CN112731285A - Cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning - Google Patents
Cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning Download PDFInfo
- Publication number
- CN112731285A CN112731285A CN202011527510.6A CN202011527510A CN112731285A CN 112731285 A CN112731285 A CN 112731285A CN 202011527510 A CN202011527510 A CN 202011527510A CN 112731285 A CN112731285 A CN 112731285A
- Authority
- CN
- China
- Prior art keywords
- data
- time
- radio signal
- geodesic
- positioning
- 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
- 238000000034 method Methods 0.000 title claims abstract description 86
- 238000013526 transfer learning Methods 0.000 title claims abstract description 12
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000013508 migration Methods 0.000 claims abstract description 15
- 230000005012 migration Effects 0.000 claims abstract description 15
- 238000009826 distribution Methods 0.000 claims description 66
- 238000005070 sampling Methods 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 6
- 230000001131 transforming effect Effects 0.000 claims description 6
- 230000000295 complement effect Effects 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 5
- 230000007613 environmental effect Effects 0.000 abstract description 3
- 230000006872 improvement Effects 0.000 abstract description 3
- 230000006978 adaptation Effects 0.000 description 11
- 238000003066 decision tree Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000001186 cumulative effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000010295 mobile communication Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000001413 cellular effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007500 overflow downdraw method Methods 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000002542 deteriorative effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 239000012466 permeate Substances 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0252—Radio frequency fingerprinting
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention provides a cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning, which comprises the following steps: step one, multi-source radio signal data fusion; step two, calculating a cross-time geodesic flow core; step three, integrating classifiers; and step four, integrating the fingerprint positioning of the classifier. According to the invention, through the fusion and migration of the data characteristics of the multi-source radio signal, the positioning stability and the positioning precision can be improved, and the radio signal fingerprint change between different times is smoothed. The problem of present because of wireless signal fingerprint positioning signal source is single and easily receive the environmental impact is solved, the improvement positioning accuracy that can be very big.
Description
Technical Field
The invention relates to the technical field of radio positioning, in particular to a cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning.
Background
The accurate position information has strong economic, social and military significance in a plurality of fields, and various positioning technologies are widely applied to the fields of transportation, surveying and mapping, geographic information, forest fire prevention, communication time system, disaster relief and reduction, emergency search and rescue and the like and gradually permeate into aspects of production and life of human society. The most common outdoor positioning method at present is satellite navigation positioning, which can provide high-precision geographical position for users. At present, there are four satellite navigation systems, which are the global positioning system GPS in the united states, the beidou navigation system BDS in china, the glonass navigation system glonass in russia, and the Galileo navigation system Galileo in europe.
In recent years, demand for high-precision indoor positioning has been increasing. Although the satellite navigation positioning system can provide a high-precision geographical position outdoors, the satellite signal is weak and is easy to be shielded, for example, in urban canyons, tunnels, indoors and other occasions, the positioning error of the satellite navigation is too large or even cannot be used, and the requirement of people in daily life cannot be met.
Indoor positioning can also be realized by utilizing a wireless communication network, and positioning calculation is carried out by measuring certain parameters of the wireless communication network. The method is further classified into a ranging method and a non-ranging method. Ranging methods are mainly used in cellular mobile communication networks, and perform positioning calculation by measuring time of arrival (TOA), time difference of arrival (TDOA), or angle of arrival (AOA) between a mobile terminal and a plurality of base stations. However, the positioning method has acceptable positioning effect only in a line-of-sight environment, and generally has positioning accuracy only up to tens of meters, which is lower in the complex wireless environment.
The non-ranging positioning method is based on the dependence of wireless signal propagation on the environment, and the wireless signals at different positions have unique distinguishable characteristics, so that the positions can be corresponding by utilizing the signal characteristics. Such methods are also referred to as electromagnetic fingerprint location or fingerprint location, and these electromagnetic signal features are also referred to as electromagnetic fingerprints, since their principle is similar to that of a finger fingerprint to identify a unique person. Document [1] proposes for the first time the use of Wi-Fi Signal Strength information (RSS) for positioning, which divides positioning into two phases: the first stage is an off-line stage, wireless signals of a plurality of reference positions in an acquisition area are collected, and signal characteristics are extracted to construct a fingerprint database; the second phase is an online phase, where the location is determined by matching the signals acquired in real time with the data in the fingerprint library. In recent years, some scholars have studied the problem of improving the fingerprint positioning accuracy by using an artificial intelligence algorithm. In document [2], position estimation is studied as a machine learning problem, and a position is estimated by a probabilistic method. Document [3] trains the fingerprint data in the offline stage by using a Support Vector Machine (SVM) in Machine learning, and performs classification prediction in the online stage. Document [4] performs position estimation using a deep learning method.
The fingerprint positioning is a non-ranging method, and solves the problem of overlarge positioning error of the ranging method under a non-direct path by utilizing the dependence of wireless signal propagation on the environment. Ideally, different locations in the area have uniquely distinguishable wireless signal characteristics. However, since wireless signals are affected by multipath effects and fast fading, it is also possible that two fingerprints that are geographically close together have a small correlation.
For example, existing fingerprint positioning is mainly based on RSS values of signals such as WiFi and bluetooth as fingerprint features, and positioning accuracy of tens of meters can be achieved within a small range. However, these methods require a large number of nodes to be deployed, a large number of reference point data to be collected, and the coverage area is relatively limited. In addition, the fingerprint positioning method based on WiFi, bluetooth and the like also faces the problems of poor environmental adaptation, large positioning error caused by signal time variation and the like.
