CN110083637A - A kind of denoising method towards bridge defect ratings data - Google Patents
A kind of denoising method towards bridge defect ratings data Download PDFInfo
- Publication number
- CN110083637A CN110083637A CN201910327313.0A CN201910327313A CN110083637A CN 110083637 A CN110083637 A CN 110083637A CN 201910327313 A CN201910327313 A CN 201910327313A CN 110083637 A CN110083637 A CN 110083637A
- Authority
- CN
- China
- Prior art keywords
- sample
- conflict
- data set
- data
- bridge defect
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The present invention relates to a kind of denoising methods towards bridge defect ratings data, it is characterised in that: deletion initial data first concentrates the feature for being difficult to differentiate between out characteristic value orbution;Secondly the sample that data concentrate all labels different is compared two-by-two respectively, obtains a conflict being made of the sample of label collision to set;Then the number occurred in set is ranked up from high to low in conflict according to sample, then successively before calculated for rank the sample of t% silhouette coefficient, it is less than the sample of ε to deletion silhouette coefficient in set and data set from conflict, then the number of middle appearance is ranked up in conflict again depending on sample, it calculates silhouette coefficient and carries out delete operation, the silhouette coefficient of the sample of t% is not less than ε before ranking;The data set for eliminating noise is finally obtained, which can effectively improve the accuracy of bridge defect data staging.
Description
Technical field
The invention belongs to technical field of data processing, it is intended to design a kind of denoising side towards bridge defect ratings data
Method.
Background technique
Since reform and opening-up, highway in China bridge welcomes big construction great development period.Currently, highway in China bridge sum connects
Nearly 800,000, bridge quantity and scale occupy first of the world.However, the servicing bridges that China steps into the maintenance interval are increasing.
According to incompletely statistics, China's servicing bridges about 40% be on active service more than 20 years, industrial grade three, four classes bridge in spite of illness be up to
30%, or even having more than 100,000 bridge blocks is unsafe bridge, security risk can not be ignored.Bridge defect situation grading be road and bridge management and
The basis of maintenance.Traditional artificial ranking method not only takes time and effort, but also accuracy is not high, and there is an urgent need to use machine learning
Technology carries out automatic measure grading to servicing bridges disease.A large amount of label noise data is usually contained in existing highway bridge data set,
In order to effectively improve the estimated performance that machine learning method carries out bridge defect grading, need to filter out the mark of initial data concentration
Sign noise data.There are two types of forms for the label noise filtering method of mainstream at present: (1) adapting sorting algorithm directly to reduce label
Influence of the noise to algorithm performance;(2) classification ballot is carried out to data using pre-classifier, then filters out the doubtful noise in part
Data.However, above two method is on filtering bridge defect label noise data, the effect is unsatisfactory.
In conclusion this crossing domain needs to design the new label noise filtering method of one kind to solve the above problems.
Summary of the invention
In view of this, effectively improving and being based on the invention discloses a kind of denoising method of bridge defect ratings data
The estimated performance of the bridge defect ranking method of machine learning.The first, feature is difficult to differentiate between out by deleting initial data concentration
The feature of value orbution obtains new data set, the characteristic value of each feature all orderliness relationships in the data set, wherein institute
State the essential information that initial data concentration includes each bridge, the bridge defect information of various species and corresponding bridge defect
Grade label;The second, it concentrates all samples to be compared two-by-two data, and mutually will form a punching by conflicting two samples
Prominent pair, and conflict all in data set conflicts to being configured to one to set;Third, statistics conflict go out sample in set
Occurrence number, and the frequency of occurrence of sample is ranked up;4th, according to sample silhouette coefficient and the sample frequency successively by height to
It is low to weed out a certain proportion of sample and obtain new data set;Finally, using stacking method respectively to raw data set and
New data set, which is trained, obtains two models, and carries out assessment verifying to the bridge defect grade forecast performance of two models,
To verify the validity of this denoising method, if confirming the validity, the data set of a relative clean has just been obtained.
