CN111356227A - Feature selection method applied to obstruction category identification in UWB indoor positioning - Google Patents
Feature selection method applied to obstruction category identification in UWB indoor positioning Download PDFInfo
- Publication number
- CN111356227A CN111356227A CN202010114879.8A CN202010114879A CN111356227A CN 111356227 A CN111356227 A CN 111356227A CN 202010114879 A CN202010114879 A CN 202010114879A CN 111356227 A CN111356227 A CN 111356227A
- Authority
- CN
- China
- Prior art keywords
- feature
- constraint
- correlation
- evaluation index
- subset
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Complex Calculations (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a feature selection method for identifying the type of an obstruction in UWB indoor positioning. The invention designs a bidirectional search feature selection algorithm based on maximum correlation, minimum redundancy and maximum calculation cost feature evaluation indexes. The invention adopts a bidirectional search strategy to effectively ensure that the selected first characteristic is the characteristic with the maximum integral contribution degree, simultaneously considers the calculation cost and the characteristic redundancy, and can effectively ensure that the selected characteristic has the characteristics of large category correlation, small characteristic redundancy and small calculation cost. In addition, the timeliness and the robustness of the system are improved by setting three constraint thresholds, namely minimum correlation constraint, maximum calculation cost constraint and maximum evaluation index constraint. Compared with the traditional other feature selection methods, the method comprehensively considers the computing power of the system, and has better timeliness and universality.
Description
Technical Field
The invention relates to the field of UWB indoor positioning, in particular to a feature selection method for identifying the type of an obstruction in UWB indoor positioning.
Background
With the rapid development of science and technology, especially in the 5G era, the popularization of intelligent robots and unmanned vehicles has promoted the generation of more positioning-dependent application services, however, compared with the technology of outdoor positioning becoming mature, such as GPS and the like, the outdoor positioning can achieve a good effect, and the indoor positioning cannot achieve an ideal effect due to the difficulty in penetrating buildings.
In the existing indoor positioning system, the Ultra Wideband (UWB) technology becomes one of the most promising indoor positioning technologies in the present day due to its high precision, good delay resolution, low power consumption, and high robustness in a complex indoor environment, and especially has significant advantages in precision compared with other indoor positioning technologies. However, the biggest reason for hindering the development of UWB indoor positioning is the non-line-of-sight NLOS state caused by a variety of obstacles in an indoor complex environment, and propagation of UWB under the non-line-of-sight NLOS causes the distance between the BS and the MS to be greater than the actual distance, so that the final positioning accuracy is sharply reduced.
The identification of the NLOS state is one of the keys for improving the UWB indoor positioning accuracy, however, the existing method has the following problems: firstly, the traditional method only distinguishes a non-line-of-sight (NLOS) state from a line-of-sight (LOS) state, neglects the contribution of the type of an obstruction to the indoor space information perception, can well provide indoor prior information according to the type of the obstruction, and assists in improving the UWB indoor positioning accuracy. Secondly, when the NLOS state and the LOS state of the UWB are identified by adopting the characteristics of the Channel Impulse Response (CIR), the first characteristic selected by the traditional characteristic selection method is often the characteristic with the maximum correlation with the category, and the first characteristic selected by the traditional characteristic selection method is not necessarily the characteristic with the maximum contribution to the whole body because the combination effect of the characteristics is ignored. In addition, the traditional method usually ignores the calculation cost of the features, and some features usually need high calculation cost, so that the identification delay is caused in practical application, and the real-time identification cannot be realized. In practical UWB indoor positioning application, an optimal feature set with high category correlation, low redundancy among features, and low computation cost is often selected as an identification feature, thereby ensuring high accuracy and high real-time performance of system detection.
Disclosure of Invention
In order to solve the problems, the invention provides a feature selection method applied to identification of the type of an obstruction in UWB indoor positioning.
The invention comprises the following steps:
the method comprises the following steps: establishing a feature selection model based on CIR signal features
Given a data set D containing k samples, D ═ D1,d2,...,dkD, samples of each type in the data set Di={Xi,LiAll have feature set Xi={xi1,xi2,...,xinWhere n is 1,2, 8 denotes 8 CIR signal characteristics of first path energy, maximum amplitude, rise time, standard deviation, mean excess delay, root mean square delay spread, kurtosis, and skewness, respectively.