Because the non-ranging method has greater dependence on the environment, the positioning error of fingerprint positioning becomes larger when the environment is changed violently; in addition, due to the time-varying nature of wireless signals, the distribution of signals acquired at different times is not the same, which makes the fingerprint characteristics at the same location not the same at different times. In practice, if the fingerprint database cannot be updated in time, the positioning error becomes large. In response to these problems, document [5] proposes to update the fingerprint database in a crowdsourced manner, and to continuously update the fingerprint database with public collected data. However, since there may be errors in the public location information itself, and such errors may cause accumulated errors after being submitted to the fingerprint database, thereby deteriorating the positioning result, the collected data needs to be filtered. Document [6] proposes to perform multi-source fusion positioning in combination with inertial navigation, but inertial navigation also accumulates errors, and requires an Inertial Measurement Unit (IMU) to have higher accuracy.
The traditional fingerprint positioning method assumes that the distribution of the acquired signals is basically unchanged, but in a real complex environment, the distribution of data is changed due to the movement of an object and the change of time. However, it is difficult to collect new data at every time, which requires migration learning such a method that data can be migrated under different data distributions.
The core of the transfer learning is to find the similarity between a new problem and an original problem, take the original problem as a source domain and the new problem as a target domain, and perform knowledge transfer between the source domain and the target domain. The problem of field self-adaptation is one of research contents of transfer learning, and the problem that the feature space is consistent, the category space is consistent and only the feature distribution is inconsistent is mainly solved. This is very similar to the problem of fingerprint localization: the fingerprint locations are spatially consistent in category but not in feature distribution. Document [7] applies migration learning to migrate data collected at different reference point distributions, so as to reduce the cost of re-collection and training. Document [8] studies the feasibility of transfer learning in reducing the temporal impact of fingerprint localization. Document [9] proposes a data edge distribution adaptive migration Component Analysis (TCA) method. The objective of the TCA method is to learn a feature mapping on a reproducible hilbert space by means of maximum mean difference, so that the data distribution after mapping is close. However, the TCA method only considers the edge distribution adaptation of data, and does not consider the conditional distribution adaptation of data. Document [10] proposes a joint distribution adaptation (JDA, JointDistributionAdaptation) method that adapts both the edge distribution and the conditional distribution of data. The edge distribution of the JDA method is adapted the same as the TCA method, and the edge distribution is approximated by generating a pseudo label when adapting the conditional distribution. In the JDA method, the edge distribution adaptation and the conditional distribution adaptation are considered to work equally, but actually, the contributions of these two parts are not equal in the domain adaptation. Document [11] proposes a Balanced Distribution Adaptation (BDA) method, in which balancing factors are added to dynamically adjust the importance of edge distribution and conditional distribution based on a JDA method, and the balancing factors in the method are respectively given by global and local a-Distance approximations of two domain data. Although the BDA method gives quantitative estimation of the contribution of edge distribution and condition distribution in the field adaptation for the first time, the BDA method does not solve the problem of accurate calculation of balance factors, has similar effect to JDA in some cases, and cannot ensure the accuracy.
Data in the domain adaptation problem comes from two different but similar domains, and how to find the similarity of the two domains is a key problem for migration. Besides adapting the data distribution, manifold learning is also an important direction of transfer learning. The basic assumption is that existing data is sampled from a high-dimensional space, and therefore, it has a low-dimensional manifold structure in the high-dimensional space. Since the features in manifold space are usually very geometrically well defined, warping can be avoided, and thus the features in original space can be transformed into manifold space.
Document [12] proposes a domain adaptive method based on manifold learning, a Geodesic Flow Kernel (GFK) method, in which a geodesic line represents a line connecting two points at the shortest distance in a high-dimensional space. The GFK method learns the differences in the two domain subspaces and the incremental changes of common features and gives a low-dimensional representation of the invariance between the two domains. The data is transformed through the low-dimensional representation, and then the transformed data is trained by using some learning algorithms, so that the self-adaption of the change in different fields can be realized.
The GFK method is shown in FIG. 1, and embeds data of a source domain S and a target domain T into a Grassmann manifold, RDGrassmann manifold of upper total d-dimensional subspace as GD,dWhere D is the data dimension and D is the subspace dimension, then GD,dCan be regarded as quotient space SO (D)/(SO (D) × SO (D-D)), where SO (D) is a D × D plum cluster. Therefore, the source domain and the target domain can be regarded as GD,dTwo points, then two geodesic lines can be considered as a single parameter exponential flow,t aexp (tB), wherein
A∈R(D-d)×dThe submatrix a represents the direction and speed of the geodesic from the starting point to the end point, and t represents the time taken for the geodesic to flow from the starting point to the end point. The GFK method integrates infinite subspaces into the geodesic flow from the source domain subspace to the target domain subspace, and the calculation of the geodesic flow kernel between two points means that the low-dimensional representation of invariance between two fields is learned.
For the conventional fingerprint positioning method, a single radio signal data source cannot provide many signal characteristics and is sensitive to radio signal changes. Various migration learning methods can only migrate between a single source domain and a single target domain, and cannot migrate between a plurality of source domains and target domains.
The references are as follows:
[1]P.Bahl and V.N.Padmanabhan,"RADAR:an in-building RF-based user location and tracking system,"Proceedings IEEE INFOCOM 2000.Conference on Computer Communications.Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies(Cat.No.00CH37064),Tel Aviv,Israel,2000,pp.775-784 vol.2,doi:10.1109/INFCOM.2000.832252.
[2]T.Roos,
H.Tirri,et al.A Probabilistic Approach to WLAN User Location Estimation.International Journal of Wireless Information Networks,pp.155–164(2002).