Technical solution of the present invention way of realization are as follows: a kind of denoising method towards bridge defect ratings data, it is logical first
The comparison two-by-two for crossing sample obtains conflict to set, is then being conflicted according to sample to the number occurred in set, in conjunction with sample
Silhouette coefficient carry out noise data rejecting, obtain filtered data set, then distinguished using same machine learning method
Model is trained on raw data set and filtered new data set, finally compares the estimated performance of two models, specifically
Step are as follows:
S1, the data that initial data is concentrated are carried out to pretreatment acquisition data set W1, by W1The spy of middle no ordering relation
Sign is removed to obtain new data set W2;
S2, it is based on data set W2, according to feature aiCharacteristic value aI, jSample with different labels is compared two-by-two,
Construction conflict is to ci;
S3, conflict is based on to ciConstruction conflict set C={ c1, c2..., cN, wherein N is rushing of including in conflict set C
It dashes forward to sum;
S4, pass through sample s in statistics conflict set CkThe frequency f of appearancek, obtain dictionary D={ sk: fk}。
S5, the sample in dictionary D is ranked up from high to low by the frequency;
S6, for the sample of t% preceding after sequence, in data set W2Middle calculating silhouette coefficient s (k) deletes s (k) less than ε's
Sample sk, obtain filtered new data set W3, delete at the same time conflict in set C comprising doubtful noisy samples skRush
Prominent pair;
S7, S4, S5, S6 are repeated, until being less than the sample of ε in step S62 without s (i);
S8, using same machines learning method, be based on data set W1And W3Model M is respectively trained out1And M3, assess and compare
Compared with model M3Estimated performance.
Further, step S1 includes:
S11, it is based on data set W1, using calorie fill method, characteristic value is lacked enough using the value complement of most like sample, most
The measure of similar sample is Wherein aI, jIt is i-th in data set
The characteristic value of j-th of feature of sample,For the characteristic value of missing;
The useless feature of S12, deletion on label value without influence;
S13, data set W is deleted1Feature of the middle characteristic value without ordering relation obtains data set W2。
Further, step S2 includes:
S21, data set W2Characteristic set be A={ a1, a2..., aNi, Ni is data set W2Feature sum;
S22, data set features aiCharacteristic value collection beWherein NaIt is data set W2It is total
Sample number and feature aiCharacteristic value sum;
S23, the label for first determining whether two samples are then skipped if they are the same and are compared the two samples, if label is different,
Size is compared correspondingly to the characteristic value under two all features of sample, calculation formula:
If f (A, B) be it is true, then have A, B constitute conflict to (A,
B);
S24, select first sample, successively by subsequent all samples in the way of step S23 with first sample
It being compared, construction conflict pair successively goes on, until iterating to the last one sample, second sample is then selected, according to
Secondary to be compared subsequent all samples with first sample in the way of step S23, construction conflict pair successively carries out down
It goes, until iterating to the last one sample;Similarly, stop iteration after selected penultimate sample is completeer.
Further, step S3 includes:
S31, all conflicts pair for constructing step S23 are configured to a conflict set C={ c1, c2..., cN, N is
The conflict of conflict set C is to sum.
Further, step S4 includes:
S41, statistics conflict are to sample s in the element of the left sidekThe number f of appearancelk;
S42, statistics conflict are to sample s in the element of the rightkThe number f of appearancerk;
S43, total frequency f is calculatedk=flk+frk;
S44, by sample skThe frequency f occurred with itkBetween mapping relations one by one, construct a dictionary D={ sk: fk,
K=1,2 ..., Na.
Further, step S5 includes:
S51, by the sample in dictionary D according to frequency fkIt is ranked up from high to low.