For the multi-classification problem, there is a set of labels L per samplei={li 1,li 2,...,li mWhere m is the number of occlusion species. li m1, -1, if li mThe representation is labeled as class m sample, otherwise-1.
The feature selection is to select a feature subset B from the feature set X under a certain constraint condition to be optimal under a certain evaluation index, wherein B ═ B1,b2,...,bn},bw1,0, (w 1, 2.., n), if b is greater than nw1 denotes the selected feature xwOn the contrary bw0 indicates no selection.
Step two: establishing a characteristic comprehensive evaluation index based on mutual information measurement:
setting a correlation D (x) between a feature and a tagiL), features and redundancy between features R (x)i,xj) Introducing a computation time cost C (B) of the selected feature subset, then
Where I (L;) represents the mutual information computation, | L | represents the number of label categories in the label set L, | X | represents the number of features in the feature set X, I (X)i;lj) Represents a feature xiAnd category label ljMutual information of, I (x)i;xj) Represents a feature xiAnd feature xjB is a selected feature subset, c (x)i) Represents a feature xiβ ofiRepresents a feature xiA relative weight parameter of the cost and the correlation is calculated.
And finally establishing a comprehensive evaluation index mRMRMC of maximum correlation, minimum redundancy and minimum calculation cost:
wherein xi∈BsRepresenting the candidate feature set BsThe selected feature subset B should satisfy the overall evaluation index
Step three, setting a minimum correlation constraint ηdComputing cost constraints ηcMaximum evaluation index constraint ηj。
Constraint 1: a minimum correlation constraint.
Setting a feature and tag correlation threshold ηdOnce a feature is associated with a tag, D (x)i,L)<ηdThen directly discard the feature xiWherein ηdThe selection of (2) is based on the correlation between the global feature vector and the label:
ηd=λdmax{D(xi;L),xi∈X} (7)
wherein 0<λd1 represents the minimum correlation threshold parameter.
Constraint 2: a cost constraint is calculated.
Setting a computational cost constraint threshold ηc,If C (B) + c (x)j)>ηcDirectly determining the optimal feature subset as BbestCalculate a cost constraint threshold η, BcThe method is determined according to the calculation complexity of the system characteristics and the calculation capability of the system in practical application.
Wherein 0<λc≦ 1 indicates calculating the cost threshold parameter.
Constraint 3: and (5) maximum evaluation index constraint.
Setting a maximum evaluation index constraint threshold ηj,0<ηj≤1,xj∈BsRepresenting the feature with the maximum evaluation criterion in the feature set to be selected ifThen it means adding xjThe recognition accuracy of the post-system can reach the actual requirement, and the optimal feature subset B can be obtained at the momentbest=B∪xj。
Step four: deleting features having a correlation less than a minimum correlation constraint threshold
If D (x)i,L)<ηdIf the correlation between the feature and the category is too small, the feature is directly deleted, and the subsequent calculation amount is reduced.
Step five: reverse deletion determines the first feature.
Firstly, calculating the relevance D of all the characteristics of the whole body and the labelall(X, L) and then deleting feature X in reverse orderiTo obtain deletion xiRelevance D of the latter feature set to the labelmiss(xi)(xiL), selecting the feature that results in the largest change in correlation as the first feature vector of the selected subset of features,
then adopting the forward direction to select the feature set B to be selected in turnsThe internal characteristics of the test piece are that the comprehensive evaluation index J ismRMRMC(xi) The largest feature is added to the selected feature subset B and finally the optimal feature subset is determined.
If it isThen the characteristic xiProvide the most information for the whole, so xiA first feature B (1) ═ x as a selected seti。
Step six: and (3) determining the next optimal feature from the feature set to be selected, and determining the optimal feature in the feature set to be selected according to the comprehensive evaluation index of the rmrmrmrmc established by the formula (5) and adding the optimal feature into the selected feature subset B.
Step seven: and determining the optimal feature subset according to the calculation cost constraint threshold and the maximum evaluation index constraint threshold.
