CN107124761B - Cellular network wireless positioning method fusing PSO and SS-ELM - Google Patents

Cellular network wireless positioning method fusing PSO and SS-ELM Download PDF

Info

Publication number
CN107124761B
CN107124761B CN201710140253.2A CN201710140253A CN107124761B CN 107124761 B CN107124761 B CN 107124761B CN 201710140253 A CN201710140253 A CN 201710140253A CN 107124761 B CN107124761 B CN 107124761B
Authority
CN
China
Prior art keywords
elm
training data
pso
particle
iteration
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.)
Active
Application number
CN201710140253.2A
Other languages
Chinese (zh)
Other versions
CN107124761A (en
Inventor
刘发贵
覃亨锐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201710140253.2A priority Critical patent/CN107124761B/en
Publication of CN107124761A publication Critical patent/CN107124761A/en
Application granted granted Critical
Publication of CN107124761B publication Critical patent/CN107124761B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • H04W64/006—Locating users or terminals or network equipment for network management purposes, e.g. mobility management with additional information processing, e.g. for direction or speed determination
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]

Abstract

The invention discloses a cellular network wireless positioning method fusing PSO and SS-ELM. The implementation of the invention comprises the steps of training the weight parameter beta of the output layer of the SS-ELM by using the labeled training data and the unlabeled training data, introducing PSO to automatically optimize the hyperparameter of the SS-ELM in the training process of the SS-ELM, and optimizing and screening the SS-ELM by using the labeled training data and the unlabeled training data in the training process of the fitness value calculation function of the PSO to obtain the optimal SS-ELM parameter as a regression model for the online positioning service. The invention discloses a separation line and an online two-part implementation. The invention reduces the dependence of cellular network positioning based on RSS fingerprint data on the labeled RSS fingerprint data, reduces the cost of manually acquiring data, and reduces the workload of manually adjusting parameters in the algorithm training process.