[3]M.Brunato,R.Battiti,Statistical learning theory for location fingerprinting in wireless LANs,Computer Networks,Volume 47,Issue 6,2005,pp.825-845.
[4]X.Wang,L.Gao,S.Mao and S.Pandey,"DeepFi:Deep learning for indoor fingerprinting using channel state information,"2015 IEEE Wireless Communications and Networking Conference(WCNC),New Orleans,LA,2015,pp.1666-1671,doi:10.1109/WCNC.2015.7127718.
[5]B.Wang,Q.Chen,L.T.Yang and H.Chao,"Indoor smartphone localization via fingerprint crowdsourcing:challenges and approaches,"in IEEE Wireless Communications,vol.23,no.3,pp.82-89,June 2016,doi:10.1109/MWC.2016.7498078.
[6]Y.Kim,Y.Chon,H.Cha.Smartphone-based collaborative and autonomous radio fingerprinting.IEEE Transactions on Systems,Man and Cybernetics Part C:Applications and Reviews,2012.42(1),112–122.
[7]S.Sorour,Y.Lostanlen,S.Valaee and K.Majeed,"Joint Indoor Localization and Radio Map Construction with Limited Deployment Load,"in IEEE Transactions on Mobile Computing,2015.14(5):1031-1043,doi:10.1109/TMC.2014.2343636.
[8]V.W.Zheng,E.W.Xiang,Q.Yang,D.Shen.Transferring Localization Models over Time.InAAAI2008.1421-1426.
[9]S.J.Pan,I.W.Tsang,J.T.Kwok and Q.Yang,"Domain Adaptation via Transfer Component Analysis,"in IEEE Transactions on Neural Networks,2011.22(2):199-210,doi:10.1109/TNN.2010.2091281.
[10]M.Long,J.Wang,G.Ding,J.Sun and P.S.Yu,"Transfer Feature Learning with Joint Distribution Adaptation,"IEEE International Conference on Computer Vision,Sydney,NSW,2013.2200-2207,doi:10.1109/ICCV.2013.274.
[11]J.Wang,Y.Chen,S.Hao,W.Feng and Z.Shen,"Balanced Distribution Adaptation for Transfer Learning,"IEEE International Conference on Data Mining(ICDM).2017.1129-1134,doi:10.1109/ICDM.2017.150.
[12]B.Gong,Y.Shi,F.Sha and K.Grauman,"Geodesic flow kernel for unsupervised domain adaptation,"IEEE Conference on Computer Vision and Pattern Recognition,2012.2066-2073,doi:10.1109/CVPR.2012.6247911.
disclosure of Invention
The invention aims to provide a cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning, so as to solve the problems of the traditional positioning method.
The invention provides a cross-time multi-source radio signal positioning method based on geodesic flow kernel transfer learning, which comprises the following steps:
step one, multi-source radio signal data fusion: carrying out multiple random sampling on the collected multi-source radio signal intensity values to obtain a plurality of data sets;
step two, calculating a cross-time geodesic flow core: respectively calculating a geodesic flow kernel and data distribution difference measurement for the plurality of data sets obtained in the step one, and fusing the geodesic flow kernels at different times by adopting the data distribution difference measurement to obtain a cross-time geodesic flow kernel;
step three, integrating classifiers: transforming the data in the plurality of data sets obtained in the step one through the cross-time geodesic flow kernel calculated in the step two, training a plurality of base classifiers, and grouping the plurality of base classifiers into an inter-group classifier and an intra-group classifier;
step four, integrating classifier fingerprint positioning: and (4) collecting multi-source radio signal data in real time, processing and converting the data in the first step and the second step, predicting through the inter-group classifier and the intra-group classifier obtained in the third step, voting on a prediction result, and obtaining a final positioning result with the largest number of votes.
Further, step one includes the following substeps:
step 11, acquiring radio signal data at different times:
begin scanning channels of airborne multi-source radio signals, set NcThe channel number of the wireless electromagnetic signal is
After the number of channels of each radio signal is determined, recording the strength values of the multi-source radio signals collected at different time at each preset position;
carrying out random sampling on N data samples of the collected multi-source radio signal strength value for K times to obtain K data sets; wherein each is randomThe sampling method comprises the following steps: randomly extracting N data samples from the collected multi-source radio signal intensity value, wherein N is less than or equal to N; then, for each data sample of the randomly extracted n data samples at aiRandomly extracting m data from the data collected by each channel, wherein m is less than or equal to ai，i∈NCThus, n × m dimensional data is obtained, and then the n × m dimensional data is spliced together as a new data set, which is recorded as:
k data sets obtained by K random sampling are recorded as: RSS ═ RSS1 RSS2 … RSSK}。
Further, step two includes the following substeps:
step 21, constructing a source domain and a target domain;
step 22, constructing a geodesic flow;
step 23, calculating a geodesic flow kernel:
step 24, calculating a data distribution difference measure:
and step 25, fusing a plurality of geodesic flow cores at different times.
Further, the method of step 21 is: setting a Time set as Time { Time ] for a plurality of data sets RSS acquired at different times in the first step1,time2,…,timeRR is the number of time set elements; a pair of source domain time and target domain time (time) is formed between different elements in the time sets,timet) Wherein S and T represent subscripts corresponding to the source domain S and the target domain T, respectively; data sets RSS in a plurality of data sets RSS acquired at different times in the step onekCorresponds to times,timetAs the source domain S and the target domain T.
