Environment adaptive positioning method and device based on improved TCA algorithm and readable medium

Abstract

The application discloses an environment self-adaptive positioning method and device based on an improved TCA algorithm and a readable medium, obtains original CSI data and position coordinates of reference fingerprint points and to-be-positioned points, and performs pretreatment to obtain CSI data; a 1D-CNN model is used to extract feature data in the CSI data, and the feature data of the reference fingerprint points and corresponding position coordinates are stored in a position fingerprint database; the feature data of the reference fingerprint points is taken as first source domain feature data, the feature data of the to-be-positioned points is taken as first target domain feature data, data mapping and dimension reduction processing are performed through the improved TCA algorithm to obtain second source domain feature data and second target domain feature data, and the optimization target of the improved TCA algorithm includes a maximum mean difference distance value, an intra-class distance value and an inter-class distance value; the position coordinates of the to-be-positioned points are predicted through a double-layer KNN positioning algorithm according to the second source domain feature data, the second target domain feature data and the position coordinates, and the positioning accuracy can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of indoor positioning, in particular to an environment adaptive positioning method and device based on an improved TCA algorithm and a readable medium. BACKGROUND

[0002] Compared with satellite signal-based positioning technology, Wi-Fi-based indoor positioning technology has many advantages and has therefore received extensive attention. In early research, indoor positioning was based on received signal strength indication (RSSI), and the accuracy of indoor positioning based on RSSI has been greatly improved due to the continuous development of new algorithms. In recent years, with the release of CSITool, it has become possible to obtain channel state information in a WiFi environment, and therefore indoor positioning methods based on CSI as a feature parameter have received increasing attention.

[0003] In the indoor positioning method based on a location fingerprint database, there are mainly four steps of fingerprint data collection, data preprocessing, model establishment, and positioning data matching. The signal feature parameters at the positions of the reference points (RPs) in the required positioning area are collected, wherein the feature parameters include CSI data segments and the coordinates of the RPs. After the collected CSI signal segments are preprocessed, the corresponding relationship between the CSI segments and the coordinates is established according to a specific algorithm, and a location fingerprint database is generated. In the online positioning stage, the collected CSI signal segments are preprocessed and input into the trained model, and the user's location is obtained in a regression or classification manner according to the established model.

[0004] However, most of the existing research on indoor positioning assumes that the probability distribution of CSI data is stable and does not change, but this assumption does not hold in real indoor environments with complex compositions. With changes in various environmental conditions, the original model cannot accurately locate the user's position. The overall probability distribution of the data collected at two different time periods is not the same, so traditional machine learning cannot be directly applied to positioning operations. SUMMARY

[0005] In view of the above-mentioned problem of indoor positioning based on CSI data due to environmental changes and instability, which makes it difficult to accurately locate, the purpose of the embodiments of the present application is to propose an environment adaptive positioning method and device based on an improved TCA algorithm and a readable medium to solve the technical problems mentioned in the background section.

[0006] In a first aspect, the present application provides an environment adaptive positioning method based on an improved TCA algorithm, comprising the following steps:

[0007] S1, obtaining the original CSI data and position coordinates of the reference fingerprint points and the to-be-positioned points, pre-processing the original CSI data of the reference fingerprint points and the to-be-positioned points to obtain CSI data;

[0008] S2, using a 1D-CNN model to extract feature data in the CSI data, and storing the feature data of the reference fingerprint points and the corresponding position coordinates in a position fingerprint database;

[0009] S3, taking the feature data of the reference fingerprint points as first source domain feature data, taking the feature data of the to-be-positioned points as first target domain feature data, and performing data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data through an improved TCA algorithm to obtain second source domain feature data and second target domain feature data, wherein the optimization objective of the improved TCA algorithm includes a maximum mean difference distance value, an intra-class distance value and an inter-class distance value;

[0010] S4, taking the second source domain feature data and the second target domain feature data as domain-invariant feature data, and predicting the position coordinates of the to-be-positioned points through a double-layer KNN positioning algorithm according to the domain-invariant feature data and the position coordinates in the position fingerprint database.

