CN115905787B - High-precision indoor positioning method based on fuzzy migration learning model - Google Patents
High-precision indoor positioning method based on fuzzy migration learning model Download PDFInfo
- Publication number
- CN115905787B CN115905787B CN202211292065.9A CN202211292065A CN115905787B CN 115905787 B CN115905787 B CN 115905787B CN 202211292065 A CN202211292065 A CN 202211292065A CN 115905787 B CN115905787 B CN 115905787B
- Authority
- CN
- China
- Prior art keywords
- data
- fuzzy
- model
- formula
- sfs
- 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
Links
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a high-precision indoor positioning method based on a fuzzy migration learning model, which comprises the steps that a target area data acquisition module acquires a fingerprint feature set of a target area; then, the data processing module of the target area carries out data processing on the collected fingerprint feature set; then, the indoor positioning module based on the fuzzy migration model receives the data processed by the data processing module, and performs analysis and calculation according to the data to obtain target result data, wherein the target result data is positioning result data; the method not only realizes high-precision indoor positioning, but also reduces various costs.
Description
Technical Field
The invention belongs to the technical field of indoor positioning, and particularly relates to a high-precision indoor positioning method based on a fuzzy migration learning model.
Background
In recent years, with the rapid development of internet of things (internet of things, IOT) and the popularization of wireless networks (wireless networks), location-based services (location based services, LBS) are becoming deeper into the aspects of people's lives. The LBS includes calculation of the positions of target objects in different environments, and the key of the LBS is accurate acquisition of the positions of users, and is divided into outdoor positioning and indoor positioning according to application scenes.
In an outdoor environment, a global navigation satellite system (global navigation satellite system, GNSS) can achieve accurate positioning, however, the positioning granularity of the GNSS in an indoor environment is large, and the requirement of high-precision positioning in the outdoor environment cannot be met. In an indoor environment, as the wireless network is widely deployed in the indoor scene, the mode of indoor positioning by using the wireless signal is more and more concerned and applied, the mode of indoor positioning by using the wireless signal comprises a positioning method based on signal fingerprints, and the positioning method based on the signal fingerprints does not need to deploy additional equipment, so that the application is convenient, the expandability is better, and the method is always a research hot spot in the technical field of indoor positioning. The signal fingerprint-based positioning method generally comprises a fingerprint map construction process and a fingerprint matching positioning process. The fingerprint map construction stage requires offline sampling of the fingerprint database for the entire sensing area, and requires a lot of labor and time for offline sampling of the entire sensing area, so that when the sampling area is large, such as when there are multiple floors, the sampling time required for the entire sensing area increases sharply, especially when some floors or rooms are not allowed to be opened due to some special reasons during the sampling process, the fingerprint information of the areas cannot be obtained manually, thereby seriously affecting the indoor positioning result. In summary, it is therefore a real problem to be solved to reasonably control the sampling cost (including time cost, labor cost and equipment cost) and ensure the positioning accuracy.
The transfer learning in the machine learning method can realize multiplexing in similar or related areas through learning of source domain data, so that target learning becomes accumulative learning, thereby reducing the construction cost of a target domain model and improving the learning effect. The source domain-oriented construction learning model is a simple but effective (simple yet effective) migration learning mode, the source domain-oriented construction learning model is to construct an initial model by utilizing a source domain sample, store corresponding parameters, then act target domain data on the model, and conduct parameter fine adjustment so as to adapt to a data set of the model. However, in a real application scenario, due to the lack and difference of the sampling space of the target area, there is a difference in data distribution between the sampled fingerprint and the source domain fingerprint, so that the positioning model obtained based on the source domain fingerprint cannot show good performance on the target domain fingerprint, and thus indoor positioning is not accurate.
Therefore, a new indoor positioning method needs to be proposed.
Disclosure of Invention
The invention aims to: in order to overcome the defects of poor indoor positioning precision, high indoor positioning cost and the like in the prior art, the invention provides the high-precision indoor positioning method based on the fuzzy migration learning model, which not only realizes high-precision indoor positioning, but also reduces various costs including time cost, labor cost and equipment cost.
The technical scheme is as follows: in order to solve the technical problems, the invention provides a high-precision indoor positioning system based on a fuzzy migration learning model, which comprises a target area data acquisition module, a target area data processing module, an indoor positioning module based on the fuzzy migration model and a data output module which are sequentially communicated;
the target area data acquisition module is used for acquiring a fingerprint feature set of the target area;
the target area data processing module is used for carrying out data processing on the collected fingerprint feature set of the target area to obtain data suitable for processing of the next module;
the indoor positioning module based on the fuzzy migration model is used for receiving the data processed by the data processing module, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
the data output module is used for outputting positioning result data for reference of a user;
the method comprises the following steps:
step 1, a target area data acquisition module in a high-precision indoor positioning system based on a fuzzy transfer learning model acquires a fingerprint feature set of a target area;
Step 2, a target area data processing module in the high-precision indoor positioning system based on the fuzzy transfer learning model processes data aiming at the collected fingerprint feature set of the target area to obtain data suitable for processing by a next module;
step 3, receiving the data processed by the data processing module by the indoor positioning module based on the fuzzy migration model in the high-precision indoor positioning system based on the fuzzy migration learning model, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
and step 4, finally, outputting positioning result data by a data output module in the high-precision indoor positioning system based on the fuzzy transfer learning model for reference of a user.
Further, the generating the indoor positioning module based on the fuzzy migration model specifically comprises the following steps:
step S1, data acquisition is carried out on a sample area through a sample area data acquisition module, and a fingerprint feature set of the sample area is acquired;
step S2, data processing is carried out on the collected fingerprint feature set of the sample area through a sample area data processing module, so that data suitable for processing of a next module are obtained;
Step S3, dividing the fingerprint feature set of the sample area after the data processing by a sample area data sample dividing module to obtain training set sample data and test set sample data;
and S4, training the basic model based on the training set sample data to obtain a trained model, wherein the model is used as an indoor positioning module based on the fuzzy migration model and used for analyzing and calculating a data set acquired by a target area to obtain positioning result data for reference of a user.