Further, remember
Is the base of the subspace of the source domain S and the target domain T in the ith data set, and all the subspaces form Grassmann manifold
The method for constructing a geodesic flow in step 22 comprises the following steps:
(1) solving the S subspace base of the source domain
Of orthogonal complement
(2) By a pair of formulas
Performing singular value decomposition to obtain an orthogonal matrix
And diagonal matrix Γ ii(t),Σi(t) its diagonal element is cos θa,simθa(ii) a Wherein a is 1,2, …, d, thetaaIs PS,PTThe angle of the main angle of (a) is,
they measure the coverage angle between subspaces;
(3) the starting point and the end point of the source domain S and the target domain T as the geodesic lines are expressed as follows:
connection of
The geodesic lines of (a) are expressed as:
wherein t is ∈ [0, 1]]。
Further, the method of step 23 is:
(1) will Si,TiIs projected to phii(t) obtaining their infinite dimensional subspace
(2) Computing
The inner product of (a) yields the geodesic flow kernel:
(4) geodesic core GiThe closed-form solution of (c) can be obtained from the result in step 22:
wherein, Λi,1To Λi,3For a diagonal matrix, the diagonal elements are:
further, the method of step 24 is:
(1) calculating all joint distribution sets psi (S, T) taking S, T as edge distribution;
(2) sample (x, y) from the joint distribution set Ψ (S, T): μ, calculating the distance ρ (x, y) between samples x and y and the expectation of the distance for the samples under μ for the joint distribution, where p represents the p-norm:
(3) computing the desired infimum bound of distance, i.e. the measure of difference between two data distributions, in all sets of joint distributions Ψ (S, T)
Further, the method of step 25 is: measure of difference between the data distributions obtained according to step 24
Weighting and adding all the geodesic flow cores to obtain the cross-time geodesic flow core
Further, the third step comprises the following sub-steps:
step 31, in the step one, obtaining K data sets RSS { RSS by randomly sampling the multi-source radio signal intensity value for K times1 RSS2 … RSSKAfter the calculation, calculating cross-time geodesic flow kernels G of the K data sets according to the method of the step twok,k＝1,2…K；
Step 32, grouping the reference points for acquiring the radio signals, roughly positioning between groups, and positioning in detail in the groups;
step 33, with GkTransforming the grouped source domain S, the transformed data using a base classifier model fk(Gk,Xk,Yk) Carry out training, Xk，YkGrouping training data and labels for the kth round; the final inter-group classifier consists of K base classifiers;
at step 34, the data in each group is trained in the same manner as in step 33 to obtain the intra-group basis classifiers.
Further, the step four includes the following substeps:
step 41, collecting multi-source radio signal data in real time;
step 42, fusing multi-source radio signal data acquired in real time according to the method in the step one, and then taking a plurality of obtained data sets as target domain data;
step 43, transforming the target domain data by using the cross-time geodesic flow core obtained by the calculation in the step two;
step 44, classifying and predicting the transformed target domain data through the inter-group base classifiers obtained in the step three, voting prediction results of all the inter-group classifiers, and taking the most voted result as a rough positioning result;
and step 45, performing classification prediction by using the corresponding intra-group classifiers according to the rough positioning result, voting the prediction results of all the intra-group classifiers, and taking the most voted result as the final positioning result.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
according to the invention, through the fusion and migration of the data characteristics of the multi-source radio signal, the positioning stability and the positioning precision can be improved, and the radio signal fingerprint change between different times is smoothed. The problem of present because of wireless signal fingerprint positioning signal source is single and easily receive the environmental impact is solved, the improvement positioning accuracy that can be very big.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
Fig. 1 is a schematic diagram of a conventional GFK method.
Fig. 2 is a flowchart of a cross-time multi-source radio signal positioning method based on geodesic core migration learning according to the present invention.
FIG. 3 is a flow chart of multi-source radio signal data fusion in accordance with the present invention.
FIG. 4 is a flow chart of the present invention for computing a cross-time geodetic flow kernel.
FIG. 5 is a flow chart of classifier integration according to the present invention.
FIG. 6 is a flow chart of integrated classifier fingerprint location of the present invention.
Fig. 7 is a schematic diagram of exemplary multi-source radio signal data fusion in accordance with the present invention.
Fig. 8 is a graph comparing the positioning distance error of an example office building-2 level according to the present invention.