[0011] Preferably, in step S1, the original CSI data of the reference fingerprint points and the to-be-positioned points is pre-processed to obtain the CSI data, which specifically comprises:

[0012] Hample filtering is used to remove obvious outliers and outliers in the original CSI data to obtain filtered original CSI data;

[0013] The filtered original CSI data is subjected to scale wavelet decomposition to obtain wavelet decomposition coefficients W, and the wavelet decomposition coefficients W are divided into approximate wavelet coefficients and detail wavelet coefficients ;

[0014] The wavelet decomposition coefficients are taken as absolute values and arranged from small to large, and the absolute values of each wavelet decomposition coefficient are squared to obtain an estimation vector For each element in the estimation vector , a risk vector is calculated according to the following formula:

[0015] ;

[0016] Wherein, represents the The risk vector of the element, finding the minimum point in the risk vector as the risk value, and recording the index value corresponding to the minimum point of the risk vector as The threshold value is calculated according to the following formula:

[0017] ;

[0018] The threshold shrinkage processing is performed on the detail wavelet coefficients obtained by the wavelet decomposition If the absolute value of the detail wavelet coefficient is lower than the threshold value, it is considered to be caused by noise, and the detail wavelet coefficient is set to 0; if the detail wavelet coefficient is greater than the threshold value, the shrinkage processing is performed on it, and the specific formula is as follows:

[0019] ;

[0020] Where represents the sign function, is the wavelet coefficient after wavelet threshold processing;

[0021] The wavelet inverse transform is performed on the wavelet coefficient and the approximate wavelet coefficient to obtain the CSI data.

[0022] As preferred, the 1D-CNN model comprises an input layer, a first composite layer, a second composite layer, a third composite layer, a first full connection layer, a second full connection layer and an output layer connected in sequence, the first composite layer comprises a first convolution layer and a first pooling layer connected in sequence, the second composite layer comprises a second convolution layer and a second pooling layer connected in sequence, and the third composite layer comprises a third convolution layer and a third pooling layer connected in sequence.

[0023] As preferred, the size of the convolution kernel in the first convolution layer is 5, the step is 2, the number of convolution kernels is 16, and the padding is set to 0, the size of the convolution kernel in the first pooling layer is 5, and the step is 1.

[0024] The size of the convolution kernel in the second convolution layer is 5, the step is 2, the number of convolution kernels is 32, and the padding is set to 0, the size of the convolution kernel in the second pooling layer is 5, and the step is 1.

[0025] The size of the convolution kernel in the third convolution layer is 5, the step is 1, the number of convolution kernels is 64, and the padding is set to 0, the size of the convolution kernel in the third pooling layer is 5, and the step is 1.

[0026] The number of neurons in the first full connection layer is 3264, and the number of neurons in the second full connection layer is 270.

[0027] As preferred, the several-dimensional feature data extracted in the second full connection layer are taken as the feature data.

[0028] As preferred, the step S3 specifically comprises:

[0029] The maximum mean difference MMD is adopted to calculate the distance between the first source domain feature data and the first target domain feature data, and the calculation formula is as follows:

[0030] ;

[0031] Wherein, is the first source domain feature data, is the first target domain feature data, and respectively represent the total data amount of the first source domain feature data and the total data amount of the first target domain feature data;

[0032] The distance calculation formula of the within-class distance value WCD of the t-th class sample in the first source domain feature data is as follows:

[0033] ;

[0034] The sum of the within-class distance of all class samples in the first source domain feature data is:

[0035] ;

[0036] Wherein, t i and t j are two samples in the t-th class sample.

[0037] The distance calculation formula of the between-class distance value BCD of the first source domain feature data is as follows:

[0038] ;

[0039] represents the feature mapping, which is used to map the original data into a high-dimensional space, and the mapping function is the domain adaptation function, is the dimension reduction matrix, is the dimension reduction vector;

[0040] The maximum mean difference MMD, the within-class distance value WCD and the between-class distance value BCD are taken as the measurement method to construct a multi-objective optimization problem, and the multi-objective optimization problem is represented by the following formula:

[0041] ;

[0042] Wherein, α, β and μ are influence factors, which are used to balance the importance of different distances;

[0043] The kernel function is introduced as and the kernel matrix K:

[0044] ;

[0045] The vector derived from the kernel function can be denoted as

[0046] ;

[0047] If the kernel matrix K is reduced to a low-dimensional latent space by a domain adaptation function , the reduced kernel matrix is denoted as:

[0048] ;

[0049] Since any vector in the latent space can be expressed as a linear combination of inner products of samples in the sample space, the following formula is established:

[0050] ;

[0051] The reduced kernel matrix can be denoted as:

[0052]

[0053]

[0054] ;

[0055] where is the domain adaptation matrix, and

[0056] The maximum mean difference MMD, the within-class distance value WCD, and the between-class distance value BCD are re-expressed as functions composed of :

