CN114501329A

CN114501329A - Environment self-adaptive training method and device of indoor positioning model based on 1D-CNN

Info

Publication number: CN114501329A
Application number: CN202111534604.0A
Authority: CN
Inventors: 林建旋; 付一夫; 方剑平; 胡思林; 闫航; 郭航瑞; 王佳昊
Original assignee: Xiamen Zhixiaojin Intelligent Technology Co ltd
Current assignee: Xiamen Zhixiaojin Intelligent Technology Co ltd
Priority date: 2021-12-15
Filing date: 2021-12-15
Publication date: 2022-05-13

Abstract

The invention discloses an environment self-adaptive training method and device of an indoor positioning model based on 1D-CNN, wherein a position fingerprint database and a 1D-CNN positioning model are established, the 1D-CNN positioning model is trained through the fingerprint database, and a convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer; and taking the position coordinates and CSI data in the position fingerprint database as first source domain data, taking the position coordinates and CSI data acquired by reference points in an actual environment as first target domain data, processing the first source domain data and the second target domain data through a TCA algorithm to obtain second source domain data and second target domain data, and respectively taking the second source domain data and the second target domain data as a training set and a test set to train the 1D-CNN positioning model to obtain an environment self-adaptive indoor positioning model. And updating the positioning model according to different environmental changes to realize good precision.

Description

Environment self-adaptive training method and device of indoor positioning model based on 1D-CNN

Technical Field

The invention relates to the field of indoor positioning, in particular to an environment self-adaptive training method and device of an indoor positioning model based on 1D-CNN.

Background

Wi-Fi based indoor positioning technology has many advantages over satellite signal based positioning technology and thus has received much attention. In the previous research, indoor positioning is based on Received Signal Strength (RSSI), and then indoor positioning accuracy based on RSSI is greatly improved due to continuous new algorithm. In recent years, with the release of CSI, it is possible to acquire channel state information in a WiFi environment, and therefore an indoor positioning method based on CSI as a characteristic parameter is increasingly gaining importance.

The indoor positioning method based on the position fingerprints mainly comprises four parts, namely fingerprint data acquisition, data preprocessing, positioning model construction and data positioning. And acquiring signal characteristic parameters of the positions of Reference nodes (RP) in a required positioning area, wherein the characteristic parameters comprise CSI data segments and position coordinates of RP points. After data preprocessing is performed on the acquired CSI signal segments, a corresponding relation between the CSI segments and the position coordinates is established according to a specific algorithm, and a position fingerprint database (LocationFingerprintDatabase) is generated. In the on-line positioning stage, the collected CSI signal segments are preprocessed and sent into a trained model, and the user position is obtained in a regression or classification mode according to the established model.

However, most of the existing research on indoor positioning assumes that the probability distribution of CSI data is stable and unchangeable, but this assumption basis is not established in a real indoor environment with a complex composition, and the original model cannot accurately position the user position with the change of various environmental conditions.

Disclosure of Invention

The technical problem mentioned above is addressed. An embodiment of the present application aims to provide a complete indoor positioning model training method that is more efficient and more convenient under different environmental changes, so as to solve the technical problems mentioned in the background art.

In a first aspect, an embodiment of the present application provides an environment adaptive training method for an indoor positioning model based on 1D-CNN, including the following steps:

s1, establishing a position fingerprint database and a 1D-CNN positioning model, training the 1D-CNN positioning model through the fingerprint database, wherein the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer;

s2, taking the position coordinates and CSI data in the position fingerprint database as first source domain data, taking the position coordinates and CSI data acquired by a reference point in an actual environment as first target domain data, and processing the first target domain data and the position coordinates and CSI data by a TCA algorithm to obtain second source domain data and second target domain data;

s3, training the 1D-CNN positioning model by using the second source domain data and the second target domain data as a training set and a test set respectively, and obtaining an environment self-adaptive indoor positioning model.

In some embodiments, step S1 specifically includes:

s11, collecting original CSI data and position coordinates of a reference point, removing obvious outliers in the original CSI data by adopting Hample filtering, and smoothing the original CSI data by adopting a wavelet threshold denoising method based on an extreme value threshold to obtain preprocessed CSI data;

and S12, forming a position fingerprint database by the preprocessed CSI data and the corresponding position coordinates.