Further, the step S2 includes the steps of:
s21, if the sampling duration of the sampling point P is Γ and the short period is τ, then in the period t, that is, in the short period τ after the time t, the RSS vector sampled by the sampling point P is expressed as formula (1):
then in the next period (t+t→t+2τ), the RSS vector sampled by the sampling point P is expressed as:
then the neighbors P of the sampling point P are sampled within a period t, i.e. a short period τ from time t i The sampled RSS vector is expressed as:
s22, calculating ISF of the sampling point P at t time through a formula (4):
ISF(P,t)＝RSS(P,t→t+τ)+RSS(P,t+τ→t+2τ)＝[ISf 1 ,ISf 2 ,…,ISf n ]formula (4)
In the above-mentioned formula (4),
s23 calculates CSF of sampling point P at time t by equation (5):
CSF(P,t)＝RSS(P,t→t+τ)+RSS(P i ,t→t+τ)＝[CSf 1 ,CSf 2 ,…,CSf n ]Formula (5)
In the above-mentioned formula (5),
s24, obtaining a short-time feature set of the sampling point P at the time T as follows: SFS (P, t) =isf (P, t)/(CSF (P, t);
s25 thus results in a short-term feature set of the sampling point P over the entire sampling period Γ expressed by formula (6):
SFS(P，Γ)＝∑ t∈Γ SFS (P, t) equation (6)
The short-time feature set SFS of the sampling points is obtained through the steps.
Further, the step S4 includes the steps of:
s41, firstly, training short-time characteristic data SFS through an Optimized long-term memory neural network to obtain an Optimized LSTM, and constructing a Pre-Model for transfer learning preliminarily;
s42, introducing the thought of an attention mechanism, and optimizing the Pre-Model of the migration learning constructed preliminarily by using a lightweight mechanism SENet facing sparse data to obtain an Optimization SE-LSTM, wherein the Optimization SE-LSTM is an optimized training Model.
Further, the step S41 includes the following steps, setting a short-time feature set Input to the LSTM network at the μmoment to sfs (μ), and obtaining an Input value by combining the Input Gate in the LSTM network with the μ -1 moment value h (μ -1) in the Memory Cell:
I(μ)＝σ(w s，I *sfs(μ)+w h，I * h (mu-1)) formula (7)
In the above formula (7), σ represents an activation function; w (w) s，I And w h，I Parameters that are a function I; sfs (mu) is a short-time feature input at mu moment, and h (mu-1) is a mu-1 moment value in the Memory Cell;
At the same time, input Gate generates a candidate vector:
in the above formula (8), w s，c And w h，c As a function ofParameters of (2); sigma represents an activation function; sfs (mu) is a short-time feature input at mu moment, and h (mu-1) is a mu-1 moment value in the Memory Cell;
then, the Forget Gate reads sfs at μ and the value h at μ -1 at μ, and outputs a value at a given interval using the activation function to indicate the acceptance of the value:
F(μ)＝σ(w s，f *sfs(μ)+w h，f * h (mu-1)) formula (9)
In the above formula (9), w s，f And w h，f Parameters that are functions F; sigma represents an activation function; sfs (mu) is a mu moment input short-time feature, and h (mu-1) is a mu-1 moment value in a Memory Cell;
next, the state information is updated according to the above-described formula (7), formula (8), and formula (9):
finally, the Output Gate outputs the result at the mu moment according to the state information, and the Memory Cell is updated at the same time, and the Memory Cell is updated through a formula (11), wherein the updating mode is as follows:
h (μ) =o (μ) ×tanh (c (μ)) formula (11)
In the above-mentioned formula (11),o (μ) is calculated by equation (12):
O(μ)＝σ(w s，O *Δsfs(μ)+w h，O * h (mu-1)) formula (12)
In the above formula (12), the vector difference Δ sfs (μ) = sfs (μ) -sfs (μ -1) at adjacent times;
the short-term characteristic data SFS is trained through the long-term and short-term memory neural network LSTM, and the Optimized LSTM is obtained and used for initially constructing the Pre-Model for migration learning.
Further, a ReLU function is employed as the activation function σ.
Further, the step S42 includes the steps of: 7. the high-precision indoor positioning method based on the fuzzy migration learning model of claim 6,
first, an embedded channel characteristic response is generated with a global distribution, allowing all layers of the training model to use, and the characteristic data is globally averaged and pooled to represent the global distribution of responses over the characteristic channels:
in the above formula (13), H represents the number of elements in the SFS, and W represents the length of each element;
then, the relation between the characteristic channels is mined, and nonlinear interaction between the channels is learned by using two layers of nonlinear activation functions so as to obtain proper weights:
S(μ)＝σ(w 2 *δ(w 1 * Z (μ))
In the above formula (14), σ represents a sigmoid activation function, δ represents a ReLU activation function, w 1 And w 2 Representing a scaling parameter;
finally, output is achieved by using channel-wise multiplication:
in the above formula (15), S (μ) is formula (14), and O (μ) is formula (12);
the pre-model O-LSTM of the migration learning which is initially constructed is optimized to obtain the OSE-LSTM by introducing a lightweight mechanism SENet facing sparse data.
Further, the OSE-LSTM is further optimized: firstly, establishing class labels for sampling points of a target domain according to the distribution condition of the sampling points in a source domain area by using a fuzzy clustering method; and then, performing transfer learning by using the target domain data marked with the class labels and the obtained pre-model PreModel of transfer learning, thereby constructing a fuzzy transfer learning model meeting different missing situations and different floors.
The beneficial effects are that: compared with the prior art, the invention has the advantages that:
1. according to the method, characteristics of sampling points are mined in a fine granularity mode, a fingerprint construction method based on Short-time features is designed, characteristics of fingerprints are displayed in a finer granularity mode, meanwhile, LSTM and SENet are optimized due to sparsity and time sequence of the Short-time features, an OSE-LSTM model is provided as a pre-model for transfer learning, sampling time in a fingerprint map construction process is shortened, and accurate positioning accuracy is guaranteed;
2. the method provides a feature migration method based on fuzzy clustering, and aims at the problem of different data distribution caused by the difference of different distribution, so that the similarity of the feature distribution of a source domain and a target domain is ensured, the robustness of migration learning is enhanced, and the indoor positioning accuracy is further improved;
3. the method provided by the invention is based on the fuzzy transfer learning model (Fuzzy Transfer Learning, FTL), realizes accurate positioning under the conditions of different floors and different sampling rates, and fully performs experimental comparison between the fuzzy transfer learning model and the prior art under different floors, different sampling rates and different devices, and the experimental result proves that the method provided by the invention has effectiveness and reliability.