Further, step S6 includes:
S61, according to formulaSilhouette coefficient is calculated, wherein
For sample skCluster in dissmilarity degree, aI, kFor sample skThe characteristic value and rk of ith feature are sample skLabel;B (k)=
min{b(k)1, b (k)2... b (k)nIt is sample skCluster between dissmilarity degree,It is sample
This skWith the n-th cluster (i.e. label be rn classification) dissimilar degree;
If S62, sample skSilhouette coefficient s (k) < ε, then record sample skNumber, and be regarded as doubtful make an uproar
Sound sample, in data set W2In delete sample sk, obtain new data set W3。
S63, it deletes comprising doubtful noisy samples s in step S62 in conflict in set CkConflict pair.
Further, step S7 includes:
S71, S4, S5, S6 are repeated, until (in this patent, ε is set to the sample in step S62 without s (i) less than ε
0)。
Further, step S8 includes:
S81, respectively by data set W1With new data set W3It is divided into three parts according to same ratio, is training set respectively, tests
Card collection and test set;
S82, with use stacking method respectively in W1And W3Train model M1And M3;
S83, comparative assessment model M3Estimated performance.
After above method strategy, the positive effect of the present invention is:
(1) present invention devises a kind of completely different for the label noise data occurred in bridge defect data set
Noise elimination algorithm has found different sample conducts according to sample conflict number using the label collision between sample and sample
The difference of noise data probability size, increases the accuracy of label noise filtering, improves the quality of data of data set.
(2) compared to traditional method for carrying out label noise filtering using sorting algorithm, the present invention is by means of data set
The specificity of immanent structure itself improves the estimated performance for finally training machine learning model.
Detailed description of the invention
Reader is after having read a specific embodiment of the invention referring to attached drawing, it will more clearly understands of the invention
Various aspects, drawings in the following description are some embodiments of the invention, for those of ordinary skill in the art, not
Under the premise made the creative labor, it can also be obtained according to these attached drawings other accompanying drawings.
Fig. 1 is the flow diagram of the noise removal embodiment of the method for the invention towards bridge defect ratings data.
Fig. 2 is the specific stream of step S2 in the noise removal embodiment of the method for the invention towards bridge defect ratings data
Journey schematic diagram.
Fig. 3 is the specific stream of step S4 in the noise removal embodiment of the method for the invention towards bridge defect ratings data
Journey schematic diagram.
Fig. 4 is the specific stream of step S6 in the noise removal embodiment of the method for the invention towards bridge defect ratings data
Journey schematic diagram.
Fig. 5 is the specific stream of step S8 in the noise removal embodiment of the method for the invention towards bridge defect ratings data
Journey schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
S1, the data progress data prediction that initial data is concentrated is obtained into data set W1, to W1The spy of middle no ordering relation
Sign is removed, and obtains data set W2。
S11, using calorie fill method, utilize the value complement of most like sample missing characteristic value enough, the degree of most like sample
Amount method is Wherein ai, j are j-th of i-th of sample in data set
The characteristic value of feature,For the characteristic value of missing;
The useless feature of S12, deletion on label value without influence;
S13, data set W is deleted1Feature of the middle characteristic value without ordering relation, obtains data set W2。
S2, according to data set W2, it is based on feature aiCharacteristic value aI, jThe sample of different labels is compared two-by-two, is constructed
Conflict is to ci。
S21, data set W2Characteristic set be A={ a1, a2..., aNi, Ni is data set W2Feature sum;
S22, data set features aiCharacteristic value collection be D={ aI, 1, aI, 2..., aI, Na, Na is data set W2Gross sample
This number and feature aiCharacteristic value sum;
S23, the label for first determining whether two samples are then skipped if they are the same and are compared the two samples, if label is different,
Size is compared correspondingly to the characteristic value under two all features of sample, calculation formula:
If f (A, B) be it is true, then have A, B constitute conflict to (A,
B)。
S24, select first sample, successively by subsequent all samples in the way of step S23 with first sample
It being compared, construction conflict pair successively goes on, until iterating to the last one sample, second sample is then selected, according to
Secondary to be compared subsequent all samples with first sample in the way of step S23, construction conflict pair successively carries out down
It goes, until iterating to the last one sample;Similarly, stop iteration after selected penultimate sample is completeer.