Description

Cellular network wireless positioning method fusing PSO and SS-ELM
Technical Field
The invention belongs to a wireless positioning method of a cellular network in the intelligent field of pattern recognition and calculation, and particularly relates to Particle Swarm Optimization (PSO) and a Semi-Supervised Extreme Learning Machine (SS-ELM).
Background
The highly developed cellular network system and cellular network signals covering the world make the cellular network system the most widely used mobile communication system, and the popularization of smart phones makes the cellular network system-based positioning technology an important outdoor positioning technology. Especially in situations where satellite positioning systems, such as the Global Positioning System (GPS), are not available, smartphones can only rely on cellular network systems for outdoor positioning. Meanwhile, with the development of the internet of things technology, more and more intelligent devices are connected to the cellular network, and obtaining the position information of the intelligent devices becomes a prerequisite condition of various application scenarios.
Compared with traditional positioning technologies based on geometrical distance, such as TOA, TDOA, AOA and the like, Received Signal Strength (RSS) and machine learning algorithms are more suitable for positioning mobile devices in a radio signal non-line-of-sight (NLOS) environment. However, because of the acquisition of RSS fingerprint data required for supervised machine learning algorithm training, i.e., "labeled training data" with location information and Received Signal Strength (RSS), it is necessary to acquire the received signal strength and corresponding location information (two-dimensional longitude and latitude or one-dimensional location area designation) in a target area to be located by a mobile signal acquisition device, and it takes more time and labor cost to collect sufficient RSS fingerprint data especially in a wide outdoor environment. On the contrary, the price for acquiring the non-tag Received Signal Strength (RSS) data without the location information is very little, and a large amount of Received Signal Strength (RSS) data can be extracted from mr (measurement report) data uploaded by the mobile phones in the target service area, but the data do not contain the location information and are called as 'non-tag training data'. In the field of pattern recognition, semi-supervised learning is adopted, label-free data is complementary to labeled training data, and the accuracy of algorithm prediction and recognition can be improved by the label-free data under the condition of less labeled training data.
The Semi-supervised extreme learning machine (SS-ELM) is a neural network with a single hidden layer, has the characteristics of high training speed and good generalization capability, and can be used for training by combining labeled training data and unlabeled training data. But SS-ELM pairs
Figure BDA0001242582050000011
Norm and manifold regularization constraints are very sensitive, no systematic theory is provided to guide us to optimize constrained hyper-parameters in practical application, and the SS-ELM training process only trains and optimizes the output layer weight beta and no systematic theory is provided to guide SS-ELM hyper-parameter optimization, so that the SS-ELM hyper-parameter optimization can be performed repeatedly only by experienced workers according to specific service scenes.
Disclosure of Invention
The implementation scheme of the invention is divided into two steps: the method aims to solve the problem that more labeled training data need to be acquired in positioning based on RSS fingerprint data in a cellular network environment, and less labeled training data need to be acquired by SS-ELM under the condition of realizing the same positioning precision by using the SS-ELM, so that the cost for manually collecting the labeled training data is reduced. Meanwhile, the invention combines SS-ELM and PSO, and utilizes PSO to automatically optimize the hyper-parameters of SS-ELM, thereby reducing manual intervention in algorithm parameter tuning and improving the efficiency of production application. The invention is realized by the following technical scheme.
The cellular network wireless positioning method fusing the PSO and the SS-ELM comprises the steps of training an output layer weight parameter beta of the SS-ELM by using labeled training data and unlabeled training data, automatically optimizing a hyper-parameter of the SS-ELM by using the PSO in the SS-ELM training process, calculating an adaptability value calculation function of the PSO, covering the labeled training data and the unlabeled training data, and using an optimal SS-ELM parameter obtained after PSO optimization screening as a regression model for online positioning service.
Further, the device Received Signal Strength (RSS) is used as an input, the SS-ELM is trained and optimized to obtain SS-ELM parameters, and the SS-ELM parameters are used as a regression model to provide online positioning service in a cellular network environment.
Further, the labeled training data and the unlabeled training data are used simultaneously in the training process of the weight parameter beta of the SS-ELM output layer.
Further, the calculation of the fitness value calculation function of the PSO algorithm covers the labeled training data and the unlabeled training data, and the particle fitness calculation function is specifically defined as:
Figure BDA0001242582050000021
wherein i, j, k and N respectively represent the number of the particles, the number of the example of the labeled training data, the number of the dimension of the label vector and the total dimension of the label vector, l is the number of the labeled training data in the training data set, u is the number of the unlabeled training data in the training data set,
Figure BDA0001242582050000022
a label prediction value representing the SS-ELM predicting the labeled training data on the training data set, Y representing the actual value of the label of the labeled training data,
Figure BDA0001242582050000023
the proportion of extreme values generated by the prediction of the SS-ELM on the training set for each iteration of training.