Drawings
Fig. 1 is a system structure diagram based on a fuzzy migration learning model of the present invention.
FIG. 2 is a flowchart of the steps for generating an indoor positioning module based on a fuzzy migration model.
Fig. 3 is a sample area structure diagram.
FIG. 4 is a block diagram of an implementation of a Pre-Model for transfer learning.
Fig. 5 is a graph of the results of the specific experiment of the migration of the positioning model to each floor in example 9.
Fig. 6 is a graph showing experimental results when the sampling rate of the target domain was 80% in example 10.
Fig. 7 is a graph showing experimental results when the sampling rate of the target domain was set to 30% in example 10.
Fig. 8 is a graph of error experimental results of positioning results at different sampling intervals in example 11.
Detailed Description
The invention will be further described with reference to the drawings and examples.
Example 1
The high-precision indoor positioning method (Fuzzy Transfer Learning Model for Accuracy Localization) based on the fuzzy migration learning model of the embodiment provides a high-precision indoor positioning system based on the fuzzy migration learning model, and referring to fig. 1, the high-precision indoor positioning system based on the fuzzy migration learning model comprises a target area data acquisition module, a target area data processing module, an indoor positioning module based on the fuzzy migration model and a data output module which are sequentially communicated;
The target area data acquisition module is used for acquiring a fingerprint feature set of a target area;
the target area data processing module is used for performing data processing on the collected fingerprint feature set of the target area to obtain data suitable for processing of the next module;
the indoor positioning module based on the fuzzy migration model is used for receiving the data processed by the data processing module, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
the data output module is used for outputting positioning result data for reference of a user.
When the high-precision indoor positioning system based on the fuzzy transfer learning model works, a target area data acquisition module acquires a fingerprint feature set of a target area; then, the target area data processing module processes data aiming at the collected fingerprint feature set of the target area to obtain data suitable for processing of the next module; then, the indoor positioning module based on the fuzzy migration model receives the data processed by the data processing module, and performs analysis and calculation according to the data to obtain target result data, wherein the target result data is positioning result data; and finally, outputting positioning result data by a data output module for reference of a user.
The high-precision indoor positioning method based on the fuzzy migration learning model of the embodiment specifically comprises the following steps:
step 1, firstly, a target area data acquisition module in a high-precision indoor positioning system based on a fuzzy transfer learning model acquires a fingerprint feature set of a target area;
step 2, a target area data processing module in the high-precision indoor positioning system based on the fuzzy transfer learning model processes data aiming at the collected fingerprint feature set of the target area to obtain data suitable for processing by a next module;
step 3, then, receiving the data processed by the data processing module by the indoor positioning module based on the fuzzy migration model in the high-precision indoor positioning system based on the fuzzy migration learning model, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
and step 4, finally, outputting positioning result data by a data output module in the high-precision indoor positioning system based on the fuzzy transfer learning model for reference of a user.
The high-precision indoor positioning system based on the fuzzy transfer learning model can realize indoor positioning of a target area, and has high positioning precision.
Example 2
In the high-precision indoor positioning method based on the fuzzy migration learning model of the present embodiment, based on embodiment 1, the indoor positioning module based on the fuzzy migration model set forth in embodiment 1 specifically includes the following steps, referring to fig. 2:
step S1, data acquisition is carried out on a sample area through a sample area data acquisition module, and a fingerprint feature set of the sample area is acquired;
step S2, data processing is carried out on the collected fingerprint feature set of the sample area through a sample area data processing module, so that data suitable for processing of a next module are obtained;
step S3, dividing the fingerprint feature set of the sample area after the data processing by a sample area data sample dividing module to obtain training set sample data and test set sample data;
and S4, training the basic model based on the training set sample data to obtain a trained model, wherein the model is used as an indoor positioning module based on the fuzzy migration model and used for analyzing and calculating a data set acquired by a target area to obtain positioning result data for reference of a user.
Example 3
In the high-precision indoor positioning method based on the fuzzy migration learning model of the embodiment, based on embodiment 2, in step S1, data acquisition is performed on a sample area through a sample area data acquisition module, and a fingerprint feature set of the sample area is acquired;
in general, each sample point is characterized by a fingerprint that is made up of the received RSS values of the APs. In the process of fingerprint map construction, each sampling point is collected for a certain period of time to obtain a signal set. The each term in the set records information such as AP name, RSS value, and timestamp. To express simplicity and suppress the influence of outliers, the RSS value of the same AP obtained at each sampling point is mostly averaged as the RSS value of the AP at that point, thereby obtaining a one-dimensional fingerprint vector. Obviously, this way of averaging all the sampled signals ignores the real-time nature of each AP, resulting in the details of the signal fluctuating to be "filtered" and thus the fingerprint characteristics of the sampled points cannot be obtained in fine granularity.
Therefore, in summary, it is necessary to obtain the fingerprint features of the sampling points in a fine granularity, and in order to obtain the features of the sampling points in a finer granularity, the embodiment constructs the Feature set of the sampling points by extracting the Short-time Feature (Short-time Feature) of the fingerprint signal set of the sampling points, that is, the Short-time Feature set of the sampling points, and displays the features of the fingerprints in a finer granularity.
