediction model
It includes:
The training sample is inputted into the saturation state prediction model, obtains unsupervised saturation state prediction result;
It is determined according to the saturation state of the unsupervised saturation state prediction result and the corresponding roadbed of each training data
Unsupervised prediction error;
Judge whether the unsupervised prediction error is less than prediction error threshold, obtains the first judging result;
When first judging result indicates that the unsupervised prediction error is greater than the prediction error threshold, according to the nothing
Monitoring forecast error carries out error back propagation adjustment to the saturation state prediction model, with described aqueous after being finely tuned
State Forecasting Model.
4. determining method according to claim 3, which is characterized in that it is described according to the unsupervised prediction error to described
Saturation state prediction model carries out error back propagation adjustment, with the saturation state prediction model after being finely tuned, specifically
Include:
Obtain verifying sample and test sample, the verifying sample includes multiple verifying samples pair, and each verifying sample is to including
One verifying roadbed saturation state of multiple verifying roadbed gamma bandpass filter group cepstrum coefficient vector sums, the test sample include
Multiple test samples pair, each test sample is to including multiple test roadbed gammas bandpass filter group cepstrum coefficient vector sum one
Test roadbed saturation state;
Error back propagation adjustment is carried out to the saturation state prediction model according to the unsupervised prediction error, is adjusted
Error after weight reversely adjusts prediction model;
Prediction model is reversely adjusted to the error using the verifying sample to be trained and cross validation, is obtained verifying and is missed
Difference;
Prediction model is reversely adjusted to the error using the test sample to test, and obtains test error;
Judge whether to meet termination condition, obtain the second judging result, the termination condition are as follows: continuous multiple validation error differences
Less than the difference threshold of setting, and continuous multiple validation errors are respectively less than corresponding test error, wherein the verifying misses
Difference is the difference of two adjacent validation errors;
When the expression of the second judging result meets termination condition, the error is reversely adjusted into prediction model as finally aqueous
State Forecasting Model;
When the second judging result sufficient termination condition with thumb down, " training sample is inputted into depth conviction net described in return
Network model carries out unsupervised training to the deepness belief network model, obtains the saturation state prediction model ".
5. determining method according to claim 1, which is characterized in that target roadbed gamma bandpass filter group cepstrum system
The determination method of number vector includes:
Obtain the Coherent Noise in GPR Record of target roadbed;
It is down-sampled according to sample rate and the sampling number progress of setting to the Coherent Noise in GPR Record, obtain down-sampled Ground Penetrating Radar
Data;
The down-sampled Coherent Noise in GPR Record is handled using gamma bandpass filter group cepstrum coefficient algorithm, obtains per pass drop
Sample the corresponding gamma bandpass filter group cepstrum coefficient vector of Coherent Noise in GPR Record.
6. a kind of determination system of roadbed saturation state, which is characterized in that the determining system includes:
Coefficient vector obtains module, for obtaining target roadbed gamma bandpass filter group cepstrum coefficient vector;
Saturation state prediction module, for being input to the target roadbed gamma bandpass filter group cepstrum coefficient vector containing watery
In state prediction model, the saturation state of target roadbed is obtained;Wherein, the input of the saturation state prediction model is target roadbed
Gamma bandpass filter group cepstrum coefficient vector, the output of the saturation state prediction model are the saturation state of target roadbed;Institute
Stating saturation state prediction model is established according to gamma bandpass filter group cepstrum coefficient algorithm and deepness belief network model algorithm
It forms;The subsystem of establishing of the saturation state prediction model specifically includes:
Training sample obtains module, and for obtaining training sample, the training sample includes multiple training datas, the trained number
According to the gamma bandpass filter group cepstrum coefficient vector for roadbed known to saturation state;
Unsupervised training module, for the training sample to be inputted deepness belief network model, to the deepness belief network
Model carries out unsupervised training, obtains the saturation state prediction model.
7. determining system according to claim 6, which is characterized in that the unsupervised training module specifically includes:
Limited Boltzmann machine training unit, for the training sample successively to be inputted each limited Boltzmann machine, to each
A limited Boltzmann machine carries out unsupervised training, obtains corresponding unsupervised limited Boltzmann machine, described unsupervised
The lower layer of limited Boltzmann machine is visual layers, and the upper layer of the unsupervised limited Boltzmann machine is hidden layer;
Prediction model generation unit, for will each unsupervised limited Boltzmann machine stacking, and described in the top layer
A classification layer is added on the hidden layer of unsupervised limited Boltzmann machine, forms the saturation state prediction model.
8. determining system according to claim 6, which is characterized in that the determining system further include:
Training sample prediction module obtains unsupervised contain for the training sample to be inputted the saturation state prediction model
Water state prediction result;
Unsupervised prediction error determination module, for according to the unsupervised saturation state prediction result and each trained number
Unsupervised prediction error is determined according to the saturation state of corresponding roadbed;
First judgment module obtains the first judgement for judging whether the unsupervised prediction error is less than prediction error threshold
As a result;
Error back propagation module, for indicating that the unsupervised prediction error is greater than the prediction when first judging result
When error threshold, error back propagation adjustment is carried out to the saturation state prediction model according to the unsupervised prediction error,
With the saturation state prediction model after being finely tuned.