And determining a calculation cost constraint threshold and a maximum evaluation index constraint threshold according to the actual system calculation capacity and demand, and if the evaluation index of the selected feature subset is greater than the maximum evaluation index constraint threshold or the calculation cost of the selected feature subset is greater than the calculation cost constraint threshold, taking the selected feature subset as the optimal feature subset.
The invention has the beneficial effects that:
firstly, in a UWB positioning system under a complex indoor environment, not only the feature selection is performed for the identification of two states of NLOS and LOS, but also the feature selection is performed by considering the identification of multiple obstruction types, and a feature set with the best identification effect on all the obstruction types and the LOS types is selected.
Secondly, not only the correlation between the features and the categories but also the redundancy between the features are considered, and meanwhile, the actual feature calculation cost is introduced to construct an mRMRMC feature comprehensive evaluation index, so that the features with higher timeliness and recognition effect in practical application can be comprehensively selected based on the index.
And finally, determining an optimal feature subset by adopting a bidirectional search strategy based on three constraints of setting minimum correlation constraint, calculating cost constraint and maximum evaluation index constraint, wherein reverse deletion ensures that the first feature is a feature which provides most information for the whole. Aiming at the identification of different shelters under the NLOS state in the complex indoor environment positioning of UWB, the feature selection method has higher timeliness and application universality.
Drawings
FIG. 1 is a graph of feature-to-category and relationship metrics between features;
FIG. 2 is a calculation result of mutual information correlation between 8 features and features under the same label;
fig. 3 is a flow chart of a bidirectional search feature selection algorithm based on an rmrmrmrmrmc comprehensive evaluation index.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The method comprises the following specific steps:
the method comprises the following steps: establishing feature selection based on CIR signal features
Given a data set D containing k samples, D ═ D1,d2,...,dkD, samples of each type in the data set Di={Xi,LiAll have feature set Xi={xi1,xi2,...,xinWhere n is 1,2, 8 denotes 8 CIR signal characteristics of first path energy, maximum amplitude, rise time, standard deviation, mean excess delay, root mean square delay spread, kurtosis, and skewness, respectively.
For the multi-classification problem, there is a set of labels L per samplei={li 1,li 2,...,li mWhere m is the number of occlusion species. li m1, -1, if li mThe representation is labeled as class m sample, otherwise-1.
The feature selection is to select a feature subset B from the feature set X under a certain constraint condition to be optimal under a certain evaluation index, wherein B ═ B1,b2,...,bn},bw1,0, (w 1, 2.., n), if b is greater than nw1 denotes the selected feature xwOn the contrary bw0 indicates no selection.
Step two: establishing a characteristic comprehensive evaluation index based on mutual information measurement:
mutual Information (MI) can be used to describe the degree of correlation between the feature set X and the tag set L.
The value of I (X; L) reflects the degree of correlation between X and L, and a larger value of I (X; L) indicates a stronger correlation between the two, whereas a smaller value of I (X; L) indicates independence between the two.
Feature and category label and featureRelationships between signatures can be classified as related, redundant, and independent according to mutual information. As shown in FIG. 1, X1,X2,X3And is related to L, X2And X3L is redundant (black area indicates redundant part), X1And X2Are independent of each other.
Then the correlation between different features under the same label can be represented by conditional mutual information, such as formula (12)
Wherein xiAnd xjRepresenting different characteristics,/kDenotes a label, p (x)i|l),p(xj|l),p(xi,xj|lk) Representing a conditional probability density function.
Setting a correlation D (x) between a feature and a tagiL), features and redundancy between features R (x)i,xj) The computational time cost c (b) of the selected feature subset is introduced. Based on the correlation between the features and the categories, a measurable multivariate variable model F can be obtained, and the specific form formula (13) of F is shown as follows:
in order to avoid loss of generality, the data is normalized and centralized to obtain an n × n-dimensional symmetric matrix, each element in the matrix represents mutual information correlation under the same label, and the mutual information correlation results among 8 different features and characteristics under the same label are calculated according to actually acquired data as shown in fig. 2, wherein feaurer 12, 8 represents first path energy, maximum amplitude, rise time, standard deviation, average excess delay, delay propagation root-mean-square, kurtosis and skewness respectively.