The embodiment carries out data processing on the collected fingerprint feature set of the sample area through a sample area data processing module to obtain a short-time feature set of the sampling point, and specifically comprises the following steps:
s21 referring to FIG. 3, n APs in the sample area are set, the position coordinates of the sampling point P in the sample area are (x, y), and the five-pointed star position in FIG. 3 is the position of the sampling point P, the neighbor P of the sampling point P i Is the coordinates of (a)Wherein i is more than or equal to 1 and less than or equal to 4, inThe positions of the ring points in fig. 3 are all neighbors P of the sampling point P i If the sampling duration of the sampling point P is Γ and the short period is τ, then the RSS vector sampled by the sampling point P in the period t, i.e. in the short period τ after the time t, is expressed as follows by equation (1):
then in the next period (t+τ→t+2τ), the RSS vector sampled by the sampling point P is expressed as:
then the neighbors P of the sampling point P are sampled within a period t, i.e. a short period τ from time t i The sampled RSS vector is expressed as:
s22, calculating ISF of the sampling point P at t time according to a formula (4) to be:
ISF(P，t)＝RSS(P，t→t+τ)+RSS(P，t+τ→t+2τ)＝[ISf 1 ，ISf 2 ，…，ISf n ]formula (4)
In the above formula (4), wherein,
s23 calculates CSF of sampling point P at time t by equation (5):
CSF(P，t)＝RSS(P，t→t+τ)+RSS(P i ，t→t+τ)＝[CSf 1 ，CSf 2 ，…，CSf n ]formula (5)
In the above formula (5), wherein,
S24, obtaining a short-time feature set of the sampling point P at the time T as follows: SFS (P, t) =isf (P, t)/(CSF (P, t);
s25 thus results in a short-term feature set of the sampling point P over the entire sampling period Γ expressed by formula (6):
SFS(P，Γ)＝∑ t∈Γ SFS (P, t) equation (6)
The short-time feature set SFS of the sampling points is obtained through the steps, the features of the sampling points are obtained in a finer granularity mode, and the accuracy of the positioning model is improved based on the model trained by the short-time feature set.
As a preferred embodiment, when calculating ISF and CSF, the situation that the same AP appears in both adjacent time periods or adjacent sampling points may appear, which is of course very little probability, and in order to unify the RSS vector lengths, the embodiment averages the RSS values appearing twice, so as to suppress the influence caused by the abnormality of the AP in a shorter time period, and further improve the accuracy of the positioning model.
Example 4
The high-precision indoor positioning method based on the fuzzy migration learning model of the embodiment is based on embodiment 3, wherein step S3, a sample region data sample dividing module is used for dividing the fingerprint feature set of the sample region after the data processing, so as to obtain training set sample data and test set sample data;
In this embodiment, the short-term feature set SFS of the sampling point obtained in embodiment 2 is divided, and the division result is not limited to one, and for example, the division result may be: 70% is divided into training set sample data, and the rest 30% is divided into test set sample data; the division result may also be: 100% is totally divided into training set sample data to obtain an effective positioning model.
In this embodiment, all sampled SFSs are trained to obtain an efficient positioning model.
In the step S4, the basic Model is trained based on the training set sample data to obtain a trained Model, and the Model is used as an indoor positioning module based on a fuzzy migration Model, and is used for analyzing and calculating a data set collected by a target area to obtain positioning result data for reference of a user, in this embodiment, in order to more effectively process time sequence data SFS, a Long Short-term Memory (LSTM) Model is selected as the basic Model, and the basic Model is trained based on all sampled SFS, and is used for constructing a Pre-Model for migration learning, which specifically includes the following steps:
s41, firstly, training SFS data by optimizing a Long Short-term Memory (LSTM) to obtain Optimized LSTM (O-LSTM) for initially constructing a Pre-Model for migration learning;
S42, because the ISF and CSF in SFS are constructed in different modes, the thought of a focus mechanism is further introduced, a sparse data oriented lightweight mechanism SENet (Squeeze-and-Excitation Network) is used for optimizing the Pre-Model of the migration learning initially constructed, so that Optimization SE-LSTM (OSE-LSTM) is obtained, and the positioning accuracy of the Pre-Model is enhanced.
The optimized training Model is obtained through the steps, and is used as a Pre-Model for transfer learning, and in the practical application process, the Pre-Model for transfer learning is transferred to a target area, and high-precision indoor positioning can be realized no matter whether the target area is adjacent to a sample area or not.
Example 5
The high-precision indoor positioning method based on the fuzzy migration learning model of the present embodiment is based on embodiment 4, wherein step S41: firstly, training SFS data by optimizing Long-term Memory (LSTM) to obtain Optimized LSTM (O-LSTM) for initially constructing Pre-Model for migration learning; the method specifically comprises the following steps: since LSTM uses Memory Cell to selectively store information and regulate and control transmitted data in real time through three gating states of Input Gate, forget Gate and Output Gate, the embodiment mainly includes the following three steps: 1) Inputting a short-time feature set sfs, and firstly generating an Input value through Input Gate; 2) Then, the Forget Gate selects to Forget the last information in the Memory Cell; 3) Finally, output Gate determines whether to Output the information at this time. The method specifically comprises the following steps: in the model training process, a short-time characteristic sfs (mu) is Input at mu moment, and then the short-time characteristic sfs (mu) passes through an Input Gate in an LSTM network and is combined with a mu-1 moment value h (mu-1) in a Memory Cell to obtain an Input value:
I(μ)＝σ(w s，I *sfs(μ)+w h，I * h (mu-1)) formula (7)
In the above formula (7), σ represents an activation function; w (w) s，I And w h，I Parameters that are a function I; sfs (mu) is a mu moment input short-time feature, and h (mu-1) is a mu-1 moment value in a Memory Cell;
at the same time, input Gate generates a candidate vector:
in the above formula (8), w s，c And w h，c As a function ofParameters of (2); sigma represents an activation function; sfs (mu) is a mu moment input short-time feature, and h (mu-1) is a mu-1 moment value in a Memory Cell;
subsequently, the Forget Gate reads sfs at μ and the value h at μ -1 at μ, and outputs a value at a given interval using the activation function to indicate the acceptance of the value:
F(μ)＝σ(w s，f *sfs(μ)+w h，f * h (mu-1)) formula (9)
In the above formula (9), w s，f And w h，f Parameters that are functions F; sigma represents an activation function; sfs (mu) is a mu moment input short-time feature, and h (mu-1) is a mu-1 moment value in a Memory Cell;
next, the state information is updated according to the above-described formula (7), formula (8), and formula (9):
in the above formula (10), F (μ) is formula (9)) The method comprises the steps of carrying out a first treatment on the surface of the I (mu) is formula (7);is formula (8); c (mu) is current state information, and c (mu-1) is state information at the last moment;
finally, the Output Gate outputs the result at the mu moment according to the state information, and the Memory Cell is updated at the same time, and the Memory Cell is updated through a formula (11), wherein the updating mode is as follows:
h (μ) =o (μ) ×tanh (c (μ)) formula (11)
In the above-mentioned formula (11),o (μ) is calculated by equation (12):
when the SFS is adopted to train the positioning model, h μ-1 And x μ Since the sparsity of the input data at adjacent times is large, the output data is generated using the vector difference Δ sfs (μ) at adjacent times from the time dependence of the input vector:
O(μ)＝σ(w s，O *Δsfs(μ)+w h，O * h (mu-1)) formula (12)
In the above formula (12), Δ sfs (μ) = sfs (μ) -sfs (μ -1);
according to the embodiment, the SFS data are updated by using the mode, and the SFS data are trained by optimizing a Long Short-term Memory (LSTM), so that the Optimized LSTM is obtained and is used for initially constructing the Pre-Model for transfer learning, multiplication operation in the training process and the loading amount of the data are greatly reduced, and the training time of the Model is shortened.