S3, according to conflict to ciConstruction conflict set C={ c1, c2..., cN, wherein N is the conflict of conflict set C to total
Number.
S31, all conflicts for constructing step S23 are to one conflict set C={ c of composition1, c2..., cN, wherein N is
The conflict of conflict set C is to sum.
S4, pass through sample s in statistics conflict set CkThe frequency f of appearancek, obtain dictionary D={ sk: fk}。
S41, statistics conflict are to sample s in the element of the left sidekThe number f of appearancelk;
S42, statistics conflict are to sample s in the element of the rightkThe number f of appearancerk;
S43, frequency f is calculatedk=flk+frk;
S44, by sample skThe frequency f occurred with itkBetween mapping relations one by one, construct a dictionary D={ sk: fk,
K=1,2 ..., Na.
S5, the sample in dictionary D is ranked up from high to low by the frequency.
S51, by the sample in dictionary D according to frequency fkIt is ranked up from high to low.
S6, for t% preceding after sequence sample in data set W2Middle calculating silhouette coefficient s (k), deleting s, (k is less than the sample of ε
This sk, obtain filtered new data set W3, while delete conflict in set C comprising doubtful noisy samples skConflict pair.
S61, according to formulaSilhouette coefficient is calculated, wherein
For sample skCluster in dissmilarity degree, aI, kFor sample skThe characteristic value and rk of ith feature are sample skLabel;B (k)=
min{b(k)1, b (k)2... b (k)nIt is sample skCluster between dissmilarity degree,It is sample
This skWith the n-th cluster (i.e. label be rn classification) dissimilar degree;
If S62, sample skSilhouette coefficient s (k) < ε, then record sample skNumber, and be regarded as doubtful make an uproar
Sound sample, in data set W2In delete sample sk, obtain new data set W3;
S63, it deletes comprising doubtful noisy samples s in step S62 in conflict in set CkConflict pair.
S7, S4, S5, S6 are repeated, until being less than the sample of ε in step S62 without s (i).
S71, S4, S5, S6 are repeated, until (in this patent, ε is set to the sample in step S62 without s (i) less than ε
0)。
S8, it is based on data set W1And W3Model M is respectively trained out with stacking method1And M3, assess simultaneously comparison model
M3Estimated performance.
S81, respectively by data set W1With new data set W3Three parts are divided into according to same ratio, respectively as training
Collection, verifying collection and test set;
S82, W is based respectively on identical machine learning algorithm1And W3Data set trains model M1And M3;
S83, assessment and comparison model M3Estimated performance.
Above, a specific embodiment of the invention is described with reference to the accompanying drawings.But those skilled in the art
It is understood that without departing from the spirit and scope of the present invention, can also make to a specific embodiment of the invention each
Kind change and replacement.These changes and replacement are all fallen within the scope of the invention as defined in the claims.
Claims (9)
1. a kind of denoising method towards bridge defect ratings data is rushed by compare two-by-two to sample data first
It is prominent then to be conflicted according to sample to the number occurred in set to set, and the silhouette coefficient of sample is combined to reject noise number
According to, obtain filtered data set, and then using stacking method respectively in raw data set and filtered new data
Model is trained on collection, finally assesses and compare the estimated performance of two models, to verify the validity of this denoising method, if
It confirms the validity, has just obtained the data set of a relative clean, specific steps are as follows:
S1, the data that initial data is concentrated are pre-processed to obtain data set W1, to W1The feature of middle no ordering relation is gone
It removes, obtains data set W2;
S2, according to data set W2, it is based on feature aiCharacteristic value aI, jThe sample of different labels is compared two-by-two, construction conflict
To ci;
S3, according to conflict to ciConstruction conflict set C={ c1, c2..., cN, N is the conflict of conflict set C to sum;
Sample s in S4, statistics conflict set CkThe frequency f of appearancek, obtain dictionary D={ sk: fk};
S5, the sample in dictionary D is ranked up from high to low by the frequency;
S6, to the sample of t% preceding after sequence in data set W2Middle calculating silhouette coefficient s (k) deletes the sample s that s (k) is less than εk,
Obtain filtered new data set W3, while delete conflict in set C comprising doubtful noisy samples skConflict pair;
S7, S4, S5, S6 are repeated, until being less than the sample of ε in step S62 without s (i);
S8, in data set W1And W3It is upper that model M is respectively trained out using same machine learning algorithm1And M3, comparative assessment model
M3Estimated performance.
2. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S1 is specifically included:
S11, it is based on data set W1, using calorie fill method, characteristic value is lacked enough using the value complement of most like sample, it is most like
The measure of sample isWherein aI, jFor i-th of sample in data set
The characteristic value of this j-th of feature,For the characteristic value of missing;
The useless feature of S12, deletion on label value without influence;
S13, data set W is deleted1Feature of the middle characteristic value without ordering relation, obtains data set W2。
3. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S2 is specifically included:
S21, data set W2Characteristic set be A={ a1, a2..., aNi, NiIt is data set W2Feature sum;
S22, data set features aiCharacteristic value collection be D={ aI, 1, aI, 2..., aI, Na, NaIt is data set W2Total number of samples,
It is also feature aiCharacteristic value sum;
S23, the label for first determining whether two samples are then skipped if they are the same and are compared the two samples, if label is different, to two samples
Characteristic value under all features compares size correspondingly, calculation formula:
If f (A, B) is very, then to have A, B to constitute conflict to (A, B);
S24, first sample is selected, successively carries out subsequent all samples with first sample in the way of step S23
Compare, construction conflict pair successively goes on, until iterating to the last one sample, then selectes second sample, successively will
Subsequent all samples are compared in the way of step S23 with first sample, and construction conflict pair successively goes on,
Until iterating to the last one sample;Similarly, stop iteration after selected penultimate sample is completeer.
4. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S3 is specifically included:
S31, all conflicts pair for constructing step S23 are configured to a conflict set C={ c1, c2..., cN, N is conflict
Collect the conflict of C to sum.
5. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S4 is specifically included:
S41, statistics conflict are to sample s in the element of the left sidekThe number f of appearancelk;
S42, statistics conflict are to sample s in the element of the rightkThe number f of appearancerk;
S43, total frequency f is calculatedk=flk+frk;
S44, by sample skThe frequency f occurred with itkBetween mapping relations one by one, construct a dictionary D={ sk: fk, k=1,
2 ..., Na.
6. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S5 is specifically included:
S51, by the sample in dictionary D according to frequency fkIt is ranked up from high to low.
7. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S6 is specifically included:
S61, according to formulaSilhouette coefficient is calculated, wherein
For sample skCluster in dissmilarity degree, aI, kFor sample skThe characteristic value and rk of ith feature are sample skLabel;B (k)=
min{b(k)1, b (k)2... b (k)nIt is sample skCluster between dissmilarity degree,It is sample
This skWith the n-th cluster (i.e. label be rn classification) dissimilar degree;
If S62, sample skSilhouette coefficient s (k) < ε, then record sample skNumber, and be regarded as doubtful noise sample
This, in data set W2In delete sample sk, obtain new data set W3;
S63, it deletes comprising doubtful noisy samples s in step S62 in conflict in set CkConflict pair.
8. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S7 is specifically included:
S71, S4, S5, S6 are repeated, until being less than the sample of ε in step S62 without s (i).(in this patent, the value of ε is set
For 0).