In some embodiments, the raw CSI data is composed of N_t×N_r×S_nIs a one-dimensional vector of amplitude values of, wherein N_t、N_rNumber of transmitting antennas and receiving antennas, S, respectively, in a MIMO system_nIndicating the number of subcarriers.

In some embodiments, the input in the 1D-CNN positioning model is CSI data and the output is location feature information, which corresponds to location coordinates in a location fingerprint database.

In some embodiments, the 1D-CNN localization model includes an input layer, a convolutional layer, a pooling layer, a fully-connected layer, and an output layer, where the input layer is used to normalize the CSI data, and then passes through a first convolutional layer, a second convolutional layer, a maximum pooling layer, a third convolutional layer, an average pooling layer, two fully-connected layers, and a Softmax output layer, where the first convolutional layer, the second convolutional layer, and the third convolutional layer are all one-dimensional convolutional layers and have convolutional kernel numbers of 4, 8, and 16, respectively.

In some embodiments, the 1D-CNN localization model includes an input layer, three one-dimensional convolutional layers, three one-dimensional maximal pooling layers, and two fully-connected layers and a Softmax output layer, where each one-dimensional convolutional layer is followed by a ReLU as an activation function and connected to one-dimensional maximal pooling layer.

In some embodiments, step S2 specifically includes:

s21, calculating to obtain two input matrixes respectively as an L matrix and an H matrix according to the first source domain data and the first target domain data;

s22, mapping the first source domain data and the first target domain data to a high-dimensional Hilbert space through a kernel function and calculating a kernel matrix K;

s23, determining the dimension value M, solving the formula (KLK + mu I)^-1The first M eigenvectors of KHK form a matrix W, the matrix W is multiplied by the kernel matrix K to obtain a matrix A, and the first n of the matrix A₁The term is the second source domain data, the last n of the matrix A₂The item is second source domain data, n₁Is the data volume of the second source domain data, n₂Is the data amount of the second target domain data.

In a second aspect, an embodiment of the present application provides an environment adaptive training apparatus based on a 1D-CNN indoor positioning model, including:

the model establishing module is configured to establish a position fingerprint database and a 1D-CNN positioning model, the 1D-CNN positioning model is trained through the fingerprint database, and the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer;

the transfer learning module is configured to take part of position coordinates and CSI data in the position fingerprint database as first source domain data, take position coordinates and CSI data acquired by a reference point in an actual environment as first target domain data, and obtain second source domain data and second target domain data through TCA algorithm processing;

and the environment self-adaption module is configured to train the 1D-CNN positioning model by respectively using the second source domain data and the second target domain data as a training set and a test set to obtain an environment self-adaption indoor positioning model.

In a third aspect, embodiments of the present application provide an electronic device comprising one or more processors; storage means for storing one or more programs which, when executed by one or more processors, cause the one or more processors to carry out a method as described in any one of the implementations of the first aspect.

In a fourth aspect, embodiments of the present application provide a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

Compared with the prior art, the invention has the following beneficial effects:

(1) compared with the prior art that only the CSI data probability distribution is assumed to be constant and unchanged, the method considers the real indoor complex scene condition and introduces the idea of transfer learning, and utilizes the original position fingerprint database to perform domain adaptation by means of the idea of the TCA algorithm.

(2) The invention can update the indoor positioning model more efficiently and more conveniently according to different environmental changes by a domain adaptation method to realize good positioning precision.

(3) The 1D-CNN positioning model adopted in the invention uses a one-dimensional convolution layer, which can effectively improve the identification performance, and the structural form is diversified, thereby being beneficial to flexible use according to specific scenes.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is an exemplary device architecture diagram in which one embodiment of the present application may be applied;

FIG. 2 is a schematic diagram of an environment adaptive training apparatus based on an indoor positioning model of 1D-CNN according to an embodiment of the present invention;

fig. 3 is a schematic structural diagram of a computer device suitable for implementing an electronic apparatus according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Fig. 1 illustrates an exemplary device architecture 100 to which the environment adaptive training method based on the indoor positioning model of 1D-CNN or the environment adaptive training device based on the indoor positioning model of 1D-CNN according to the embodiment of the present application may be applied.