In this embodiment, the ReLU function is used as the activation function σ, because it is more suitable for sparse representation, and effectively suppresses the gradient vanishing problem (gradient vanishing) generated by SFS data in training, so that the whole network is easier to converge, and positioning accuracy is further improved.
Example 6
According to the high-precision indoor positioning method based on the fuzzy migration learning Model, based on embodiment 5, SFS-oriented sparsity is adopted, ISF represents self fingerprint characteristics, CSF represents association characteristics with surrounding neighbors and is large in data size, and according to the differences, the thought of an attention mechanism is introduced in the embodiment, and a sparse data-oriented lightweight mechanism SENet (Squeeze-and-Excitation Network) is used for optimizing the preliminarily constructed Pre-Model of migration learning to obtain Optimization SE-LSTM (OSE-LSTM), so that the relation among the characteristics is better represented, and the positioning precision of the Pre-Model is enhanced. The method specifically comprises the following steps:
First, an embedded channel characteristic response is generated with a global distribution, allowing all layers of the training model to use, and global averaging pooling (global average pooling) of the characteristic data to represent the global distribution of responses over the characteristic channels:
in the above formula (13), H represents the number of elements in the SFS, and W represents the length of each element;
then, the relation between the characteristic channels is mined, and nonlinear interaction between the channels is learned by using two layers of nonlinear activation functions so as to obtain proper weights:
S(μ)＝σ(w 2 *δ(w 1 * Z (μ))
In the above formula (14), σ represents a sigmoid activation function, δ represents a ReLU activation function, w 1 And w 2 Representing a scaling parameter;
finally, output is achieved by using channel-wise multiplication:
in the above formula (15), S (μ) is formula (14), and O (μ) is formula (12).
Referring to FIG. 4, FIG. 4 is a block diagram showing the implementation of a Pre-Model for transfer learning, wherein O (μ) is obtained by training LSTM based on a short-term feature set SFS, and by introducing SENet (Squeeze-and-Excitation Network) as wellS (mu) to finally obtainThe method can give consideration to the time sequence and sparsity of short-time data, and simultaneously considers the data with different ISF and CSF generation strategies in SFS, thereby realizing accurate positioning accuracy.
Example 7
According to the high-precision indoor positioning method based on the fuzzy transfer learning model, based on embodiment 6, the transfer learning can transfer the knowledge learned when the task is solved in the source domain to the target domain, and a model with generalization capability is constructed by utilizing a small amount of target domain data. Because the source domain interval is complete, a PreModel constructed using the sampled data of this region can be used to migrate and multiplex to the target domain interval of missing samples. The effect of the transfer learning depends greatly on the distribution similarity of the source domain data and the target domain data. However, the degree of absence (position, size, etc.) of the target domain is uncertain, resulting in different variability between the distribution of the sampled data and the source domain data, and thus accurate positioning of the target domain is difficult to achieve by simply using fine tuning.
The present embodiment further optimizes the fuzzy migration learning model. Firstly, establishing class labels (class labels) for sampling points of a target domain according to the distribution condition of the sampling points in a source domain area by utilizing a fuzzy clustering method, so that source domain data and target domain data are segmented according to classes, and the effect of migration learning is prevented from being influenced by the difference of the overall data distribution of the source domain data and the target domain data; and then, performing transfer learning by using the target domain data marked with the class labels and the obtained pre-model PreModel of transfer learning, thereby constructing a fuzzy transfer learning model meeting different missing situations and different floors. The method specifically comprises the following steps:
(1) And clustering the sampling points of the source domain by using a traditional clustering mode, and marking class labels. Wherein, the feature of the sampling point is a short-time feature set SFS;
(2) Clustering the target domain data by using a fuzzy clustering method according to the category number of the source domain, wherein each sampling point is obtainedClustering label TL f ；
(3) Because the sampling space of the target domain data is similar (even the same) as that of the source domain data, the sampling points at the corresponding positions in the target domain are marked according to the class labels of the sampling points in the source domain, and a theoretical class label TL is obtained t ；
(4) TL of element SFS (P) of short-term feature set SFS of sampling point P in source domain data f With TL (T) t Identical or adjacent, the category label of sfs (P) is set to TL f ；
(5) If TL is f And TL (T) t Not adjacent, then there are:
when there is a class TL x Adjacent to them, the class label of sfs (P) is set to TL x ；
If there are no classes adjacent to them, the record is deleted, i.e., SFS (P) =sfs (P) - { SFS (P) };
through the steps (1) - (5), the target domain data is added with a category label. And performing transfer learning by using the target domain data marked with the class labels and the PreModel, so as to construct fuzzy transfer learning models meeting different missing situations and different floors, and further optimizing the fuzzy transfer learning models.
In the practical application process, the optimized fuzzy transfer learning model is transferred to a target area, and high-precision indoor positioning can be realized no matter whether the target area is adjacent to a sample area or not.