[0057] ;

[0058] For simplicity, let

[0059] ;

[0060] Therefore is finally simplified as:

[0061] ;

[0062] Similarly, the within-class distance value WCD is simplified to the following formula:

[0063] ;

[0064] wherein, ;

[0065] Similarly, the inter-class distance value BCD is simplified to the following formula:

[0066] ;

[0067] wherein, ;

[0068] After sorting, the multi-objective optimization problem can be finally converted into the following form:

[0069] Adding the constraint of W to the optimization objective and the data divergence as the constraint of the optimization objective, the multi-objective optimization problem is converted into the following form:

[0070] ;

[0071] The extreme value is solved by using the Lagrange multiplier method, and the second source domain feature data and the second target domain feature data are obtained by solving the generalized eigenvalue decomposition problem. Assuming that the dimension value after dimension reduction is M, the second source domain feature data and the second target domain feature data are the eigenvectors corresponding to the first M eigenvalues of .

[0072] As preferred, step S4 specifically comprises:

[0073] The Mahalanobis distance between the second source domain feature data and the second target domain feature data is calculated, and the specific calculation formula of the Mahalanobis distance between two points is as follows:

[0074] ;

[0075] The K reference fingerprint points with the smallest Mahalanobis distance between the second source domain feature data and the second target domain feature data are taken, and the Mahalanobis distance calculated is arranged from small to large. Taking the position coordinates of the reference fingerprint point with the largest similarity as a reference, the Euclidean distance between the position coordinates corresponding to the remaining K-1 reference fingerprint points is calculated:

[0076] ;

[0077] The Euclidean distances of the K-1 position coordinates obtained Take the average value:

[0078] ;

[0079] According to the average value obtained, the standard deviation of the K-1 position coordinates is calculated: :

[0080] ;

[0081] According to the standard deviation obtained, the K reference fingerprint points selected are secondarily screened, and reference fingerprint points greater than the standard deviation are deleted, and only the nearest neighbor reference fingerprint points less than or equal to the standard deviation are retained. The coordinate values of the reference fingerprint points obtained after the secondary screening are averaged to obtain the position coordinates of the point to be positioned.

[0082] The coordinate values of the reference fingerprint points obtained after the secondary screening are averaged to obtain the position coordinates of the point to be positioned.

[0083] In a second aspect, the present application provides an environment adaptive positioning device based on an improved TCA algorithm, comprising:

[0084] A preprocessing module is configured to obtain original CSI data and position coordinates of reference fingerprint points and a point to be positioned, and to preprocess the original CSI data of the reference fingerprint points and the point to be positioned to obtain CSI data.

[0085] A feature extraction module is configured to extract feature data in the CSI data by using a 1D-CNN model, and to store the feature data of the reference fingerprint points and the corresponding position coordinates in a position fingerprint database.

[0086] A dimension reduction module is configured to take the feature data of the reference fingerprint points as first source domain feature data, to take the feature data of the point to be positioned as first target domain feature data, and to perform data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data by using an improved TCA algorithm to obtain second source domain feature data and second target domain feature data, wherein the optimization objectives of the improved TCA algorithm include a maximum mean discrepancy distance value, an intra-class distance value and an inter-class distance value.

[0087] A positioning module is configured to take the second source domain feature data and the second target domain feature data as domain-invariant feature data, and to predict the position coordinates of the point to be positioned by using a double-layer KNN positioning algorithm according to the domain-invariant feature data and the position coordinates in the position fingerprint database.

[0088] In a third aspect, the present application provides an electronic device, comprising one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementation manners of the first aspect.

[0089] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, when the computer program is executed by a processor, the method described in any of the implementation manners of the first aspect is implemented.

[0090] ​Compared with the prior art, the present application has the following beneficial effects:

[0091] (1) The environment adaptive positioning method based on the improved TCA algorithm proposed in the present application uses Hample filtering to remove obvious abnormal points and outliers in the original CSI data collected by the reference fingerprint point and the to-be-positioned point, and uses the wavelet denoising method based on the Rigrsure threshold value to smooth the data segments, thereby retaining the specific differentiated features of the position fingerprint, which is conducive to analyzing non-stationary data and extracting local features in the data.

[0092] (2) The environment adaptive positioning method based on the improved TCA algorithm proposed in the present application inputs the CSI data obtained after preprocessing into a 1D-CNN model to extract feature data, and constructs a position fingerprint database of the feature data and the coordinate position of the reference fingerprint point, which facilitates subsequent regression calculation for positioning.