FIG. 9 is a graph comparing the positioning distance error of 0 floor of office building according to an exemplary embodiment of the present invention
FIG. 10 is a graph comparing the positioning distance error of the 1 st floor of the office building according to the example of the present invention
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
The time-crossing multi-source radio signal positioning method based on the geodesic stream core transfer learning provided by the embodiment comprises four steps of multi-source radio signal data fusion, time-crossing geodesic stream core fusion, classifier integration and integrated classifier fingerprint positioning, and referring to fig. 2, the multi-source radio signal positioning method specifically comprises the following steps:
step one, multi-source radio signal data fusion:
in the first step, collected multi-source radio signal strength values are randomly sampled for multiple times to obtain multiple data sets, so that the collected multi-source radio signal strength values are fused together, referring to fig. 3, and the method comprises the following substeps:
step 11, acquiring radio signal data at different times:
initiating scanning of airborne multi-source radio signalsChannel of (1), set NcThe channel number of the wireless electromagnetic signal is
After the number of channels of each radio signal is determined, recording the strength values of the multi-source radio signals collected at different time at each preset position; since the air radio signals are ubiquitous, there are a large number of radio signals both indoors and outdoors. The main types of these radio signals include broadcast television signals of television stations, broadcast signals of broadcast stations, communication signals transmitted by mobile communication base stations, radar signals, and the like. Compared with indoor Wi-Fi signals, digital broadcast television signals, broadcast station signals, mobile communication base station signals and the like can provide wider coverage range than the Wi-Fi signals, and the frequency bands of the radio signals are licensed, so that the radio signals can be prevented from co-channel interference emitted by surrounding equipment, and have the characteristics of less interference and higher positioning accuracy;
carrying out random sampling on N data samples of the collected multi-source radio signal strength value for K times to obtain K data sets; wherein, the method of each random sampling comprises the following steps: randomly extracting N data samples from the collected multi-source radio signal intensity value, wherein N is less than or equal to N; then, for each data sample of the randomly extracted n data samples at aiRandomly extracting m data from the data collected by each channel, wherein m is less than or equal to ai，i∈NCThus, n × m dimensional data is obtained, and then the n × m dimensional data is spliced together as a new data set, which is recorded as:
k data sets obtained by K random sampling are recorded as: RSS ═ RSS1 RSS2 … RSSK}。
Step two, calculating a cross-time geodesic flow core:
when RSS data for a multi-source radio signal is collected at different times, the distribution of the RSS data at different points in time changes due to the time-varying nature of the signal. In order to reduce the influence of positioning accuracy caused by data distribution change, invariance characteristics of data between different times need to be obtained. Because migration between multiple times is needed, multiple geodesic flow cores need to be calculated, namely, the cross-time geodesic flow core fusion method (mkkfk) proposed by the present invention. Therefore, in the second step, the geodesic flow kernel and the data distribution difference metric are respectively calculated for the plurality of data sets obtained in the first step, and the geodesic flow kernels at different times are fused by adopting the data distribution difference metric to obtain the cross-time geodesic flow kernel, referring to fig. 4, which includes the following sub-steps:
step 21, constructing a source domain and a target domain:
setting a Time set as Time { Time ] for a plurality of data sets RSS acquired at different times in the first step1,time2,…,timeRR is the number of time set elements;
a pair of source domain time and target domain time (time) is formed between different elements in the time sets,timet) Wherein S and T represent subscripts corresponding to the source domain S and the target domain T, respectively;
data sets RSS in a plurality of data sets RSS acquired at different times in the step onekCorresponds to times,timetAs the source domain S and the target domain T.
Step 22, constructing a geodesic flow:
note the book
Is the base of the subspace of the source domain S and the target domain T in the ith data set, and all the subspaces form Grassmann manifold
The method for constructing a geodesic flow in step 22 comprises the following steps:
(1) solving the S subspace base of the source domain
Of orthogonal complement
(2) By a pair of formulas
Performing singular value decomposition to obtain an orthogonal matrix
And diagonal matrix Γi(t),Σi(t) its diagonal element is cos θa,sinθa(ii) a Wherein a is 1,2, …, d, thetaaIs PS,PTThe angle of the main angle of (a) is,
they measure the coverage angle between subspaces;
(3) the starting point and the end point of the source domain S and the target domain T as the geodesic lines are expressed as follows:
connection of
The geodesic lines of (a) are expressed as:
wherein t is ∈ [0, 1]]。
Step 23, calculating a geodesic flow kernel:
(1) will Si,TiIs projected to phii(t) obtaining their infinite dimensional subspace
(2) Computing
Inner product of (2)Obtaining a geodesic flow kernel:
(5) geodesic core GiThe closed-form solution of (c) can be obtained from the result in step 22:
wherein, Λi,1To Λi,3For a diagonal matrix, the diagonal elements are:
step 24, calculating a data distribution difference measure:
the RSS data distribution differences of a plurality of data sets acquired at different times are different, and the importance degree of the difference degree of two domains to the final core needs to be considered when the geodesic flow core is fused. The data distribution difference metric is a measure of the difference between two data distributions, i.e., whether two data distributions are similar to each other. The method for calculating the data distribution difference metric in step 24 comprises the following sub-steps:
(1) calculating all joint distribution sets psi (S, T) taking S, T as edge distribution;
(2) sample (x, y) from the joint distribution set Ψ (S, T): μ, calculating the distance ρ (x, y) between samples x and y and the expectation of the distance for the samples under μ for the joint distribution, where p represents the p-norm:
(3) computing the desired infimum bound of distance, i.e. the measure of difference between two data distributions, in all sets of joint distributions Ψ (S, T)
the multi-core model has higher flexibility than the single-core model. The high-dimensional space after being mapped by the plurality of kernel functions is a combined space formed by combining a plurality of feature spaces. The combined space can combine different feature mapping capabilities of each subspace, different feature components in heterogeneous data can be mapped through the most appropriate single kernel function, and finally the data can be more accurately and reasonably expressed in a new combined space, so that the classification accuracy or prediction accuracy of sample data is improved. Measure of difference between the data distributions obtained according to step 24
Weighting and adding all the geodesic flow cores to obtain the cross-time geodesic flow core
Step three, integrating classifiers:
the third step is to train a plurality of base classifiers after transforming the data in the plurality of data sets obtained in the first step through the cross-time geodesic flow kernel calculated in the second step, and to group the plurality of base classifiers into an inter-group classifier and an intra-group classifier, referring to fig. 5, including the following substeps:
step 31, in the step one, obtaining K data sets RSS { RSS by randomly sampling the multi-source radio signal intensity value for K times1 RSS2 … RSSKAfter the calculation, calculating cross-time geodesic flow kernels G of the K data sets according to the method of the step twok,k＝1,2…K；
Step 32, grouping the reference points for acquiring the radio signals, roughly positioning between groups, and positioning in detail in the groups;
step 33, with GkTransforming the grouped source domain S, the transformed data using a base classifier model fk(Gk,Xk,Yk) Carry out training, Xk，YkGrouping training data and labels for the kth round; the final inter-group classifier consists of K base classifiers;
at step 34, the data in each group is trained in the same manner as in step 33 to obtain the intra-group basis classifiers.