Example 8
The high-precision indoor positioning method based on the fuzzy transfer learning model of the embodiment selects a teaching building as a positioning experimental scene based on the embodiment 7, wherein the teaching building has five floors with similar structures, and the area of each floor is 1450m 2 The experiment was performed using three types of equipment, 5 smartphones, 2 tablets and 1 PDA, respectively. In order to fully verify the positioning effect of the invention, the experiment completely samples 5 floors, the sampling point density is 1.2 x 1.2 meters, and the sampling height is 1 meter. The acquisition time for each sampling point was 3 minutes. The experiment was better tested than the performance of the different positioning models (LSTM, SE-LSTM and OSE-LSTM)The robustness of the positioning model proposed herein is demonstrated, and comparative experiments are performed on different positioning methods. The sampling interval is fixed to be 1.2 meters, a floor is randomly selected as an experimental area, five devices of HUAWEI Mate7, HUAWEI Mate8, honor, VIVO x6 and MI 6 are used for carrying out model performance comparison, and the devices are respectively numbered as devices 1-5. The average error distance AED is used as a positioning performance evaluation criterion. The experimental results are shown in the following table 1, and table 1 is the average positioning error distance (unit: m) of different positioning methods:
TABLE 1
As shown in the above Table 1, the OSE-LSTM method provided by the invention has the lowest error of the positioning result, and further illustrates that the OSE-LSTM method provided by the invention can realize high-precision indoor positioning.
Example 9
In the high-precision indoor positioning method based on the fuzzy migration learning model of the embodiment, based on embodiment 7, in order to better compare the positioning performance difference of migration of the positioning model, a first layer corridor of a teaching building is selected as an experimental area. In order to avoid negative effects of device isomerism on the current experiment, the average value of fingerprint characteristics of data collected by five devices, namely HUAWEI Mate7, HUAWEI Mate8, honor, VIVO x6 and MI 6, at the floor is used as initial data to perform model training and positioning accuracy test, and the sampling interval is 1.2 meters. Wherein the training set is 70% and the test set is 30%. The average error distance AED is used as a positioning performance evaluation criterion. The base positioning model OSE-LSTM is selected after the data training of the full fingerprint points of the first floor is completed, and then the positioning model OSE-LSTM is migrated to other floors. And fine tuning training is performed by adopting different sampling rates on the target domain data, and experimental verification is performed by using a traditional migration learning method and a fuzzy migration method proposed herein respectively. Finally, the experiment compares the positioning results of the traditional WKNN algorithm at different sampling rates of different floors. The specific experimental results of the migration of the positioning model to each floor are shown in table 2 and fig. 5, wherein AF represents an adjacent floor, NAF represents a non-adjacent floor (three, four and five floors), TTL represents conventional migration learning, FTL represents fuzzy migration learning, and table 2 uses different sampling rates for average positioning error distances (unit: m) at different floors for conventional migration learning and fuzzy migration learning:
TABLE 2
From the above experimental results, it can be seen that the model positioning accuracy also appears to slide down to different degrees with the decrease of the sampling density. However, compared with the traditional transfer learning method, the fuzzy transfer learning method provided by the invention has higher precision, and the performance of the fuzzy transfer learning method when being transferred to the adjacent floor is obviously better than that of the fuzzy transfer learning method when not being transferred to the adjacent floor, so that the fuzzy transfer learning method can be more suitable for complex and changeable environments.
Example 10
According to the high-precision indoor positioning method based on the fuzzy migration learning model, based on embodiment 7, the positioning performance result condition of different devices in fuzzy migration to different floors is verified through an OSE-LSTM model. The data from the pre-trained model was derived from all fingerprints of the first floor using five devices, HUAWEI Mate7, MI 6, HUAWEI Honor T1-823, MI Pad 3, UROVO i6300A, numbered devices 1-5, respectively. The sampling rate of the target domain is respectively 80% and 30%, the training method uniformly uses fine tuning training, and the average error distance AED is used as a positioning performance evaluation standard. The experimental results for the target domain at 80% sampling rate are shown in the following tables 3 and 6, and the average positioning error distance (unit: m) of different floors at 80% sampling rate is shown in table 3:
TABLE 3 Table 3
The experimental results for the target domain at a sampling rate of 30% are shown in the following tables 4 and 7, and the average positioning error distance (unit: m) for different floors at a sampling rate of 80% is shown in table 3:
TABLE 4 Table 4
From the experimental results, the positioning error is larger when the mobile station is migrated to the non-adjacent floors, because the farther the floors are, the larger the difference of the spatial structure distribution is, and the larger the data distribution between the source domain and the target domain is. Meanwhile, when the sampling rate is low, the result of the positioning model after being migrated to different floors has certain jitter in positioning errors among different devices. The fuzzy migration learning method provided by the embodiment has higher robustness when migrating to other floors.
Example 11
In the high-precision indoor positioning method based on the fuzzy migration learning model according to the embodiment 7, in order to verify the positive and negative effects of different sampling densities on the positioning model, data of different sampling points at different intervals collected by five devices in a certain floor of a teaching building are selected as original data, an OSE-LSTM positioning model is uniformly used, and minimum sampling rectangles are respectively 1.2 x 1.2, 1.2 x 2.4 and 2.4 x 2.4 (unit: meters). And selecting a training set of 70% and a test set of 30%. The experimental results are shown in FIG. 8. As can be seen from fig. 8, the error is in an ascending trend as the sampling interval is enlarged. But in general the different sampling intervals have less effect on the error of the positioning result.
It can be seen from a combination of the above experiments that the present invention has a high robustness between different devices. When the sampling rate is 80%, the positioning error of the adjacent layer is only 1.38 m, and when the sampling rate is 30%, the positioning error of the adjacent layer is only 1.92 m, so that the sampling work of fingerprint data can be greatly reduced on the premise of ensuring the positioning accuracy. At the same time, compared with the traditional migration, the invention improves by 18.1 percent and 12.6 percent respectively.
The foregoing is only a preferred embodiment of the invention, it being noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the invention.