D (x) can be determined by calculating formula (11) and formula (1-4) from mutual informationi,L)、R(xi,xj) And C (B). Finally, the maximum correlation and the minimum correlation shown in the formula (5) are establishedComprehensive evaluation index J of redundancy and minimum calculation costmRMRMCUnder this comprehensive evaluation index, all features within the selected feature subset B are determined according to equation (6), see fig. 3.
Step three, setting a minimum correlation constraint ηdComputing cost constraints ηcMaximum evaluation index constraint ηj。
Constraint 1: a minimum correlation constraint.
Setting a feature and tag correlation threshold η according to equation (7)dOnce a feature is associated with a tag, D (x)i,L)<ηdThen directly discard the feature xi。
Constraint 2: a cost constraint is calculated.
Determining the calculation cost constraint threshold η shown in formula (8) according to the calculation complexity of the system characteristics and the calculation capability of the system in practical applicationc,If C (B) + c (x)j)>ηcDirectly determining the optimal feature subset as Bbest=B。
Constraint 3: and (5) maximum evaluation index constraint.
Setting a maximum evaluation index constraint threshold η according to the actual demand of the systemj,0<ηj≤1,xj∈BsRepresenting the feature with the maximum evaluation criterion in the feature set to be selected ifThen it means adding xjThe recognition accuracy of the post-system can reach the actual requirement, and the optimal feature subset B can be obtained at the momentbest=B∪xj。
Step four: deleting features having a correlation less than a minimum correlation constraint threshold
If D (x)i,L)<ηdIf the correlation between the feature and the category is too small, the feature is directly deleted, and the subsequent calculation amount is reduced.
Step five: reverse deletion determines the first feature.
First, the correlation D of all the characteristics of the whole and the label is calculated according to the formula (9)all(X, L) and then deleting feature X in reverse orderiCalculating the missing x according to the formula (10)iRelevance of the latter feature set to the labelThe feature that results in the greatest change in correlation is selected as the first feature vector of the selected subset of features. Then adopting the forward direction to select the feature set B to be selected in turnsInner characteristics, will evaluate index JmRMRMC(xi) The largest feature is added to the selected feature subset B and finally the optimal feature subset is determined.
If it isThen the characteristic xiProvide the most information for the whole, so xiA first feature B (1) ═ x as a selected seti。
Step six: and (3) determining the next optimal feature from the feature set to be selected, and determining the optimal feature in the feature set to be selected according to the comprehensive evaluation index of the rmrmrmrmc established by the formula (5) and adding the optimal feature into the selected feature subset B.
Step seven: and determining the optimal feature subset according to the calculation cost constraint threshold and the maximum evaluation index constraint threshold.
And determining a calculation cost constraint threshold and a maximum evaluation index constraint threshold according to the actual system calculation capacity and demand, and if the evaluation index of the selected feature subset is greater than the maximum evaluation index constraint threshold or the calculation cost of the selected feature subset is greater than the calculation cost constraint threshold, taking the selected feature subset as the optimal feature subset.
Claims (6)
1. The feature selection method applied to the identification of the type of the obstruction in UWB indoor positioning is characterized by comprising the following steps:
the method comprises the following steps: establishing a feature selection model based on CIR signal features
Given a data set D containing k samples, D ═ D1,d2,...,dkD, samples of each type in the data set Di={Xi,LiAll have feature set Xi={xi1,xi2,...,xin8, where n is 1,2, 8, respectively, representing 8 CIR signal characteristics of first path energy, maximum amplitude, rise time, standard deviation, mean excess delay, root mean square delay spread, kurtosis, and skewness;
for the multi-classification problem, there is a set of labels L per samplei={li 1,li 2,...,li mM is the number of types of obstacles; li m1, -1, if li m1, the representation is marked as an m-class sample, otherwise-1;
the feature selection is to select a feature subset B from the feature set X under a certain constraint condition to be optimal under a certain evaluation index, wherein B ═ B1,b2,...,bn},bw1,0, (w 1, 2.., n), if b is greater than nw1 denotes the selected feature xwOn the contrary bw0 indicates no selection;
step two: establishing characteristic evaluation index based on mutual information measurement