[0093] (3) The environment adaptive positioning method based on the improved TCA algorithm proposed in the present application takes the feature data of the reference fingerprint point as the first source domain feature data and the feature data of the to-be-positioned point as the first target domain feature data, uses the improved TCA algorithm to map the first source domain feature data and the first target domain feature data to a high-dimensional space, and the probability distributions thereof are consistent, and then reduces the dimension to a low-dimensional latent space, takes the maximum mean difference distance value and the intra-class distance value as the optimization target, obtains the second source domain feature data and the second target domain feature data, and finally combines the coordinate position of the position fingerprint data to perform positioning, which can improve the positioning accuracy and still obtain accurate positioning in the case of environmental change and instability. BRIEF DESCRIPTION OF DRAWINGS

[0094] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0095] Figure 1 is an exemplary device architecture diagram to which an embodiment of the present application can be applied;

[0096] Figure 2 is a flowchart of the environment adaptive positioning method based on the improved TCA algorithm of the embodiment of the present application;

[0097] Figure 3 is a flowchart of the environment adaptive positioning method based on the improved TCA algorithm of the embodiment of the present application;

[0098] Figure 4This is a schematic diagram of an environment adaptive positioning device based on an improved TCA algorithm, according to an embodiment of this application.

[0099] Figure 5 This is a schematic diagram of the structure of a computer device suitable for implementing the electronic device of the present application. Detailed Implementation

[0100] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0101] Figure 1 An exemplary device architecture 100 is shown that can be applied to the environmental adaptive localization method or the environmental adaptive localization device based on the improved TCA algorithm according to the embodiments of this application.

[0102] like Figure 1 As shown, the device architecture 100 may include terminal devices 101, 102, and 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, and 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0103] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications, such as data processing applications and file processing applications, can be installed on terminal devices 101, 102, and 103.

[0104] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.

[0105] The server 105 can be a server providing various services, such as a background data processing server processing files or data uploaded by the terminal devices 101, 102, and 103. The background data processing server can process the obtained files or data to generate a processing result.

[0106] It should be noted that the environment adaptive positioning method based on the improved TCA algorithm provided in the embodiments of the present application can be executed by the server 105 or the terminal devices 101, 102, and 103, and correspondingly, the environment adaptive positioning apparatus based on the improved TCA algorithm can be arranged in the server 105 or the terminal devices 101, 102, and 103.

[0107] It should be understood that, Figure 1 The number of terminal devices, networks, and servers in the above apparatus architecture is only illustrative. Any number of terminal devices, networks, and servers can be provided according to implementation needs. In the case where the data to be processed does not need to be obtained remotely, the above apparatus architecture can not include a network, but only a server or a terminal device.

[0108] Figure 2 An environment adaptive positioning method based on an improved TCA algorithm provided by an embodiment of the present application is shown, which includes the following steps:

[0109] S1, obtaining original CSI data and position coordinates of reference fingerprint points and to-be-positioned points, and pre-processing the original CSI data of the reference fingerprint points and the to-be-positioned points to obtain CSI data.

[0110] In a specific embodiment, the original CSI data of the reference fingerprint points and the to-be-positioned points is pre-processed in step S1 to obtain the CSI data, which specifically includes:

[0111] The original CSI data is processed using Hample filtering to remove obvious abnormal points and outliers in the original CSI data to obtain filtered original CSI data;

[0112] The filtered original CSI data is subjected to scale wavelet decomposition to obtain wavelet decomposition coefficients W, and the wavelet decomposition coefficients W are divided into approximate wavelet coefficients and detail wavelet coefficients ;

[0113] The wavelet decomposition coefficients are taken as absolute values, arranged from small to large, and squared to obtain an estimated vector Each element in the estimated vector is calculated according to the following formula to obtain a risk vector:

[0114] ;

[0115] wherein, represents the risk vector of the th element, the minimum point in the risk vector is found as the risk value, and the subscript value corresponding to the minimum point of the risk vector is recorded as , and the threshold value is calculated according to the following formula:

[0116] ;

[0117] The threshold shrinkage processing is performed on the detail wavelet coefficients obtained through the wavelet decomposition, if the absolute value of the detail wavelet coefficient is lower than the threshold value, it can be considered that it is caused by noise, and the detail wavelet coefficient is set to 0; if the detail wavelet coefficient is greater than the threshold value, the shrinkage processing is performed on it, and the specific formula is as follows:

[0118] ;

[0119] wherein represents a sign function, is the wavelet coefficient after the wavelet threshold processing;

[0120] The wavelet inverse transform is performed on the wavelet coefficient and the approximate wavelet coefficient after the wavelet threshold processing to obtain the CSI data.