Claims (6)
1. The high-precision indoor positioning method based on the fuzzy migration learning model is characterized by providing a high-precision indoor positioning system based on the fuzzy migration learning model, wherein the system comprises a target area data acquisition module, a target area data processing module, an indoor positioning module based on the fuzzy migration model and a data output module which are sequentially communicated;
the target area data acquisition module is used for acquiring a fingerprint feature set of the target area;
The target area data processing module is used for carrying out data processing on the collected fingerprint feature set of the target area to obtain data suitable for processing of the next module;
the indoor positioning module based on the fuzzy migration model is used for receiving the data processed by the data processing module, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
the data output module is used for outputting positioning result data for reference of a user;
the method comprises the following steps:
step 1, a target area data acquisition module in a high-precision indoor positioning system based on a fuzzy transfer learning model acquires a fingerprint feature set of a target area;
step 2, a target area data processing module in the high-precision indoor positioning system based on the fuzzy transfer learning model processes data aiming at the collected fingerprint feature set of the target area to obtain data suitable for processing by a next module;
step 3, receiving the data processed by the data processing module by the indoor positioning module based on the fuzzy migration model in the high-precision indoor positioning system based on the fuzzy migration learning model, and analyzing and calculating according to the data to obtain target result data, wherein the target result data is positioning result data;
Step 4, finally outputting positioning result data by a data output module in the high-precision indoor positioning system based on the fuzzy transfer learning model for reference of a user;
the indoor positioning module based on the fuzzy migration model specifically comprises the following steps:
step S1, data acquisition is carried out on a sample area through a sample area data acquisition module, and a fingerprint feature set of the sample area is acquired;
step S2, data processing is carried out on the collected fingerprint feature set of the sample area through a sample area data processing module, so that data suitable for processing of a next module are obtained;
step S3, dividing the fingerprint feature set of the sample area after the data processing by a sample area data sample dividing module to obtain training set sample data and test set sample data;
step S4, training a basic model based on training set sample data to obtain a trained model, wherein the model is used as an indoor positioning module based on a fuzzy migration model and used for analyzing and calculating a data set acquired by a target area to obtain positioning result data for reference of a user;
the step S2 includes the steps of:
s21, if the sampling duration of the sampling point P is Γ and the short period is τ, then in the period t, that is, in the short period τ after the time t, the RSS vector sampled by the sampling point P is expressed as formula (1):
Then in the next period (t+τ→t+2τ), the RSS vector sampled by the sampling point P is expressed as:
then within the t period of time, i.e. a short period of time from time tWithin τ, the neighbors P of the sampling point P i The sampled RSS vector is expressed as:
s22, calculating ISF of the sampling point P at t time through a formula (4):
ISF(P,t)＝RSS(P,t→t+τ)+RSS(P,t+τ→t+2τ)＝[ISf 1 ,ISf 2 ,…,ISf n ]formula (4)
In the above-mentioned formula (4),
s23 calculates CSF of sampling point P at time t by equation (5):
CSF(P,t)＝PSS(P,t→t+τ)+RSS(P i ,t→t+τ)＝[CSf 1 ,CSf 2 ,…,CSf n ]formula (5)
In the above-mentioned formula (5),
s24, obtaining a short-time feature set of the sampling point P at the time T as follows: SFS (P, t) =isf (P, t)/(CSF (P, t);
s25 thus results in a short-term feature set of the sampling point P over the entire sampling period Γ expressed by formula (6):
SFS(P,Γ)＝∑ t∈Γ SFS (P, t) equation (6)
The short-time feature set SFS of the sampling points is obtained through the steps.
2. The high-precision indoor positioning method based on the fuzzy migration learning model of claim 1, wherein the step S4 comprises the steps of:
s41, firstly, training short-time characteristic data SFS through an Optimized long-term memory neural network to obtain an Optimized LSTM, and constructing a Pre-Model for transfer learning preliminarily;
s42, introducing the thought of an attention mechanism, and optimizing the Pre-Model of the migration learning constructed preliminarily by using a lightweight mechanism SENet facing sparse data to obtain an Optimization SE-LSTM, wherein the Optimization SE-LSTM is an optimized training Model.
3. The high-precision indoor positioning method based on the fuzzy migration learning model according to claim 2, wherein the step S41 comprises the steps of, assuming that a short-time feature set Input to the LSTM network at a μmoment is sfs (μ), obtaining an Input value by means of Input Gate in the LSTM network and combining with a μ -1 moment value h (μ -1) in a Memory Cell:
I(μ)＝σ(w s，I *sfs(μ)+w h，I * h (mu-1)) formula (7)
In the above formula (7), σ represents an activation function; w (w) s，I And w h，I Parameters that are a function I; sfs (mu) is a short-time feature input at mu moment, and h (mu-1) is a mu-1 moment value in the Memory Cell;
at the same time, input Gate generates a candidate vector:
in the above formula (8), w s，c And w h，c As a function ofParameters of (2); sigma represents an activation function; sfs (mu) is a short-time feature input at mu moment, and h (mu-1) is a mu-1 moment value in the Memory Cell;
then, the Forget Gate reads sfs at μ and the μ -1 value h at μ, and outputs a value F (μ) at a given interval using the activation function:
F(μ)＝σ(w s，f *sfs(μ)+w h，f * h (mu-1)) formula (9)
In the above formula (9), w s，f And w h，f Parameters that are functions F; sigma represents an activation function; sfs (mu) is a mu moment input short-time feature, and h (mu-1) is a mu-1 moment value in a Memory Cell;
next, the state information is updated according to the above-described formula (7), formula (8), and formula (9):
In the above formula (10), F (μ) is formula (9); i (mu) is formula (7);is formula (8);
finally, the Output Gate outputs the result at the mu moment according to the state information, and the Memory Cell is updated at the same time, and the Memory Cell is updated through a formula (11), wherein the updating mode is as follows:
h (μ) =o (μ) ×tanh (c (μ)) formula (11)
In the above-mentioned formula (11),o (μ) is calculated by equation (12):
O(μ)＝σ(w s，O *Δsfs(μ)+w h，O * h (mu-1)) formula (12)
In the above formula (12), the vector difference Δ sfs (μ) = sfs (μ) -sfs (μ -1) at adjacent times;
the short-term characteristic data SFS is trained through the long-term and short-term memory neural network LSTM, and the Optimized LSTM is obtained and used for initially constructing the Pre-Model for migration learning.
4. A high-precision indoor positioning method based on a fuzzy migration learning model according to claim 3, wherein a ReLU function is adopted as the activation function σ.