9. determining system according to claim 8, which is characterized in that the error back propagation module specifically includes:
Sample acquisition unit includes multiple verifying samples pair for acquisition verifying sample and test sample, the verifying sample, often
A verifying sample to include one verifying roadbed saturation state of multiple verifying roadbed gamma bandpass filter group cepstrum coefficient vector sums,
The test sample includes multiple test samples pair, and each test sample is fallen to including multiple test roadbed gamma bandpass filter groups
One test roadbed saturation state of spectral coefficient vector sum;
Error back propagation unit, for carrying out error to the saturation state prediction model according to the unsupervised prediction error
Backpropagation adjustment, the error after obtaining adjustment weight reversely adjust prediction model;
Validation error determination unit, for using the verifying sample to the error reversely adjust prediction model be trained and
Cross validation obtains validation error;
Test error determination unit is tested for reversely adjusting prediction model to the error using the test sample,
Obtain test error;
Second judgment unit meets termination condition for judging whether, obtains the second judging result, the termination condition are as follows: even
Continue the difference threshold that multiple validation error differences are less than setting, and continuous multiple validation errors are respectively less than corresponding test and miss
Difference, wherein the validation error difference is the difference of two adjacent validation errors;
When the expression of the second judging result meets termination condition, the error is reversely adjusted into prediction model as finally aqueous
State Forecasting Model;
When the second judging result sufficient termination condition with thumb down, " training sample is inputted into depth conviction net described in return
Network model carries out unsupervised training to the deepness belief network model, obtains the saturation state prediction model ".
10. determining system according to claim 6, which is characterized in that the target roadbed gamma bandpass filter group cepstrum
Really stator system includes: coefficient vector
Coherent Noise in GPR Record obtains module, for obtaining the Coherent Noise in GPR Record of target roadbed;
Down-sampled module, it is down-sampled for being carried out to the Coherent Noise in GPR Record according to the sample rate and sampling number of setting, it obtains
Obtain down-sampled Coherent Noise in GPR Record;
Coefficient vector determining module, for using gamma bandpass filter group cepstrum coefficient algorithm to the down-sampled Ground Penetrating Radar number
According to being handled, the corresponding gamma bandpass filter group cepstrum coefficient vector of the down-sampled Coherent Noise in GPR Record of per pass is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810844761.3A CN109034246B (en) | 2018-07-27 | 2018-07-27 | Method and system for determining roadbed water-containing state |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810844761.3A CN109034246B (en) | 2018-07-27 | 2018-07-27 | Method and system for determining roadbed water-containing state |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109034246A true CN109034246A (en) | 2018-12-18 |
CN109034246B CN109034246B (en) | 2021-04-16 |
Family
ID=64646324
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810844761.3A Active CN109034246B (en) | 2018-07-27 | 2018-07-27 | Method and system for determining roadbed water-containing state |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109034246B (en) |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060078383A1 (en) * | 2004-09-16 | 2006-04-13 | Novak John F | Continuous method and apparatus for microwave-based dryer |
CN103745002A (en) * | 2014-01-24 | 2014-04-23 | 中国科学院信息工程研究所 | Method and system for recognizing hidden paid posters on basis of fusion of behavior characteristic and content characteristic |
CN105702250A (en) * | 2016-01-06 | 2016-06-22 | 福建天晴数码有限公司 | Voice recognition method and device |
CN106096005A (en) * | 2016-06-23 | 2016-11-09 | 康佳集团股份有限公司 | A kind of rubbish mail filtering method based on degree of depth study and system |
CN106328121A (en) * | 2016-08-30 | 2017-01-11 | 南京理工大学 | Chinese traditional musical instrument classification method based on depth confidence network |
CN106405640A (en) * | 2016-08-26 | 2017-02-15 | 中国矿业大学(北京) | Automatic microseismic signal arrival time picking method based on depth belief neural network |
US20170042504A1 (en) * | 2015-08-12 | 2017-02-16 | Sonetics Ultrasound, Inc. | Method and system for measuring pressure using ultrasound |
CN106448684A (en) * | 2016-11-16 | 2017-02-22 | 北京大学深圳研究生院 | Deep-belief-network-characteristic-vector-based channel-robust voiceprint recognition system |
CN106898362A (en) * | 2017-02-23 | 2017-06-27 | 重庆邮电大学 | The Speech Feature Extraction of Mel wave filters is improved based on core principle component analysis |
CN106952644A (en) * | 2017-02-24 | 2017-07-14 | 华南理工大学 | A kind of complex audio segmentation clustering method based on bottleneck characteristic |
CN107229787A (en) * | 2017-05-24 | 2017-10-03 | 南京航空航天大学 | A kind of gamma-ray spectrum analysis method based on approximation coefficient and deep learning |