As shown in fig. 1, the apparatus architecture 100 may include

terminal devices

101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. Various applications, such as data processing type applications, file processing type applications, etc., may be installed on the

terminal apparatuses

101, 102, 103.

The

terminal apparatuses

101, 102, and 103 may be hardware or software. When the

terminal devices

101, 102, 103 are hardware, they may be various electronic devices including, but not limited to, smart phones, tablet computers, laptop portable computers, desktop computers, and the like. When the

terminal apparatuses

101, 102, 103 are software, they can be installed in the electronic apparatuses listed above. It may be implemented as multiple pieces of software or software modules (e.g., software or software modules used to provide distributed services) or as a single piece of software or software module. And is not particularly limited herein.

The server 105 may be a server that provides various services, such as a background data processing server that processes files or data uploaded by the

terminal devices

101, 102, 103. The background data processing server can process the acquired file or data to generate a processing result.

It should be noted that the environment adaptive training method based on the indoor positioning model of 1D-CNN provided in the embodiment of the present application may be executed by the server 105, or may also be executed by the

terminal devices

101, 102, and 103, and accordingly, the environment adaptive training apparatus based on the indoor positioning model of 1D-CNN may be disposed in the server 105, or may also be disposed in the

terminal devices

101, 102, and 103.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation. In the case where the processed data does not need to be acquired from a remote location, the above device architecture may not include a network, but only a server or a terminal device.

The embodiment of the application provides an environment self-adaptive training method of an indoor positioning model based on 1D-CNN, which comprises the following steps:

s1, establishing a position fingerprint database and a 1D-CNN positioning model, training the 1D-CNN positioning model through the fingerprint database, wherein the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer.

In a specific embodiment, step S1 specifically includes:

In a specific embodiment, the raw CSI data is composed of N_t×N_r×S_nIs a one-dimensional vector of amplitude values of, wherein N_t、N_rNumber of transmitting antennas and receiving antennas, S, respectively, in a MIMO system_nIndicating the number of subcarriers. The input of the 1D-CNN positioning model is CSI data, and the output is position characteristic information which corresponds to position coordinates in a position fingerprint database.

Specifically, the step is a training step in an off-line stage, and fingerprint data samples of each reference point are collected, where the fingerprint data samples are composed of channel state information values of the point, that is, original CSI data, and position coordinates of the room where the fingerprint point is located. In one embodiment, the collection of the original CSI data is mainly to extract CSI data values of each reference point from the Wi-Fi signal through commercial network cards Intel5300 and Atheros 9000. In order to solve the problem of frequency domain resource shortage, a MIMO system is introduced, and N is used_tA transmitting antenna and N_rN is formed by taking a receiving antenna as an example_t×N_rThe original CSI data in each collected data packet is composed of N_t×N_r×S_n(S_nRepresenting the number of subcarriers) a sample matrix of amplitude values. When a positioning target is set at a reference point i, a magnitude vector extracted from a data packet is generally subjected to a one-dimensional processing to obtain a fingerprint vector

(χ_iIn a fixed transceiver antennaMagnitude of amplitude of a certain subcarrier) and the probability distribution of the reference point data is P (X)_i) Each positioning will take 1000 data packets at the reference point to form a fingerprint data sample.

Due to the hardware defects of commercial Wi-Fi equipment, various electromagnetic interferences in the environment, inevitable white noise and other problems, the acquired original CSI data has a lot of noise components, and if the original CSI data is not processed, the establishment of an indoor positioning model is seriously influenced subsequently. Therefore, after the raw CSI data is collected, the data must be subjected to noise reduction processing. There are many noise reduction methods, and most commonly, a combination of various filters is used to extract desired time domain information or frequency domain components. In the embodiment of the application, obvious outliers in the data segment are removed by using Hample filtering, and the original characteristics of signals are kept while the data segment is subjected to smoothing processing by using a wavelet threshold denoising method based on a Minimax threshold, so that the analysis of non-stationary signals and the extraction of local characteristics of abrupt change signals are facilitated.