[0121] Specifically, the position fingerprint data of each reference fingerprint point and the to-be-positioned point is collected. The position fingerprint data is composed of two parts, which are the original CSI data and the position coordinates. The original CSI data mentioned in the embodiments of the present application can be regarded as the amplitude data in the CSI signal extracted from the Wi-Fi signal using the Intel WiFi Link 5300 network card. In order to solve the problem of lack of frequency domain resources and improve the dimension of available data, the embodiments of the present application introduce a MIMO system, taking an example of having transmit antennas and receive antennas, since each CSI signal contains at most 30 subcarriers, the original CSI data mentioned in the embodiments of the present application can be regarded as a one-dimensional vector with a size of N .

[0122] Due to the existence of many problems such as the hardware defects of commercial Wi-Fi devices, various electromagnetic interferences in the environment, and inevitable white noise, the collected original CSI data contains many noise components, and if it is not processed, it will seriously affect the establishment of the indoor positioning model. Therefore, after the original CSI data is collected, the noise reduction processing of the data must be performed. There are many noise reduction methods, and the most commonly used is the combination of various filters, which is used to extract the data components.

[0123] For details, please refer to Figure 3 Hample filtering is used to remove obvious outliers and isolated points from the original CSI data segments. The maximum likelihood estimation method is used to impute missing values. Wavelet denoising based on Rigrsure thresholding is used to smooth the data segments and remove noise, thereby preserving the specific differential features of the location fingerprint. This is beneficial for analyzing non-stationary data and extracting local features from the data.

[0124] Specifically, the steps for using Hample to detect and remove outliers include: processing the original CSI data... ( The amplitude of a subcarrier in the CSI signal is used as a sequence. An observation window is generated around each element. Assuming the length of half a window is d, the width of the entire window is 2d+1 (including the center element). The median of all elements in the window is calculated. The standard deviation of each sample relative to the median is estimated using the median. If the difference between a sample and the median is greater than three standard deviations, the sample is replaced by the median.

[0125] In real-world scenarios, the acquired CSI signal will not contain only one type of noise signal; it will often be mixed with noise of varying degrees from the channel. A length of... Noisy CSI signals The measured noisy data is ( The original signal, (This is a noise signal). Due to noise interference, direct use will lead to unsatisfactory positioning accuracy. Therefore, the signal must be filtered.

[0126] In environments where CSI signals are acquired, there is often a lot of random noise in the actual home environment. Therefore, the wavelet thresholding denoising method, which is commonly used and effective in wavelet transform, is used to process the raw CSI data. The denoising process of this filtering algorithm is as follows:

[0127] First, the filtered raw CSI data is subjected to scaled wavelet decomposition, specifically including: performing DWT transform on the noisy signal. In the embodiments of this application, the Daubechies 10 wavelet basis is used to decompose the filtered raw CSI data to the fifth level to obtain the required wavelet decomposition coefficients W. The wavelet decomposition coefficients W are further subdivided into two categories: approximate wavelet coefficients. and detailed wavelet coefficients Two categories. Then, the detail wavelet coefficients are processed using wavelet thresholding to obtain the wavelet thresholded detail wavelet coefficients. Finally, wavelet coefficient reconstruction is used to reconstruct the detailed wavelet coefficients after wavelet thresholding. approximate wavelet coefficients performing inverse wavelet transform to reconstruct the signal, and the reconstructed signal can be regarded as the original CSI data after removing noise interference.

[0128] S2, a 1D-CNN model is used to extract feature data in the CSI data, and the feature data of the reference fingerprint point and the corresponding position coordinates are stored in the position fingerprint database.

[0129] In specific embodiments, the 1D-CNN model includes an input layer, a first composite layer, a second composite layer, a third composite layer, a first full connection layer, a second full connection layer and an output layer connected in sequence, the first composite layer includes a first convolution layer and a first pooling layer connected in sequence, the second composite layer includes a second convolution layer and a second pooling layer connected in sequence, and the third composite layer includes a third convolution layer and a third pooling layer connected in sequence.