Further, the key parameters of the PSO to the SS-ELM are used in the SS-ELM training and optimization process:
Figure BDA0001242582050000024
norm over parameter cβAdjusting and optimizing a manifold regularization hyper-parameter lambda; PSO vs SS-ELM hyperparameter cβThe optimization calculation mode of lambda is as follows: the PSO randomly generates particles in a specified search space, and each particle moves to a position where an optimal solution is searched globally at a certain speed; in each iteration of the particle swarm optimization, each particle is positioned according to the momentum and the optimal position PbAnd global optimum PgThe influence factor of the position adjusts the speed of the user, and the position of the user in the iteration is calculated; the dimension of the search space of the particle swarm is 2, the total number of the particle swarm is n, and a certain particle ithThe position of each iteration can be represented as a vector Xi＝(xi1,xi2),xi1∈[-10,0],xi2∈[-10,0](ii) a The relationship between particle position and hyper-parameter is expressed as
Figure BDA0001242582050000031
Figure BDA0001242582050000032
The individual optimal position of the particle from the beginning of the search to the present is denoted as Pib＝(pi1,pi2) The velocity of movement of the particles being expressed as a vector Vi＝(vi1,vi2)g＝(pg1,pg2) The particle updates the current velocity and position at each iteration and uses a moderating function f (X) based on the current position of the particlei) Calculating the fitness of the particles, and updating the individual optimal position and the global optimal position; the updating of the PSO variables in an iteration can be formulated as follows:
Figure BDA0001242582050000033
Figure BDA0001242582050000034
vid(t+1)＝vid(t)+c1*rand()*[pid(t)-xid(t)]+c2*and()*[pgd(t)-Xid(t)] (4)
xid(t+1)＝xid(t)+vid(t+1)1≤i≤n,1≤d≤2 (5)
Figure BDA0001242582050000035
wherein a positive number c1,c2Is an acceleration factor, and rand () is a random number between 0 and 1; v. ofmaxAnd vminThe upper bound and the lower bound of the particle speed are respectively, and t is an algebra of algorithm iteration; and stopping iterative computation when the convergence of the particle fitness corresponding to the global optimal position does not change along with the iteration in the iteration process, and obtaining the optimal hyperparameter through the corresponding relation between the global optimal particle position and the hyperparameter.
Compared with the prior art, the invention has the beneficial effects that:
1) compared with a supervised learning method, the method has the advantage that better positioning accuracy can be obtained by using the semi-supervised learning algorithm SS-ELM under the condition of less training data.
2) The SS-ELM is sensitive to selection of the hyper-parameters, and usually needs experienced technicians to repeatedly adjust the parameters according to different service scenes.
3) Compared with the traditional PSO which only uses label training data, the PSO of the invention carries out iterative training based on label training data and label-free training data, and adopts a formula (1) as a fitness value calculation function to ensure that the SS-ELM optimized by the PSO is more robust and stable and the positioning of a user is more accurate.
Drawings
FIG. 1 is a flow chart of an off-line training portion of an embodiment of the method of the present invention.
FIG. 2 is a flow chart of the online prediction portion of the implementation of the method of the present invention.
FIG. 3 is a diagram of a SS-ELM neural network topology used in the present invention.
FIG. 4 is a diagram illustrating a variation trend of the PSO fitness value in the iterative training according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided in connection with the accompanying drawings and examples, but the invention is not limited thereto. The invention is divided into two major steps by the following embodiments:
the method comprises the steps that firstly, a model of the SS-ELM algorithm is subjected to offline training by fusing PSO (particle swarm optimization) by using part of labeled training data and unlabeled training data, and different from a traditional manual parameter adjusting method, the fused PSO and SS-ELM aim to search and screen out the optimal super-parameter of the SS-ELM in different service scenes by utilizing the PSO, and the optimal model is obtained.
And secondly, carrying out online positioning on the equipment of the user by utilizing the optimal SS-ELM model.
The key calculation method for SS-ELM and PSO training in the first step is introduced as follows:
suppose that a given set of l tagged RSS fingerprint data samples { (r)k,yk)|k＝1,2,3,…,l}lAnd u sets of unlabeled RSS data samples, { rk|k＝1,2,3,…,u}uWherein the r component is Received Signal Strength (RSS), the y component is position information, and a sigmoid function is selected as a hidden layer activation function of the SS-ELM, so that the training requirement of the SS-ELM meets the following objective function:
Figure BDA0001242582050000041
s.t.F＝Hβ
wherein
Figure BDA0001242582050000042
Figure BDA0001242582050000043
Figure BDA0001242582050000044
a is the weight of the input layer connecting the hidden layer, b is the bias of the hidden layer, beta is the weight of the SS-ELM connecting the hidden layer and the output layer, cβAnd λ respectively correspond to
Figure BDA0001242582050000046
Norm and the hyperparameter of the manifold regularization term. CeIs a weight matrix of (l + u) × (l + u) [ Ce]jj1, j-1, 2 … l, the remaining elements being equal to zero, l and u being the number of labeled and unlabeled training data, respectively. Tr (-) denotes the trace of the matrix, L is the Laplace matrix, and D is the diagonal matrix, where. w is aijIs xiAnd xjSimilarity of attributes therebetween:
Figure BDA0001242582050000045
Figure BDA0001242582050000051
L＝D-W(12)
since the extreme learning machine only needs to train and correct the weight β between the output layer and the hidden layer, the training of SS-ELM is equivalent to solving for the weight β satisfying equation (7). Therefore, taking the derivative of equation (7) to 0 yields the following equation:
Figure BDA0001242582050000052
if the number of rows of the matrix H is greater than the number of columns