In particular, for the fingerprint vector mentioned in the above step

As a sequence, an observation window is generated around each element, assuming that the length of a half window is K, the width of the whole window is 2K +1 (including the central element), the median of all elements of the window is calculated, the standard deviation of each sample to the median is estimated by using median calculation, and if the difference between a certain sample and the median is greater than three standard deviations, the sample is replaced by the median.

In practical situations, the received signal may not contain only one noise signal, and the signal may be mixed with different levels of noise in the channel. A noisy signal of length N, the noisy data measured being X_n＝f_n+e_n(f_nAs the original signal, e_nA noise signal) and therefore the signal has to be filtered.

In an environment of acquiring CSI data, because random noise is more in an actual household environment, a wavelet threshold denoising method which is commonly used in wavelet transformation and has a better effect is adopted to process original CSI data, and the denoising process of the filtering algorithm is as follows:

(1) for signal X with noise_nAnd performing discrete wavelet transform on the plurality of data packets. Decomposing an original signal by using Daubechies wavelets to obtain a group of wavelet decomposition coefficients W_j,k；

(2) The wavelet decomposition coefficient W obtained by the above steps_j,kThe method is processed by an extreme value threshold principle (a threshold is selected by adopting a maximum and minimum value principle to generate an extreme value of a minimum mean square error), in addition, in the embodiment of the application, in order to solve the problem of local jitter possibly generated after denoising caused by a hard threshold method, a soft threshold method is particularly adopted as a method for correcting the wavelet coefficient, when the absolute value of the wavelet coefficient is smaller than a given threshold value, the wavelet coefficient is made to be zero, and when the absolute value of the wavelet coefficient is larger than the threshold value, the wavelet coefficient is made to subtract the threshold value, namely the threshold value is subtracted

Wherein, sgn (W)_j,k) Is a sign function, λ is a threshold.

(3) And finally, performing wavelet reconstruction to obtain the processed CSI data. And constructing the processed CSI data and the corresponding position coordinates into a position fingerprint database.

In a specific embodiment, as can be known from the distribution of the filtered CSI data, the CSI data of different reference points have an obvious characteristic of aggregate distribution, and the obtained CSI data is N_t×N_r×S_nOne-dimensional vectors formed by the amplitudes of the subcarriers cannot well maintain the position distinguishing information if the CSI data is directly used. The embodiment of the application extracts the position characteristic information of the 1D-CNN positioning model by means of a one-dimensional convolution kernel in the positioning model. The 1D-CNN positioning model consists of an input layer, a convolution layer, a pooling layer, a full-connection layer and an output layer, and specifically comprises the following parts which are connected in sequence:

(1) an input layer: because the input data is a one-dimensional vector and the input channel is 1, the data can be sent to the convolutional layer after normalization processing.

(2) A first layer of convolutional layers: one-dimensional convolutional layers contain fewer convolution vectors than two-dimensional convolutional layers, and therefore a larger convolution kernel should be selected, so the convolution kernel size is set to 7, the step size is set to 1, and the number of convolution kernels is set to 4. Since Zero-padding (Zero-padding) can make the output feature sequence longer, retain more data features and better promote the identification performance of the one-dimensional convolutional neural network, padding is set to 0;

(3) a second layer of convolutional layers: the number of convolution kernels is 8 according to the parameters above;

(4) maximum pooling layer: the size of the pooling kernel is set to be 5, the step size is 1, and padding is 0 as well;

(5) a third layer of convolutional layers: the parameters are the same as above, the number of convolution kernels is 16;

(6) average pooling layer: the size of the pooling kernel is set to be 5, the step size is 1, and padding is 0 as well;

(7) two full-connected layers: the number of the neurons is 1024 and 512 respectively;

(8) softmax output layer: as a classification problem, the number of output neurons is set as the number of reference points in the last layer.