Step four, integrating classifier fingerprint positioning:
after multi-source radio signal data collected in real time are processed and transformed in the first step and the second step, prediction is carried out through the inter-group classifier and the intra-group classifier obtained in the third step, a prediction result is voted, and the final positioning result with the largest number of votes is obtained, and the method comprises the following sub-steps of:
step 41, collecting multi-source radio signal data in real time;
step 42, fusing multi-source radio signal data acquired in real time according to the method in the step one, and then taking a plurality of obtained data sets as target domain data;
step 43, transforming the target domain data by using the cross-time geodesic flow core obtained by the calculation in the step two;
step 44, classifying and predicting the transformed target domain data through the inter-group base classifiers obtained in the step three, voting prediction results of all the inter-group classifiers, and taking the most voted result as a rough positioning result;
and step 45, performing classification prediction by using the corresponding intra-group classifiers according to the rough positioning result, voting the prediction results of all the intra-group classifiers, and taking the most voted result as the final positioning result.
Example (c):
the experimental simulation takes an office building data set as an example, and the office building has three layers, namely-2 layers, 0 layer and 1 layer.
Step one, in the multi-source wireless electromagnetic signal data fusion, by taking the collection of a broadcast television signal (DVB-T), a frequency modulation broadcast signal (FM) and a cellular base station signal (GSM) as an example, RSS values of the DVB-T, FM and GSM signals collected in 23 time periods in one year are collected. The frequency range of the DVB-T signal is 498-602 MHz, the channel width is 8MHz, and the DVB-T signal is divided into 6 channels. The frequency range of the FM signal is 87.5-108.5 MHz, 100kHz is taken as a frequency band, and the frequency band is divided into 210 frequency bands. The frequency range of GSM is 925-960MHz, 200kHz is taken as one frequency band, and 175 frequency bands are divided in total. Therefore, the number of DVB-T signal strength values is 6, the number of FM signal strength values is 210, the number of GSM signal strength values is 175, and 391 signal strength values are obtained in total, and the three collected data are combined after 15 rounds of random sampling, and each round of random sampling combination is shown in fig. 7.
And step two, in the calculation of the cross-time geodesic flow kernel, constructing source domain and target domain data for the data acquired in 23 time intervals respectively and calculating a fusion cross-time geodesic flow kernel of the source domain and the target domain data.
And step three, dividing the-2-layer reference points of the office building into four groups in the classifier integration, dividing the 0 layer into four groups and dividing the 1 layer into three groups, and taking a decision tree as a base classifier. The maximum number of partitions of the decision tree is 54, the minimum number of leaf nodes is 1, and 15 rounds of training are performed. And training the grouped source domain data to respectively obtain 15 inter-group decision tree-based classifiers and 15 intra-group decision tree-based classifiers.
And step four, predicting target domain data by the trained 15 inter-group basis classifiers for the three-floor building in the integrated classifier positioning, voting and counting the obtained result, and taking the result with the largest number of votes as a final positioning result.
Experiments respectively compare the distance error results of the Multi-Kernel Geodesic Flow Kernel algorithm and the decision tree method, and the TCA, JDA and BDA methods provided by the invention. The distance errors of the office building-2 floor, 0 floor and 1 floor are respectively shown in fig. 8, 9 and 10. The experimental result shows that the cross-time multi-measurement ground flow core fusion MKGFK method provided by the invention has better performance than that of a decision tree method and other migration learning methods which are directly used. The distance errors of positioning of the-2 layer, the 0 layer and the 1 layer are respectively improved by 16.61%, 12.93% and 26.07% when the accumulated distribution is 90%. The cumulative distribution at 90% in the comparison to the distance error between TCA increased by 6.19%, 14.85%, 15.84%, respectively. The distance error compared with JDA is increased by 16.61%, 19.08% and 35.95% when the cumulative distribution is 90%. The distance error compared with BDA is increased by 16.61%, 19.08% and 34.05% when the cumulative distribution is 90%. Therefore, the invention can obviously improve the positioning accuracy and smooth the positioning errors at different moments: compared with the traditional method for fingerprint positioning by using wireless electromagnetic signals such as Wi-Fi and Bluetooth from a single source, the multi-source radio signal data fusion greatly improves the number of the fingerprint characteristics of the radio signals; the cross-time geodesic flow core fusion method takes invariance of fingerprint signal characteristics between different times into consideration, and enhances stability of radio signal fingerprint positioning influenced by time variation; the sample and characteristic diversity brought by the classifier integration can well improve the classification accuracy and improve the positioning precision.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A time-crossing multi-source radio signal positioning method based on geodesic flow kernel transfer learning is characterized by comprising the following steps:
step one, multi-source radio signal data fusion: carrying out multiple random sampling on the collected multi-source radio signal intensity values to obtain a plurality of data sets;
step two, calculating a cross-time geodesic flow core: respectively calculating a geodesic flow kernel and data distribution difference measurement for the plurality of data sets obtained in the step one, and fusing the geodesic flow kernels at different times by adopting the data distribution difference measurement to obtain a cross-time geodesic flow kernel;
step three, integrating classifiers: transforming the data in the plurality of data sets obtained in the step one through the cross-time geodesic flow kernel calculated in the step two, training a plurality of base classifiers, and grouping the plurality of base classifiers into an inter-group classifier and an intra-group classifier;
step four, integrating classifier fingerprint positioning: and (4) collecting multi-source radio signal data in real time, processing and converting the data in the first step and the second step, predicting through the inter-group classifier and the intra-group classifier obtained in the third step, voting on a prediction result, and obtaining a final positioning result with the largest number of votes.