9. a kind of denoising method towards bridge defect ratings data according to claim 1, which is characterized in that described
Step S8 is specifically included:
S81, respectively by data set W1With new data set W3It is divided into three parts according to same ratio, is training set, verifying collection respectively
And test set;
S82, W is based respectively on identical machine learning algorithm1And W3Data set trains model M1And M3;
S83, assessment and comparison model M3Estimated performance.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910327313.0A CN110083637B (en) | 2019-04-23 | 2019-04-23 | Bridge disease rating data-oriented denoising method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910327313.0A CN110083637B (en) | 2019-04-23 | 2019-04-23 | Bridge disease rating data-oriented denoising method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110083637A true CN110083637A (en) | 2019-08-02 |
CN110083637B CN110083637B (en) | 2023-04-18 |
Family
ID=67416114
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910327313.0A Active CN110083637B (en) | 2019-04-23 | 2019-04-23 | Bridge disease rating data-oriented denoising method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110083637B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111352966A (en) * | 2020-02-24 | 2020-06-30 | 交通运输部水运科学研究所 | Data tag calibration method in autonomous navigation |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5341142A (en) * | 1987-07-24 | 1994-08-23 | Northrop Grumman Corporation | Target acquisition and tracking system |
US20070297645A1 (en) * | 2004-07-30 | 2007-12-27 | Pace Charles P | Apparatus and method for processing video data |
CN201023495Y (en) * | 2007-04-29 | 2008-02-20 | 余亚莉 | Vehicle mounted railway traffic passenger transport capacity and operation safety intelligence monitoring and prewarning system |
CN101639934A (en) * | 2009-09-04 | 2010-02-03 | 西安电子科技大学 | SAR image denoising method based on contour wave domain block hidden Markov model |
CN102340884A (en) * | 2010-07-26 | 2012-02-01 | 中兴通讯股份有限公司 | Base station and method for preventing random access conflicts |
CN102509083A (en) * | 2011-11-19 | 2012-06-20 | 广州大学 | Detection method for body conflict event |
CN103150409A (en) * | 2013-04-08 | 2013-06-12 | 深圳市宜搜科技发展有限公司 | Method and system for recommending user search word |
CN104077749A (en) * | 2014-06-17 | 2014-10-01 | 长江大学 | Seismic data denoising method based on contourlet transformation |
CN104966090A (en) * | 2015-07-21 | 2015-10-07 | 公安部第三研究所 | Visual word generation and evaluation system and method for realizing image comprehension |
CN105551028A (en) * | 2015-12-09 | 2016-05-04 | 中山大学 | Method and system for dynamically updating geographic space data based on remote sensing image |
US20160314583A1 (en) * | 2013-12-19 | 2016-10-27 | Axon Dx, Llc | Cell detection, capture and isolation methods and apparatus |
CN108073940A (en) * | 2016-11-18 | 2018-05-25 | 北京航空航天大学 | A kind of method of 3D object instance object detections in unstructured moving grids |
CN108549904A (en) * | 2018-03-28 | 2018-09-18 | 西安理工大学 | Difference secret protection K-means clustering methods based on silhouette coefficient |
CN109389143A (en) * | 2018-06-19 | 2019-02-26 | 北京九章云极科技有限公司 | A kind of Data Analysis Services system and method for automatic modeling |
-
2019
- 2019-04-23 CN CN201910327313.0A patent/CN110083637B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5341142A (en) * | 1987-07-24 | 1994-08-23 | Northrop Grumman Corporation | Target acquisition and tracking system |
US20070297645A1 (en) * | 2004-07-30 | 2007-12-27 | Pace Charles P | Apparatus and method for processing video data |
CN201023495Y (en) * | 2007-04-29 | 2008-02-20 | 余亚莉 | Vehicle mounted railway traffic passenger transport capacity and operation safety intelligence monitoring and prewarning system |
CN101639934A (en) * | 2009-09-04 | 2010-02-03 | 西安电子科技大学 | SAR image denoising method based on contour wave domain block hidden Markov model |
CN102340884A (en) * | 2010-07-26 | 2012-02-01 | 中兴通讯股份有限公司 | Base station and method for preventing random access conflicts |