In another embodiment, the 1D-CNN localization model comprises an input layer, three one-dimensional convolutional layers, three one-dimensional maximal pooling layers, and two fully-connected layers and a Softmax output layer, wherein each one-dimensional convolutional layer is followed by a ReLU as an activation function and connected with one-dimensional maximal pooling layer. The convolution kernel size of each one-dimensional convolution layer is 3, and the step size is 1. The network structure setting of the 1D-CNN positioning model can adopt different forms and has more flexibility.

And (5) training in an off-line stage by adopting the position fingerprint database in the step S1, and finishing the training and storing corresponding parameters in the 1D-CNN positioning model when the error is reduced to a certain degree until the task of establishing the 1D-CNN positioning model is completed.

S2, taking part of position coordinates and CSI data in the position fingerprint database as first source domain data, taking the position coordinates and CSI data acquired by reference points in an actual environment as first target domain data, and obtaining second source domain data and second target domain data through TCA algorithm processing.

In a specific embodiment, step S2 specifically includes:

Specifically, the user determines whether the 1D-CNN localization model needs to be retrained according to actual requirements to perform environment adaptive adjustment on the 1D-CNN localization model, and if the 1D-CNN localization model established and trained according to step S1 can accurately obtain indoor localization, the 1D-CNN localization model does not need to be retrained. When the user judges that the environment self-adaptation is needed, the relevant program is started, and the 1D-CNN positioning model can be retrained by using a self-adaptation method based on the TCA algorithm according to part of target domain data after the environment is changed, so that the function of adapting to the environment change is achieved.

Specifically, the assumption underlying the TCA algorithm is that the edge distribution is different, namely P (X)_S)≠P(X_T) But the conditional probability distributions are the same P (Y)_S|X_S)≠P(Y_T|X_T). The idea of the TCA algorithm is to map the data of both domains together into a high-dimensional regenerated hilbert space when the source domain and the target domain are in different data distributions. In this space, the source domain data and the target domain are minimizedThe distance between the data while preserving their respective internal properties to the maximum extent that the distribution P (X) of the mapped data is_S))≈P(φ(X_T) Further, the condition distribution is P (Y)_S|φ(X_S))≈P(Y_T|φ(X_T)). Wherein X_S、Y_SBeing data and tags of the source domain, X_T、Y_TPhi () represents a mapping method for mapping from an original space to a high-dimensional space, and is kernel function mapping.

To minimize the distance between the source domain data and the target domain data, the distance between the source domain data and the target domain data is calculated with reference to the maximum mean difference MMD, the distance formula is as follows:

wherein the source domain data

For tagged data, target domain data

Is unlabeled data. Thus is provided with

Wherein, dist (X'_src,X′_tar) I.e. an empirically estimated distance between the two domains. n1 is the data size of the source domain data, and n2 is the data size of the target domain data.

To process the mapping relationship, the mapping function is solved in the form of a kernel function, and a kernel matrix K is introduced:

and L:

the MMD distance is converted into the form:

trace(KL)-λtrace(K)；

for efficient calculation, the TCA algorithm firstly decomposes a kernel matrix K and constructs a result in a dimension reduction mode, which is as follows:

after the arrangement, the final optimization target of the TCA algorithm is as follows:

s.t.W^TKHKW＝I_m。

MMD is a loss function widely used in transfer learning, especially domain adaptive, and is mainly used to measure the distance between two different but correlated distributions. In the embodiment of the application, a non-deep transfer learning method is adopted, and transfer learning is realized by using matrix decomposition and MMD distance to perform data dimension reduction and feature extraction.

Therefore, the environment self-adaption method based on the TCA algorithm comprises the following steps:

(1) selecting few data samples in the position fingerprint database as source domain data in transfer learning and data samples collected when the self-adaptive method is started as target domain data, calculating L and H matrixes through the formula,

h matrix is W as described above^TKHKW＝I_mThe formula of the H matrix is as follows:

the H matrix is the central matrix.

(2) An appropriate kernel function (such as Linear, RBF, etc.) is selected to map the source domain data and the target domain data to a high-dimensional hilbert space and compute a kernel matrix K.