2. The cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning of claim 1, wherein step one comprises the following sub-steps:
step 11, acquiring radio signal data at different times:
begin scanning channels of airborne multi-source radio signals, set NcThe channel number of the wireless electromagnetic signal is
After the number of channels of each radio signal is determined, recording the strength values of the multi-source radio signals collected at different time at each preset position;
step 12, constructing fingerprint features through random sampling and fusing the fingerprint features:
carrying out random sampling on N data samples of the collected multi-source radio signal strength value for K times to obtain K data sets; wherein, the method of each random sampling comprises the following steps: randomly extracting N data samples from the collected multi-source radio signal intensity value, wherein N is less than or equal to N; then, for each data sample of the randomly extracted n data samples at aiRandomly extracting m data from the data collected by each channel, wherein m is less than or equal to ai，i∈NCThus, n × m dimensional data is obtained, and then the n × m dimensional data is spliced together as a new data set, which is recorded as:
k data sets obtained by K random sampling are recorded as: RSS ═ RSS1 RSS2 … RSSK}。
3. The cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning of claim 2, wherein step two comprises the following sub-steps:
step 21, constructing a source domain and a target domain;
step 22, constructing a geodesic flow;
step 23, calculating a geodesic flow kernel:
step 24, calculating a data distribution difference measure:
and step 25, fusing a plurality of geodesic flow cores at different times.
4. The method of claim 3, wherein the step 21 comprises: setting a Time set as Time { Time ] for a plurality of data sets RSS acquired at different times in the first step1，time2，…，timeRR is the number of time set elements; a pair of source domain time and target domain time (time) is formed between different elements in the time sets，timet) Wherein S and T represent subscripts corresponding to the source domain S and the target domain T, respectively; data sets RSS in a plurality of data sets RSS acquired at different times in the step onekCorresponds to times，timetAs the source domain S and the target domain T.
5. The method of claim 4, wherein the cross-time multi-source radio signal location method based on geodesic flow kernel migration learning is characterized by being implemented by
Is the base of the subspace of the source domain S and the target domain T in the ith data set, and all the subspaces form Grassmann manifold
This step 22 constructs a geodesic surveyThe method of line flow comprises the steps of:
(1) solving the S subspace base of the source domain
Of orthogonal complement
(2) By a pair of formulas
Singular value decomposition is carried out to obtain an orthogonal matrix Ui，1∈Rd×d，Ui，2∈R(D-d)×dAnd diagonal matrix Γi(t)，∑i(t) its diagonal element is cos θa，sinθa(ii) a Wherein, a is 1,2aIs PS，PTThe angle of the main angle of (a) is,
they measure the coverage angle between subspaces;
(3) the starting point and the end point of the source domain S and the target domain T as the geodesic lines are expressed as follows:
connection of
The geodesic lines of (a) are expressed as:
wherein t is ∈ [0, 1]]。
6. The method of claim 5, wherein the method of step 23 is:
(1) will Si，TiIs projected to phii(t) obtaining their infinite dimensional subspace
(2) Computing
The inner product of (a) yields the geodesic flow kernel:
(3) geodesic core GiThe closed-form solution of (c) can be obtained from the result in step 22:
wherein, Λi，1To Λi，3For a diagonal matrix, the diagonal elements are:
7. the method of claim 6, wherein the method of step 24 is:
(1) calculating all joint distribution sets psi (S, T) taking S, T as edge distribution;
(2) sample (x, y) from the joint distribution set Ψ (S, T): μ, calculating the distance ρ (x, y) between samples x and y and the expectation of the distance for the samples under μ for the joint distribution, where p represents the p-norm:
(3) computing the desired infimum bound of distance, i.e. the measure of difference between two data distributions, in all sets of joint distributions Ψ (S, T)
8. The method of claim 7, wherein the method of step 25 is: measure of difference between the data distributions obtained according to step 24
Weighting and adding all the geodesic flow cores to obtain the cross-time geodesic flow core
9. The cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning of claim 8, wherein step three comprises the following sub-steps:
step 31, in the step one, obtaining K data sets RSS { RSS by randomly sampling the multi-source radio signal intensity value for K times1 RSS2 … RSSKAfter the calculation, calculating cross-time geodesic flow kernels G of the K data sets according to the method of the step twok，k＝1，2...K；
Step 32, grouping the reference points for acquiring the radio signals, roughly positioning between groups, and positioning in detail in the groups;
step 33, with GkTransforming the grouped source domain S, the transformed data using a base classifier model fk(Gk，Xk，Yk) Carry out training, Xk，YkGrouped training number for k-th roundAccording to the label; the final inter-group classifier consists of K base classifiers;
at step 34, the data in each group is trained in the same manner as in step 33 to obtain the intra-group basis classifiers.