Figure BDA0001242582050000053
The approximate solution of β in equation (13) is:
Figure BDA0001242582050000054
if the number of columns of H is greater than the number of rows
Figure BDA0001242582050000055
β*＝HT(cβI(l+u)×(1+u)+CeHHT+λLHHT)-1CeT (15)
Wherein I is
Figure BDA0001242582050000056
A dimension unit matrix.
PSO randomly generates particles in a predetermined search space, and each particle moves to a position where an optimal solution is globally searched at a certain speed. Each particle will have its own momentum and its own optimal position (P) in each iteration of the PSOb) And global optimum (P)g) The influence factors of the position adjust the speed of the user, and the position of the user in the iteration is calculated. Suppose that the dimension of our search space is D, the total number of particle groups is n, and a particle ithThe position of each iteration can be represented as a vector Xi＝(xi1,xi2,…,xiD) The individual optimal position of the particle from the beginning of the search to the present is denoted as Pib＝(pi1,pi2,…,piD) The velocity of movement of the particles being expressed as a vector Vi＝(vi1,vi2,…,viD) Global optimum position Pg＝(pg1,pg2,…,pgD) The particle updates the current velocity and position at each iteration and uses a moderating function f (X) based on the current position of the particlei) A gracefully-updated individual optimal position and a global optimal position of the particle are computed. The updating of the PSO variables in an iteration can be represented by the following equations:
Figure BDA0001242582050000057
Figure BDA0001242582050000058
vid(t+1)＝vid(t)+c1*rand()*[pid(t)-xid(t)]+c2*rand()*[pgd(t)-Xid(t)] (18)
xid(t+1)＝xid(t)+vid(t+1)1≤i≤n,1≤d≤D(19)
Figure BDA0001242582050000059
wherein a positive number c1,c2Is an acceleration factor, and rand () is a random number between 0 and 1; v. ofmaxAnd vminRespectively, the upper and lower bounds of the particle velocity, and t is the algebra of the algorithm iteration.
The embodiments of the present invention are further described below by taking a two-dimensional space as an example.
Step 1: randomly initializing a batch of n particles, each particle having an initial position Xi＝(xi1,xi2) Wherein x isi1∈[-10,0]、xi2∈[-10,0]And i represents the number of the particle. Correspondingly, one SS-ELM instance was constructed for each particle:
Figure BDA0001242582050000061
wherein, ai、bi、βi、cβi、λiRespectively representing input layer weights, hidden layer offsets, output layer weights,
Figure BDA0001242582050000062
Norm hyperparameter, and hyperparameter of manifold regularization term. Setting the number of SS-ELM input layer neurons according to input parameter dimensions
Figure BDA0001242582050000063
Setting hidden layer neuron number
Figure BDA0001242582050000064
Input layer weights a to SS-ELMiBias of the hidden layer biRandomly assigning a value range of [0,1 ]]In SS-ELM
Figure BDA0001242582050000065
The hyperparameters of the norm and manifold regularization terms are respectively expressed as an exponential representation with a base 10:
Figure BDA00012425820500000614
initializing weight matrix CeDiagonal element of [ C ]e]jj1, j 1,2 … l, j indicating the subscript of the matrix element, l indicating the number of labeled training data, matrix CeThe remaining elements are initialized to 0.
Step 2: table 1 shows an example of RSS fingerprint data for locating a target area, where RSS information of a device in the first 35 columns in table 1 is represented by a letter r, and two-dimensional position coordinates of a corresponding device in the second two columns are represented by a letter y, then the labeled training data set in this example can be represented as { (r)k,yk)|k＝1,2,3,…,l}lWhere l is the number of labeled training data. And respectively carrying out normalization processing on the RSS of the first 35 columns and the position coordinate information of the second two columns, and then calculating the hidden layer output H and the Laplace matrix L of the SS-ELM according to formulas (8) to (12).
TABLE 1 RSS data examples
Figure BDA0001242582050000066
And step 3: example the corresponding output layer weight for each SS-ELM example is calculated according to equation (14) or equation (15)
Figure BDA0001242582050000067
And the output weight of the iterative training is used for predicting on the training data set according to the following formula
Figure BDA0001242582050000068
Second, the statistical prediction set
Figure BDA0001242582050000069
Middle extreme sample ratio
Figure BDA00012425820500000610
Extreme values are defined as: for arbitrary
Figure BDA00012425820500000611
If not satisfied
Figure BDA00012425820500000612
The sample belongs to the extreme value. Assuming the number of extreme samples is k, it can be obtained
Figure BDA00012425820500000613
Finally, calculating the fitness value of each particle according to the formula (22)
Figure BDA0001242582050000071
Updating the position and speed of each particle and the variables such as the global optimum position of the individual optimum position according to the formulas (16) to (20), and recalculating the hyper-parameter c of the SS-ELM according to the updated particlesβiAnd λi。
And 4, step 4: the calculation of step 3 is repeated until the fitness value of the globally optimal particle converges and no further changes occur as shown in fig. 3.
And 5: selecting the globally optimal particle P in the above stepsgAnd corresponding SS-ELM parameters
Figure BDA0001242582050000072
Figure BDA0001242582050000073
As a final regression model for online localization.
Step 6: inputting the received signal strength information of the current equipment for the online positioning request, and obtaining the normalized received signal strength data according to the normalization method used in the step 2
Figure BDA0001242582050000074
Calculated by the following formula
Figure BDA0001242582050000075
Figure BDA0001242582050000076
To pair
Figure BDA0001242582050000077
And obtaining the final position of the equipment after inverse normalization.
The steps 1 to 5 in the above are off-line training parts, and the step 6 is an on-line positioning part. The above-mentioned procedures are preferred embodiments of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention shall be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (4)