Setting a correlation D (x) between a feature and a tagiL), features and redundancy between features R (x)i,xj) Introducing a computation time cost C (B) of the selected feature subset; then
Where I (indicates mutual information calculation, | L | TableThe number of label categories in the label set L, | X | represents the number of features in the feature set X, I (X)i;lj) Represents a feature xiAnd category label ljMutual information of, I (x)i;xj) Represents a feature xiAnd feature xjB is a selected feature subset, c (x)i) Represents a feature xiβ ofiRepresents a feature xiCalculating relative weight parameters of the cost and the correlation;
the established maximum correlation, minimum redundancy and minimum calculation cost comprehensive evaluation index mRMRMC:
wherein xi∈BsRepresenting the candidate feature set BsThe selected feature subset B should satisfy the overall evaluation index
Step three, setting a minimum correlation constraint ηdComputing cost constraints ηcMaximum evaluation index constraint ηj
Constraint 1: constraint of minimum correlation
Setting a feature and tag correlation threshold ηdOnce a feature is associated with a tag, D (x)i,L)<ηdThen directly discard the feature xi;
Constraint 2: computing cost constraints
Setting a computational cost constraint threshold ηc,If C (B) + c (x)j)>ηcDirectly determining the optimal feature subset as Bbest=B;
Constraint 3: maximum evaluation index constraint
Setting a maximum evaluation index constraint threshold ηj,0<ηj≤1,xj∈BsRepresenting the characteristic with the maximum evaluation standard in the characteristic set to be selected; if it is notThen it means adding xjThe recognition accuracy of the post-system reaches the actual requirement, and the optimal feature subset B is obtained at the momentbest=B∪xj;
Step four: deleting features having a correlation less than a minimum correlation constraint threshold
Step five: reverse deletion determines the first feature;
step six: determining next optimal features from the feature set to be selected, and determining the optimal features in the feature set to be selected according to the comprehensive evaluation index mRMC and adding the optimal features into the selected feature subset B;
step seven: and determining the optimal feature subset according to the calculation cost constraint threshold and the maximum evaluation index constraint threshold.
3. the method of claim 1, wherein the tag correlation threshold η is a threshold for selecting the identity of an obstruction class in UWB indoor positioningdThe calculation formula of (2) is as follows:
ηd=λdmax{D(xi;L),xi∈X}
wherein 0<λd1 represents the minimum correlation threshold parameter.
4. The feature selection method applied to identification of the class of an obstruction in UWB indoor positioning according to claim 1, characterized in that:
compute cost constraint threshold ηcDetermining according to the calculation complexity of the system characteristics and the system calculation capacity in practical application
Wherein 0<λc≦ 1 indicates calculating the cost threshold parameter.
5. The feature selection method applied to identification of the class of an obstruction in UWB indoor positioning according to claim 1, characterized in that: the fifth step is specifically as follows:
firstly, calculating the relevance D of all the characteristics of the whole body and the labelall(X, L) and then deleting feature X in reverse orderiTo obtain deletion xiRelevance of the latter feature set to the labelThe feature that results in the largest change in relevance is selected as the first feature vector of the selected subset of features,
then adopting the forward direction to select the feature set B to be selected in turnsInner characteristics, will evaluate index JmRMRMC(xi) Adding the maximum feature to the selected feature subset B, and finally determining the optimal feature subset;
6. The feature selection method applied to identification of the class of an obstruction in UWB indoor positioning according to claim 1, characterized in that: the seventh step is specifically:
and determining a calculation cost constraint threshold and a maximum evaluation index constraint threshold according to the actual system calculation capacity and demand, and if the evaluation index of the selected feature subset is greater than the maximum evaluation index constraint threshold or the calculation cost of the selected feature subset is greater than the calculation cost constraint threshold, taking the selected feature subset as the optimal feature subset.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010114879.8A CN111356227B (en) | 2020-02-25 | 2020-02-25 | Feature selection method applied to obstruction category identification in UWB indoor positioning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010114879.8A CN111356227B (en) | 2020-02-25 | 2020-02-25 | Feature selection method applied to obstruction category identification in UWB indoor positioning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111356227A true CN111356227A (en) | 2020-06-30 |