(3) Determining the dimension value M after dimension reduction (M is not fixed and needs to be tested for an optimal value for many times), and then solving a formula (KLK + mu I)^-1KHK of the first M feature vectors, (KLK + muI)^-1The first M-dimension eigenvector of the eigenvector matrix obtained from KHK forms a matrix W, the matrix W is multiplied by the kernel matrix K obtained from the mapping calculation in the step 2 to obtain a matrix A, and the first n of the matrix A₁The term is the source domain data processed by the TCA algorithm, the last n of the matrix A₂The term is the target domain data processed by the TCA algorithm.

(4) And (3) retraining the 1D-CNN positioning model by using the source domain data and the target domain data obtained in the step (2) as a training set and a test set so as to obtain the environment-adaptive indoor positioning model.

After the environment changes, the CSI data in the position fingerprint database is used as source domain data to be subjected to domain adaptation with target domain data by using a migration component analysis (TCA) algorithm based on maximum mean difference, so that an indoor positioning model after migration is realized, the indoor positioning accuracy is higher through the indoor positioning model, and the adaptability to the environment is stronger.

With further reference to fig. 2, the present application provides an embodiment of an apparatus for environment adaptive training based on an indoor positioning model of 1D-CNN, which may be specifically applied to various electronic devices.

The embodiment of the application provides an environment self-adaptive training device of an indoor positioning model based on 1D-CNN, which comprises:

the model establishing module 1 is configured to establish a position fingerprint database and a 1D-CNN positioning model, the 1D-CNN positioning model is trained through the fingerprint database, and the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer;

the transfer learning module 2 is configured to use part of the position coordinates and the CSI data in the position fingerprint database as first source domain data, use the position coordinates and the CSI data acquired by a reference point in an actual environment as first target domain data, and obtain second source domain data and second target domain data through TCA algorithm processing;

and the environment self-adaption module 3 is configured to train the 1D-CNN positioning model by respectively using the second source domain data and the second target domain data as a training set and a test set to obtain an environment self-adaption indoor positioning model.

Referring now to fig. 3, a schematic diagram of a computer apparatus 400 suitable for use in implementing an electronic device (e.g., the server or terminal device shown in fig. 1) according to an embodiment of the present application is shown. The electronic device shown in fig. 3 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.

As shown in fig. 3, the computer apparatus 400 includes a Central Processing Unit (CPU)401 and a Graphic Processor (GPU) 402, which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM)403 or a program loaded from a storage section 409 into a Random Access Memory (RAM) 404. In the RAM 404, various programs and data necessary for the operation of the apparatus 400 are also stored. The CPU 401, GPU402, ROM 403, and RAM 404 are connected to each other via a bus 405. An input/output (I/O) interface 406 is also connected to bus 405.

The following components are connected to the I/O interface 406: an input portion 407 including a keyboard, a mouse, and the like; an output section 408 including a display such as a Liquid Crystal Display (LCD) and a speaker; a storage portion 409 including a hard disk and the like; and a communication section 410 including a network interface card such as a LAN card, a modem, or the like. The communication section 410 performs communication processing via a network such as the internet. The driver 411 may also be connected to the I/O interface 406 as needed. A removable medium 412 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 411 as necessary, so that a computer program read out therefrom is mounted into the storage section 409 as necessary.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 410, and/or installed from the removable medium 412. The computer program performs the above-described functions defined in the method of the present application when executed by a Central Processing Unit (CPU)401 and a Graphics Processing Unit (GPU) 402.

It should be noted that the computer readable medium described herein can be a computer readable signal medium or a computer readable medium or any combination of the two. The computer readable medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor device, apparatus, or any combination of the foregoing. More specific examples of the computer readable medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution apparatus, device, or apparatus. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution apparatus, device, or apparatus. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations for aspects of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based devices that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present application may be implemented by software or hardware. The modules described may also be provided in a processor.

As another aspect, the present application also provides a computer-readable medium, which may be contained in the electronic device described in the above embodiments; or may exist separately without being assembled into the electronic device. The computer readable medium carries one or more programs which, when executed by the electronic device, cause the electronic device to: establishing a position fingerprint database and a 1D-CNN positioning model, training the 1D-CNN positioning model through the fingerprint database, wherein the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer; taking the position coordinates and CSI data in the position fingerprint database as first source domain data, taking the position coordinates and CSI data acquired by a reference point in an actual environment as first target domain data, and processing by a TCA algorithm to obtain second source domain data and second target domain data; and training the 1D-CNN positioning model by respectively using the second source domain data and the second target domain data as a training set and a test set to obtain an environment self-adaptive indoor positioning model.