1. The cellular network wireless positioning method fusing the PSO and the SS-ELM is characterized by comprising the steps of training an output layer weight parameter beta of the SS-ELM by using labeled training data and unlabeled training data, automatically optimizing a hyper-parameter of the SS-ELM by using the PSO in the SS-ELM training process, calculating an fitness value calculation function of the PSO, covering the labeled training data and the unlabeled training data, and using an optimal SS-ELM parameter obtained after PSO optimization screening as a regression model for online positioning service; the calculation of the fitness value calculation function of the PSO algorithm covers the labeled training data and the unlabeled training data, and the specific definition of the particle fitness calculation function is as follows:
Figure FDA0002784139740000011
wherein i, j, k and N respectively represent the number of the particles, the number of the example of the labeled training data, the number of the dimension of the label vector and the total dimension of the label vector, l is the number of the labeled training data in the training data set, u is the number of the unlabeled training data in the training data set,
Figure FDA0002784139740000012
a label prediction value representing the SS-ELM predicting the labeled training data on the training data set, Y representing the actual value of the label of the labeled training data,
Figure FDA0002784139740000013
the proportion of extreme values generated by the prediction of the SS-ELM on the training set for each iteration of training.
2. The PSO and SS-ELM converged cellular network wireless positioning method of claim 1, wherein: and training and optimizing the SS-ELM by taking the signal intensity received by the equipment as input to obtain SS-ELM parameters serving as a regression model to provide online positioning service in a cellular network environment.
3. The PSO and SS-ELM converged cellular network wireless positioning method of claim 1, wherein: and the labeled training data and the unlabeled training data are simultaneously used in the training process of the weight parameter beta of the SS-ELM output layer.
4. The PSO and SS-ELM converged cellular network wireless positioning method of claim 1, wherein:
key parameters of PSO to SS-ELM were used during SS-ELM training and optimization:
Figure FDA0002784139740000014
norm over parameter cβAdjusting and optimizing a manifold regularization hyper-parameter lambda; PSO vs SS-ELM hyperparameter cβThe optimization calculation mode of lambda is as follows: the PSO randomly generates particles in a specified search space, and each particle moves to a position where an optimal solution is searched globally at a certain speed; in each iteration of the particle swarm optimization, each particle can be positioned according to the momentum and the individual optimal position PibAnd a global optimum position PgAdjusting the speed of the user and calculating the position of the user in the iteration; the dimension of the search space of the particle swarm is 2, the total number of the particle swarm is n, and a certain particle ithThe position of each iteration can be represented as a vector Xi＝(xi1,xi2),xi1∈[-10,0],xi2∈[-10,0](ii) a The relationship between particle position and hyper-parameter is expressed as
Figure FDA0002784139740000015
The individual optimal position of the particle from the beginning of the search to the present is denoted as Pib＝(pi1,pi2) The velocity of movement of the particles being expressed as a vector Vi＝(vi1,vi2)g＝(pg1,pg2) The particle updates the current velocity and position at each iteration and uses a moderating function f (X) based on the current position of the particlei) Calculating the fitness of the particles, and updating the individual optimal position and the global optimal position; the updating of the PSO variables in an iteration can be formulated as follows:
Figure FDA0002784139740000021
Figure FDA0002784139740000022
vid(t+1)＝vid(t)+c1*rand()*[pid(t)-xid(t)]+c2*rand()
xid(t+1)＝xid(t)+vid(t+1)1≤i≤n,1≤d≤2 (5)
Figure FDA0002784139740000023
wherein a positive number c1,c2Is an acceleration factor, and rand () is a random number between 0 and 1; v. ofmaxAnd vminThe upper bound and the lower bound of the particle speed are respectively, and t is an algebra of algorithm iteration; and stopping iterative computation when the convergence of the particle fitness corresponding to the global optimal position does not change along with the iteration in the iteration process, and obtaining the optimal hyperparameter through the corresponding relation between the global optimal particle position and the hyperparameter.
CN201710140253.2A 2017-03-10 2017-03-10 Cellular network wireless positioning method fusing PSO and SS-ELM Active CN107124761B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710140253.2A CN107124761B (en) 2017-03-10 2017-03-10 Cellular network wireless positioning method fusing PSO and SS-ELM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710140253.2A CN107124761B (en) 2017-03-10 2017-03-10 Cellular network wireless positioning method fusing PSO and SS-ELM