10. The cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning of claim 9, wherein step four comprises the sub-steps of:
step 41, collecting multi-source radio signal data in real time;
step 42, fusing multi-source radio signal data acquired in real time according to the method in the step one, and then taking a plurality of obtained data sets as target domain data;
step 43, transforming the target domain data by using the cross-time geodesic flow core obtained by the calculation in the step two;
step 44, classifying and predicting the transformed target domain data through the inter-group base classifiers obtained in the step three, voting prediction results of all the inter-group classifiers, and taking the most voted result as a rough positioning result;
and step 45, performing classification prediction by using the corresponding intra-group classifiers according to the rough positioning result, voting the prediction results of all the intra-group classifiers, and taking the most voted result as the final positioning result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011527510.6A CN112731285B (en) | 2020-12-22 | 2020-12-22 | Cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011527510.6A CN112731285B (en) | 2020-12-22 | 2020-12-22 | Cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112731285A true CN112731285A (en) | 2021-04-30 |
| CN112731285B CN112731285B (en) | 2023-12-08 |
Family
ID=75605656
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011527510.6A Active CN112731285B (en) | 2020-12-22 | 2020-12-22 | Cross-time multi-source radio signal positioning method based on geodesic flow kernel migration learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112731285B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160078359A1 (en) * | 2014-09-12 | 2016-03-17 | Xerox Corporation | System for domain adaptation with a domain-specific class means classifier |
| CN106599922A (en) * | 2016-12-16 | 2017-04-26 | 中国科学院计算技术研究所 | Transfer learning method and transfer learning system for large-scale data calibration |
| CN109348410A (en) * | 2018-11-16 | 2019-02-15 | 电子科技大学 | Indoor orientation method based on global and local joint constraint transfer learning |
| US20200211284A1 (en) * | 2018-12-28 | 2020-07-02 | National Tsing Hua University | Indoor scene structural estimation system and estimation method thereof based on deep learning network |
| CN111625441A (en) * | 2019-02-27 | 2020-09-04 | 中国矿业大学 | Unsupervised heterogeneous defect prediction method based on geodesic flow kernel |
| CN111885703A (en) * | 2020-07-21 | 2020-11-03 | 电子科技大学 | Indoor positioning method |
-
2020
- 2020-12-22 CN CN202011527510.6A patent/CN112731285B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160078359A1 (en) * | 2014-09-12 | 2016-03-17 | Xerox Corporation | System for domain adaptation with a domain-specific class means classifier |
| CN106599922A (en) * | 2016-12-16 | 2017-04-26 | 中国科学院计算技术研究所 | Transfer learning method and transfer learning system for large-scale data calibration |
| CN109348410A (en) * | 2018-11-16 | 2019-02-15 | 电子科技大学 | Indoor orientation method based on global and local joint constraint transfer learning |
| US20200211284A1 (en) * | 2018-12-28 | 2020-07-02 | National Tsing Hua University | Indoor scene structural estimation system and estimation method thereof based on deep learning network |
| CN111625441A (en) * | 2019-02-27 | 2020-09-04 | 中国矿业大学 | Unsupervised heterogeneous defect prediction method based on geodesic flow kernel |
| CN111885703A (en) * | 2020-07-21 | 2020-11-03 | 电子科技大学 | Indoor positioning method |
Non-Patent Citations (4)
| Title |
|---|
| BOQING GONG: "Geodesic flow kernel for unsupervised domain adaptation", 2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION * |
| 刘鑫鹏;栾悉道;谢毓湘;黄明哲;: "迁移学习研究和算法综述", 长沙大学学报, no. 05 * |
| 张笑凯: "基于迁移学习和指纹的室内定位算法", 中国硕士学位论文全文数据库 信息科技辑 * |
| 李冬;张宝贤;: "基于指纹的室内定位技术", 中兴通讯技术, no. 06 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112731285B (en) | 2023-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ibrahim et al. | CNN based indoor localization using RSS time-series | |
| CN104185275B (en) | A kind of indoor orientation method based on WLAN | |
| Elbakly et al. | TrueStory: Accurate and robust RF-based floor estimation for challenging indoor environments | |
| CN109672973B (en) | Indoor positioning fusion method based on strongest AP | |
| CN110536245B (en) | Deep learning-based indoor wireless positioning method and system | |
| Cengiz | Comprehensive analysis on least-squares lateration for indoor positioning systems | |
| Adege et al. | Applying Deep Neural Network (DNN) for large-scale indoor localization using feed-forward neural network (FFNN) algorithm | |
| Ning et al. | Outdoor location estimation using received signal strength-based fingerprinting | |
| Zhao et al. | GraphIPS: Calibration-free and map-free indoor positioning using smartphone crowdsourced data | |
| Yang et al. | Research on Wi-Fi indoor positioning in a smart exhibition hall based on received signal strength indication | |
| CN104581945B (en) | The WLAN indoor orientation methods of semi-supervised APC clustering algorithms based on distance restraint | |
| Siyang et al. | WKNN indoor Wi-Fi localization method using k-means clustering based radio mapping | |
| CN112616184B (en) | Mobile equipment position estimation method based on multi-base station channel state information fusion | |
| CN113596989A (en) | Indoor positioning method and system for intelligent workshop | |
| Tarekegn et al. | D fops: Deep-learning-based fingerprinting outdoor positioning scheme in hybrid networks | |
| Zhang et al. | WiFi fingerprint positioning based on clustering in mobile crowdsourcing system | |
| Zhang et al. | Feature fusion using stacked denoising auto-encoder and GBDT for Wi-Fi fingerprint-based indoor positioning | |
| Anisetti et al. | Landmark-assisted location and tracking in outdoor mobile network | |
| Nguyen | A performance guaranteed indoor positioning system using conformal prediction and the WiFi signal strength | |
| Alitaleshi et al. | Affinity propagation clustering-aided two-label hierarchical extreme learning machine for Wi-Fi fingerprinting-based indoor positioning | |
| Tarekegn et al. | Applying long short-term memory (LSTM) mechanisms for fingerprinting outdoor positioning in hybrid networks | |
| Mukhtar et al. | Machine learning-enabled localization in 5g using lidar and rss data | |
| Soro et al. | Performance comparison of indoor fingerprinting techniques based on artificial neural network | |
| Li et al. | Outdoor location estimation using received signal strength feedback | |
| Zhang et al. | Towards floor identification and pinpointing position: A multistory localization model with wifi fingerprint |
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 |