CN102509083A (en) * | 2011-11-19 | 2012-06-20 | 广州大学 | Detection method for body conflict event |
CN103150409A (en) * | 2013-04-08 | 2013-06-12 | 深圳市宜搜科技发展有限公司 | Method and system for recommending user search word |
US20160314583A1 (en) * | 2013-12-19 | 2016-10-27 | Axon Dx, Llc | Cell detection, capture and isolation methods and apparatus |
CN104077749A (en) * | 2014-06-17 | 2014-10-01 | 长江大学 | Seismic data denoising method based on contourlet transformation |
CN104966090A (en) * | 2015-07-21 | 2015-10-07 | 公安部第三研究所 | Visual word generation and evaluation system and method for realizing image comprehension |
CN105551028A (en) * | 2015-12-09 | 2016-05-04 | 中山大学 | Method and system for dynamically updating geographic space data based on remote sensing image |
CN108073940A (en) * | 2016-11-18 | 2018-05-25 | 北京航空航天大学 | A kind of method of 3D object instance object detections in unstructured moving grids |
CN108549904A (en) * | 2018-03-28 | 2018-09-18 | 西安理工大学 | Difference secret protection K-means clustering methods based on silhouette coefficient |
CN109389143A (en) * | 2018-06-19 | 2019-02-26 | 北京九章云极科技有限公司 | A kind of Data Analysis Services system and method for automatic modeling |
Non-Patent Citations (2)
Title |
---|
P. SATTIGERI 等: ""De-noising and event extraction for silicon pore sensors using matrix decomposition"", 《 SENSOR SIGNAL PROCESSING FOR DEFENCE》 * |
游丹丹 等: ""我国网络舆情热点话题发现研究综述"", 《现代情报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111352966A (en) * | 2020-02-24 | 2020-06-30 | 交通运输部水运科学研究所 | Data tag calibration method in autonomous navigation |
Also Published As
Publication number | Publication date |
---|---|
CN110083637B (en) | 2023-04-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105913661B (en) | A kind of express highway section traffic state judging method based on charge data | |
US10482093B2 (en) | Data mining method | |
CN106528865A (en) | Quick and accurate cleaning method of traffic big data | |
WO2017143919A1 (en) | Method and apparatus for establishing data identification model | |
Mat et al. | The impact of infrastructure on foreign direct investment: The case of Malaysia | |
CN110197588A (en) | A kind of truck driving behavior appraisal procedure and device based on GPS track data | |
CN102637258B (en) | Method for creating online surface quality detection system defect library | |
CN110378869A (en) | A kind of rail fastening method for detecting abnormality of sample automatic marking | |
CN108090499A (en) | Data active mask method and system based on maximum information triple screening network | |
Hagler | Defining US megaregions | |
CN110120036A (en) | A kind of multiple dimensioned tire X-ray defect detection method | |
CN106898142B (en) | A kind of path forms time reliability degree calculation method considering section correlation | |
CN105426441B (en) | A kind of automatic preprocess method of time series | |
CN110362557A (en) | A kind of miss path restorative procedure based on machine learning and license plate identification data | |
CN103679214B (en) | Vehicle checking method based on online Class area estimation and multiple features Decision fusion | |
CN110400462B (en) | Track traffic passenger flow monitoring and early warning method and system based on fuzzy theory | |
CN111489008A (en) | Traffic accident influence factor spatial effect analysis method and application thereof | |
CN110503070A (en) | Traffic automation monitoring method based on Aerial Images object detection process technology | |
CN110059952A (en) | Vehicle insurance methods of risk assessment, device, equipment and storage medium | |
TW202127371A (en) | Image-based defect detection method and computer readable medium thereof | |
CN110083637A (en) | A kind of denoising method towards bridge defect ratings data | |
CN113327248A (en) | Tunnel traffic flow statistical method based on video | |
CN112308148A (en) | Defect category identification and twin neural network training method, device and storage medium | |
CN109829887B (en) | Image quality evaluation method based on deep neural network | |
CN109543712B (en) | Method for identifying entities on temporal data set |
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 |