Publications (2)

Publication Number Publication Date
CN107124761A CN107124761A (en) 2017-09-01
CN107124761B true CN107124761B (en) 2021-02-19

Family

ID=59718289

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710140253.2A Active CN107124761B (en) 2017-03-10 2017-03-10 Cellular network wireless positioning method fusing PSO and SS-ELM

Country Status (1)

Country Link
CN (1) CN107124761B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107770719A (en) * 2017-09-27 2018-03-06 无锡神探电子科技有限公司 A kind of object localization method based on drive test data and machine learning
CN109100682A (en) * 2018-08-01 2018-12-28 南京邮电大学 Localization method based on multicore extreme learning machine and received signal strength
CN109871850A (en) * 2019-01-21 2019-06-11 北京大学 A kind of classification method of the mobile lidar data based on neural network model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101820676A (en) * 2010-04-12 2010-09-01 中国科学技术大学 Sensor node positioning method
CN103002575A (en) * 2012-10-16 2013-03-27 桂林电子科技大学 Method for locating underwater wireless sensor network nodes on the basis of particle swarm optimization
CN104992184A (en) * 2015-07-02 2015-10-21 东南大学 Multiclass image classification method based on semi-supervised extreme learning machine
WO2016187746A1 (en) * 2015-05-22 2016-12-01 华为技术有限公司 Method and device for improving positioning performance of artificial neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101820676A (en) * 2010-04-12 2010-09-01 中国科学技术大学 Sensor node positioning method
CN103002575A (en) * 2012-10-16 2013-03-27 桂林电子科技大学 Method for locating underwater wireless sensor network nodes on the basis of particle swarm optimization
WO2016187746A1 (en) * 2015-05-22 2016-12-01 华为技术有限公司 Method and device for improving positioning performance of artificial neural network
CN104992184A (en) * 2015-07-02 2015-10-21 东南大学 Multiclass image classification method based on semi-supervised extreme learning machine

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于深度图像和彩色图像融合的头姿估计算法研究;王润元;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160315;全文 *

Also Published As

Publication number Publication date
CN107124761A (en) 2017-09-01

Similar Documents

Publication Publication Date Title
Song et al. A novel convolutional neural network based indoor localization framework with WiFi fingerprinting
Njima et al. Indoor localization using data augmentation via selective generative adversarial networks
CN107679557B (en) Driving model training method, driver identification method, device, equipment and medium
CN105809672B (en) A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring
CN107124761B (en) Cellular network wireless positioning method fusing PSO and SS-ELM
CN114449452B (en) Wi-Fi indoor positioning method based on CNN-RNN
Bhattarai et al. Geomagnetic field based indoor landmark classification using deep learning
CN109327797A (en) Mobile robot indoor locating system based on WiFi network signal
Li et al. Multi‐scale attention encoder for street‐to‐aerial image geo‐localization
Cui et al. Improved genetic algorithm to optimize the Wi-Fi indoor positioning based on artificial neural network
CN109143161B (en) High-precision indoor positioning method based on mixed fingerprint quality evaluation model
CN110795522A (en) Method and device for predicting track position of mobile user
Wei et al. Accurate visible light positioning technique using extreme learning machine and meta-heuristic algorithm
CN109766969B (en) RFID indoor positioning algorithm based on asynchronous dominant motion evaluation
Yang et al. Research on high precision indoor positioning method based on low power Bluetooth technology
CN117241215A (en) Wireless sensor network distributed node cooperative positioning method based on graph neural network
CN114710831B (en) RFID label positioning system based on deep learning
CN113805142B (en) Building floor indoor positioning method based on federal learning
CN109803234A (en) Unsupervised fusion and positioning method based on the constraint of weight different degree
Guo et al. A hybrid indoor positioning algorithm for cellular and Wi-Fi networks
Li et al. Finding an optimal geometric configuration for TDOA location systems with reinforcement learning
CN113923589A (en) Non-contact multi-target positioning method
CN108495262B (en) Sparse representation and matching positioning method for indoor space ubiquitous positioning signal fingerprint database
CN113011597A (en) Deep learning method and device for regression task
Wang et al. A Model Stacking Algorithm for Indoor Positioning System using WiFi Fingerprinting.

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