5. The high-precision indoor positioning method based on the fuzzy migration learning model of claim 4, wherein S42 comprises the steps of:
first, an embedded channel characteristic response is generated with a global distribution, allowing all layers of the training model to use, and the characteristic data is globally averaged and pooled to represent the global distribution of responses over the characteristic channels:
In the above formula (13), H represents the number of elements in the SFS, and W represents the length of each element;
then, the relation between the characteristic channels is mined, and nonlinear interaction between the channels is learned by using two layers of nonlinear activation functions so as to obtain proper weights:
S(μ)＝σ(w 2 *δ(w 1 * Z (μ))
In the above formula (14), σ represents a sigmoid activation function, δ represents a ReLU activation function, w 1 And w 2 Representing a scaling parameter;
finally, output is achieved by using channel-wise multiplication:
in the above formula (15), S (μ) is formula (14), and O (μ) is formula (12);
the pre-model O-LSTM of the migration learning which is initially constructed is optimized to obtain the OSE-LSTM by introducing a lightweight mechanism SENet facing sparse data.
6. The method for high-precision indoor positioning based on a fuzzy migration learning model of claim 5, wherein the OSE-LSTM is further optimized: firstly, establishing class labels for sampling points of a target domain according to the distribution condition of the sampling points in a source domain area by using a fuzzy clustering method; and then, performing transfer learning by using the target domain data marked with the class labels and the obtained pre-model PreModel of transfer learning, thereby constructing a fuzzy transfer learning model meeting different missing situations and different floors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211292065.9A CN115905787B (en) | 2022-10-21 | 2022-10-21 | High-precision indoor positioning method based on fuzzy migration learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211292065.9A CN115905787B (en) | 2022-10-21 | 2022-10-21 | High-precision indoor positioning method based on fuzzy migration learning model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115905787A CN115905787A (en) | 2023-04-04 |
| CN115905787B true CN115905787B (en) | 2023-09-29 |
Family
ID=86473433
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211292065.9A Active CN115905787B (en) | 2022-10-21 | 2022-10-21 | High-precision indoor positioning method based on fuzzy migration learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115905787B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016136099A (en) * | 2015-01-23 | 2016-07-28 | サトーホールディングス株式会社 | Positioning system and positioning method |
| CN110958569A (en) * | 2019-12-11 | 2020-04-03 | 军事科学院系统工程研究院网络信息研究所 | Indoor positioning method based on MIMO channel characteristic value |
| CN111898523A (en) * | 2020-07-29 | 2020-11-06 | 电子科技大学 | Remote sensing image special vehicle target detection method based on transfer learning |
| CN112584311A (en) * | 2020-12-15 | 2021-03-30 | 西北工业大学 | Indoor three-dimensional space fingerprint positioning method based on WKNN fusion |
| CN114269006A (en) * | 2021-12-24 | 2022-04-01 | 河海大学 | Indoor AP cluster selection method and device based on information gain rate |
| CN114710742A (en) * | 2022-02-28 | 2022-07-05 | 盐城师范学院 | Indoor positioning method for constructing fingerprint map based on multi-chain interpolation |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11438886B2 (en) * | 2020-02-27 | 2022-09-06 | Psj International Ltd. | System for establishing positioning map data and method for the same |
-
2022
- 2022-10-21 CN CN202211292065.9A patent/CN115905787B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016136099A (en) * | 2015-01-23 | 2016-07-28 | サトーホールディングス株式会社 | Positioning system and positioning method |
| CN110958569A (en) * | 2019-12-11 | 2020-04-03 | 军事科学院系统工程研究院网络信息研究所 | Indoor positioning method based on MIMO channel characteristic value |
| CN111898523A (en) * | 2020-07-29 | 2020-11-06 | 电子科技大学 | Remote sensing image special vehicle target detection method based on transfer learning |
| CN112584311A (en) * | 2020-12-15 | 2021-03-30 | 西北工业大学 | Indoor three-dimensional space fingerprint positioning method based on WKNN fusion |
| CN114269006A (en) * | 2021-12-24 | 2022-04-01 | 河海大学 | Indoor AP cluster selection method and device based on information gain rate |
| CN114710742A (en) * | 2022-02-28 | 2022-07-05 | 盐城师范学院 | Indoor positioning method for constructing fingerprint map based on multi-chain interpolation |
Non-Patent Citations (1)
| Title |
|---|
| "基于迁移学习和指纹的室内定位算法";张笑凯;《中国优秀硕士学位论文全文数据库信息科技辑》;第二章-第三章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115905787A (en) | 2023-04-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mirowski et al. | Probabilistic radio-frequency fingerprinting and localization on the run | |
| Njima et al. | Indoor localization using data augmentation via selective generative adversarial networks | |
| CN110414732B (en) | Travel future trajectory prediction method and device, storage medium and electronic equipment | |
| Yang et al. | ImgSensingNet: UAV vision guided aerial-ground air quality sensing system | |
| Torres-Sospedra et al. | A comprehensive and reproducible comparison of clustering and optimization rules in wi-fi fingerprinting | |
| Xing et al. | Mapping human activity volumes through remote sensing imagery | |
| Li et al. | Supreme: Fine-grained radio map reconstruction via spatial-temporal fusion network | |
| CN109059919B (en) | Indoor positioning method based on crowdsourcing sample weighted surface fitting | |
| CN106951828B (en) | Urban area function attribute identification method based on satellite images and network | |
| Niu et al. | RecNet: A convolutional network for efficient radiomap reconstruction | |
| Canovas et al. | Detecting indoor/outdoor places using WiFi signals and AdaBoost | |
| CN114449452A (en) | Indoor positioning algorithm for heterogeneous equipment | |
| CN106600652A (en) | Panorama camera positioning method based on artificial neural network | |
| Eagle et al. | Location Segmentation, Inference and Prediction for Anticipatory Computing. | |
| He et al. | Indoor localization with spatial and temporal representations of signal sequences | |
| Liu et al. | Indoor localization with adaptive signal sequence representations | |
| Naveed et al. | Assessing deep generative models on time series network data | |
| CN109121081B (en) | Indoor positioning method based on position candidate set and EM algorithm | |
| Li et al. | Model and transfer spatial-temporal knowledge for fine-grained radio map reconstruction | |
| CN109141420B (en) | Indoor positioning method based on crowdsourcing data | |
| Eagle et al. | Methodologies for continuous cellular tower data analysis | |
| Tian et al. | A survey on clustering based meteorological data mining | |
| Liu et al. | A radio map self-updating algorithm based on mobile crowd sensing | |
| CN115905787B (en) | High-precision indoor positioning method based on fuzzy migration learning model | |
| Xu | Autonomous Indoor Localization Using Unsupervised Wi-Fi 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 |