The above description is only a preferred embodiment of the application and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention herein disclosed is not limited to the particular combination of features described above, but also encompasses other arrangements formed by any combination of the above features or their equivalents without departing from the spirit of the invention. For example, the above features may be replaced with (but not limited to) features having similar functions disclosed in the present application.

Claims

1. An environment self-adaptive training method of an indoor positioning model based on 1D-CNN is characterized by comprising the following steps:

s1, establishing a position fingerprint database and a 1D-CNN positioning model, and training the 1D-CNN positioning model through the fingerprint database, wherein the convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer;

s3, training the 1D-CNN positioning model by respectively using the second source domain data and the second target domain data as a training set and a test set to obtain an environment self-adaptive indoor positioning model.

2. The method for environment adaptive training of an indoor positioning model based on 1D-CNN as claimed in claim 1, wherein the step S1 specifically includes:

and S12, forming the position fingerprint database by the preprocessed CSI data and the corresponding position coordinates.

3. The method of claim 2, wherein the raw CSI data is N-based_t×N_r×S_nIs formed as a one-dimensional vector of amplitudes, where N is_t、N_rNumber of transmitting antennas and receiving antennas, S, respectively, in a MIMO system_nIndicating the number of subcarriers.

4. The method as claimed in claim 1, wherein the 1D-CNN-based indoor positioning model has CSI data as input and location feature information as output, and the location feature information corresponds to the location coordinates in the location fingerprint database.

5. The method of claim 1, wherein the 1D-CNN-based indoor positioning model comprises an input layer, a convolution layer, a pooling layer, a fully-connected layer and an output layer, the input layer is used for performing normalization processing on the CSI data, and the normalized CSI data passes through a first convolution layer, a second convolution layer, a maximum pooling layer, a third convolution layer, an average pooling layer, two fully-connected layers and a Softmax output layer, wherein the first convolution layer, the second convolution layer and the third convolution layer are one-dimensional convolution layers, and the number of convolution cores is 4, 8 and 16.

6. The method of claim 1, wherein the 1D-CNN positioning model comprises an input layer, three one-dimensional convolutional layers, three one-dimensional maximal pooling layers, two fully-connected layers, and a Softmax output layer, wherein each one-dimensional convolutional layer is followed by a ReLU as an activation function and connected to one-dimensional maximal pooling layer.

7. The method for environment adaptive training of an indoor positioning model based on 1D-CNN as claimed in claim 1, wherein the step S2 specifically includes:

s22, mapping the first source domain data and the first target domain data to a high-dimensional Hilbert space through a kernel function, and calculating a kernel matrix K;

s23, determining the dimension value M, solving the formula (KLK + mu I)^-1The first M eigenvectors of the KHK form a matrix W, the matrix W is multiplied by the kernel matrix K to obtain a matrix A, and the first n of the matrix A₁The item is the second source domain data, the last n of the matrix A₂The item isThe second target domain data, n₁Is the data volume, n, of the second source domain data₂Is the data amount of the second target domain data.

8. An environment adaptive training device based on an indoor positioning model of 1D-CNN is characterized by comprising:

the model establishing module is configured to establish a position fingerprint database and a 1D-CNN positioning model, the 1D-CNN positioning model is trained through the fingerprint database, and a convolution layer of the 1D-CNN positioning model adopts a one-dimensional convolution layer;

the transfer learning module is configured to use part of position coordinates and CSI data in the position fingerprint database as first source domain data, use position coordinates and CSI data acquired by a reference point in an actual environment as first target domain data, and obtain second source domain data and second target domain data through TCA algorithm processing;

and the environment self-adaptation module is configured to train the 1D-CNN positioning model by respectively using the second source domain data and the second target domain data as a training set and a test set to obtain an environment self-adaptation indoor positioning model.

9. An electronic device, comprising:

one or more processors;

a storage device to store one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-7.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-7.