CN111356227B CN111356227B (en) | 2021-08-03 |
Family
ID=71197151
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010114879.8A Active CN111356227B (en) | 2020-02-25 | 2020-02-25 | Feature selection method applied to obstruction category identification in UWB indoor positioning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111356227B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112215520A (en) * | 2020-10-28 | 2021-01-12 | 润联软件系统(深圳)有限公司 | Multi-target fusion passing method and device, computer equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101053269A (en) * | 2004-10-29 | 2007-10-10 | 艾利森电话股份有限公司 | Resource allocation in communication networks |
CN101300504A (en) * | 2005-11-01 | 2008-11-05 | M/A-Com公司 | Radio frequency locator system |
US20120221921A1 (en) * | 2011-02-25 | 2012-08-30 | Kabushiki Kaisha Toshiba | Memory device having multiple channels and method for accessing memory in the same |
-
2020
- 2020-02-25 CN CN202010114879.8A patent/CN111356227B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101053269A (en) * | 2004-10-29 | 2007-10-10 | 艾利森电话股份有限公司 | Resource allocation in communication networks |
CN101300504A (en) * | 2005-11-01 | 2008-11-05 | M/A-Com公司 | Radio frequency locator system |
US20120221921A1 (en) * | 2011-02-25 | 2012-08-30 | Kabushiki Kaisha Toshiba | Memory device having multiple channels and method for accessing memory in the same |
Non-Patent Citations (1)
Title |
---|
EBTESAM ALMAZROUEI: "Measurement and Analysis of NLOS Identification Metrics for WLAN Systems", 《2014 IEEE 25TH INTERNATIONAL SYMPOSIUM ON PERSONAL, INDOOR AND MOBILE RADIO COMMUNICATIONS》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112215520A (en) * | 2020-10-28 | 2021-01-12 | 润联软件系统(深圳)有限公司 | Multi-target fusion passing method and device, computer equipment and storage medium |
CN112215520B (en) * | 2020-10-28 | 2021-06-08 | 润联软件系统(深圳)有限公司 | Multi-target fusion passing method and device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN111356227B (en) | 2021-08-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5235913B2 (en) | How to train a multi-class classifier | |
CN114449452B (en) | Wi-Fi indoor positioning method based on CNN-RNN | |
CN110163060B (en) | Method for determining crowd density in image and electronic equipment | |
CN109379780A (en) | Wireless sensor network locating method based on adaptive differential evolution algorithm | |
CN114218292A (en) | Multi-element time sequence similarity retrieval method | |
CN111356227B (en) | Feature selection method applied to obstruction category identification in UWB indoor positioning | |
Wen et al. | End-to-end semi-supervised learning for differentiable particle filters | |
CN113657472A (en) | Multi-source remote sensing data fusion method based on subspace learning | |
CN117350330A (en) | Semi-supervised entity alignment method based on hybrid teaching | |
CN110070120B (en) | Depth measurement learning method and system based on discrimination sampling strategy | |
CN117173217A (en) | Video multi-target tracking method based on multi-scale channel feature aggregation | |
CN116955650A (en) | Information retrieval optimization method and system based on small sample knowledge graph completion | |
CN115658927A (en) | Time sequence knowledge graph-oriented unsupervised entity alignment method and device | |
CN115661539A (en) | Less-sample image identification method embedded with uncertainty information | |
CN114051202A (en) | Automatic learning method for wireless positioning signal fingerprint sample library | |
CN114494972A (en) | Target tracking method and system combining channel selection and position optimization | |
CN108805162A (en) | A kind of saccharomycete multiple labeling feature selection approach and device based on particle group optimizing | |
CN111328015B (en) | Wireless sensor network target tracking method based on Fisher information distance | |
CN110543896B (en) | Heterogeneous crowdsourcing fingerprint labeling method based on semi-supervised naive Bayes | |
CN113724325A (en) | Multi-scene monocular camera pose regression method based on graph convolution network | |
CN114067293B (en) | Vehicle weight identification rearrangement method and system based on dual attributes and electronic equipment | |
CN113572732B (en) | Multi-step attack modeling and prediction method based on VAE and aggregated HMM | |
CN108984691B (en) | Road network moving object range query optimization method based on probability pruning | |
Zhang et al. | Deep Trajectory Similarity Model: A Fast Method for Trajectory Similarity Computation | |
Luo et al. | Fuzzy Clustering Algorithm for Interval Data Based on Feedback RBF 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 |