CN107527067A (en) * | 2017-08-01 | 2017-12-29 | 中国铁道科学研究院铁道建筑研究所 | A kind of Railway Roadbed intelligent identification Method based on GPR |
-
2018
- 2018-07-27 CN CN201810844761.3A patent/CN109034246B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060078383A1 (en) * | 2004-09-16 | 2006-04-13 | Novak John F | Continuous method and apparatus for microwave-based dryer |
CN103745002A (en) * | 2014-01-24 | 2014-04-23 | 中国科学院信息工程研究所 | Method and system for recognizing hidden paid posters on basis of fusion of behavior characteristic and content characteristic |
US20170042504A1 (en) * | 2015-08-12 | 2017-02-16 | Sonetics Ultrasound, Inc. | Method and system for measuring pressure using ultrasound |
CN105702250A (en) * | 2016-01-06 | 2016-06-22 | 福建天晴数码有限公司 | Voice recognition method and device |
CN106096005A (en) * | 2016-06-23 | 2016-11-09 | 康佳集团股份有限公司 | A kind of rubbish mail filtering method based on degree of depth study and system |
CN106405640A (en) * | 2016-08-26 | 2017-02-15 | 中国矿业大学(北京) | Automatic microseismic signal arrival time picking method based on depth belief neural network |
CN106328121A (en) * | 2016-08-30 | 2017-01-11 | 南京理工大学 | Chinese traditional musical instrument classification method based on depth confidence network |
CN106448684A (en) * | 2016-11-16 | 2017-02-22 | 北京大学深圳研究生院 | Deep-belief-network-characteristic-vector-based channel-robust voiceprint recognition system |
CN106898362A (en) * | 2017-02-23 | 2017-06-27 | 重庆邮电大学 | The Speech Feature Extraction of Mel wave filters is improved based on core principle component analysis |
CN106952644A (en) * | 2017-02-24 | 2017-07-14 | 华南理工大学 | A kind of complex audio segmentation clustering method based on bottleneck characteristic |
CN107229787A (en) * | 2017-05-24 | 2017-10-03 | 南京航空航天大学 | A kind of gamma-ray spectrum analysis method based on approximation coefficient and deep learning |
CN107527067A (en) * | 2017-08-01 | 2017-12-29 | 中国铁道科学研究院铁道建筑研究所 | A kind of Railway Roadbed intelligent identification Method based on GPR |
Non-Patent Citations (3)
Title |
---|
LANYUE ZHANG 等: "Feature Extraction of Underwater Target Signal Using Mel Frequency Cepstrum Coefficients Based on Acoustic Vector Sensor", 《RESEARCH ARTICLE》 * |
刘杰: "铁路路基含水状态的探地雷达检测方法研究", 《中国博士学位论文全文数据库工程科技Ⅱ辑》 * |
陈喆 等: "非平稳信号度量方法综述", 《JOURNAL OF DATA ACQUISITION AND PROCESSING》 * |
Also Published As
Publication number | Publication date |
---|---|
CN109034246B (en) | 2021-04-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110533631B (en) | SAR image change detection method based on pyramid pooling twin network | |
Langer et al. | Automatic classification and a-posteriori analysis of seismic event identification at Soufrière Hills volcano, Montserrat | |
CN110097755A (en) | Freeway traffic flow amount state identification method based on deep neural network | |
CN105863628B (en) | A kind of phase of development subtle hydrocarbon reservoir method | |
CN111967712B (en) | Traffic risk prediction method based on complex network theory | |
CN107783185A (en) | A kind of processing method and processing device of tomographic statics | |
Oldoni et al. | Context-dependent environmental sound monitoring using SOM coupled with LEGION | |
CN114220458B (en) | Voice recognition method and device based on array hydrophone | |
CN112362756A (en) | Concrete structure damage monitoring method and system based on deep learning | |
CN112685950A (en) | Method, system and equipment for detecting abnormality of ocean time sequence observation data | |
Liu et al. | Microseismic event detection and classification based on convolutional neural network | |
CN115565019A (en) | Single-channel high-resolution SAR image ground object classification method based on deep self-supervision generation countermeasure | |
CN109669017A (en) | Refinery's distillation tower top based on deep learning cuts water concentration prediction technique | |
CN110212975A (en) | A kind of OTDR fault signature judgment method based on differential evolution neural network | |
CN109034246A (en) | A kind of the determination method and determining system of roadbed saturation state | |
CN116884435A (en) | Voice event detection method and device based on audio prompt learning | |
CN116753471A (en) | Water supply pipeline leakage multi-domain feature extraction and fusion identification method | |
CN116777046A (en) | Traffic pre-training model construction and traffic prediction method and device and electronic equipment | |
CN110118994A (en) | A kind of nonmarine source rock quantitative forecasting technique based on seismic inversion and machine learning | |
CN115220098A (en) | Automatic recognition method and device for broken and crack-controlled carbonatite hole body | |
CN113050191A (en) | Shale oil TOC prediction method and device based on double parameters | |
CN111949945A (en) | Method, device, equipment and system for determining total organic carbon content of hydrocarbon source rock | |
Jiang et al. | Automatic phase picking from microseismic recordings using feature extraction and neural network | |
Li et al. | Research on voiceprint recognition technology based on deep neural network | |
CN112396161B (en) | Lithologic profile construction method, system and equipment based on convolutional 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 |