[0130] In specific embodiments, the size of the convolution kernel in the first convolution layer is 5, the step is 2, the number of convolution kernels is 16, and padding is set to 0, and the size of the convolution kernel in the first pooling layer is 5, the step is 1;

[0131] The size of the convolution kernel in the second convolution layer is 5, the step is 2, the number of convolution kernels is 32, and padding is set to 0, and the size of the convolution kernel in the second pooling layer is 5, the step is 1;

[0132] The size of the convolution kernel in the third convolution layer is 5, the step is 1, the number of convolution kernels is 64, and padding is set to 0, and the size of the convolution kernel in the third pooling layer is 5, the step is 1;

[0133] The number of neurons in the first full connection layer is 3264, and the number of neurons in the second full connection layer is 270.

[0134] Specifically, the 1D-CNN model is a trained model, and the CSI data obtained after preprocessing is input into the 1D-CNN model, and the feature data extracted from the second full connection layer of the 1D-CNN model is used as the feature data. The feature data and position coordinates of the reference fingerprint point are used to construct a position fingerprint database.

[0135] S3, the feature data of the reference fingerprint point is used as the first source domain feature data, and the feature data of the to-be-positioned point is used as the first target domain feature data, and the improved TCA algorithm is used for data mapping and dimension reduction processing according to the first source domain feature data and the first target domain feature data, to obtain the second source domain feature data and the second target feature domain data. The optimization target of the improved TCA algorithm includes the maximum mean difference distance value, the intra-class distance value and the inter-class distance value.

[0136] In specific embodiments, step S3 specifically includes:

[0137] The distance between the first source domain feature data and the first target domain feature data is calculated by using the maximum mean difference MMD, and the calculation formula is as follows:

[0138]

[0139] Wherein, is the first source domain feature data, is the first target domain feature data, and respectively represent the total data amount of the first source domain feature data and the total data amount of the first target domain feature data;

[0140] The distance calculation formula of the intra-class distance value WCD of the t-th class sample in the first source domain feature data is as follows:

[0141]

[0142] The sum of the intra-class distances of all class samples in the first source domain feature data is:

[0143]

[0144] Wherein, t i and t j are two of the t-th class samples.

[0145] The distance calculation formula of the inter-class distance value BCD of the first source domain feature data is as follows:

[0146]

[0147] represents a feature mapping, which is used to map the original data into a high-dimensional space, and the mapping function is a domain adaptation function, is a dimension reduction matrix, is a dimension reduction vector;

[0148] The maximum mean difference MMD, the intra-class distance value WCD and the inter-class distance value BCD are used as the measurement method to construct a multi-objective optimization problem, and the multi-objective optimization problem is represented by the following formula:

[0149]

[0150] Wherein, α, β and μ are influence factors, which are used to balance the importance of different distances;

[0151] The kernel function is introduced as , and the kernel matrix K is:

[0152] ;​​​​

[0153] By , the kernel function The vector derived from

[0154] ;

[0155] If the kernel matrix K is reduced to a low-dimensional latent space by a domain adaptation function , then the reduced kernel matrix is represented as:

[0156] ;

[0157] Since any vector in the latent space can be represented as a linear sum of inner products of samples in the sample space, the following equation holds:

[0158] ;

[0159] By this equation, the reduced kernel matrix can be represented as:

[0160]

[0161]

[0162] ;

[0163] where is the domain adaptation matrix, which is the weight matrix of the samples;

[0164] The maximum mean discrepancy MMD, the within-class distance value WCD, and the between-class distance value BCD are re-expressed as functions composed of :

[0165] ;

[0166] For simplicity, let:

[0167] ;

[0168] Therefore is finally simplified to:

[0169] ;

[0170] The within-class distance value WCD is simplified, and finally we get:

[0171] ;

[0172] where ;

[0173] Similarly, the inter-class distance value BCD is simplified to obtain the following formula:

[0174] ;

[0175] wherein, ;

[0176] After sorting, the multi-objective optimization problem can be finally converted into the following form:

[0177] To make the result of W not too complex, the constraint of W is added to the optimization objective To make the mapped data maintain their respective data characteristics, the data divergence is added as a constraint of the optimization objective to ensure the consistency of the data. After sorting the optimization objectives, the multi-objective optimization problem can be finally converted into the following form:

[0178] ;

[0179] The extreme value is solved by using the Lagrange multiplier method, and the second source domain feature data and the second target domain feature data are obtained by solving the generalized eigenvalue decomposition problem. If the dimension value after dimension reduction is M, then the second source domain feature data and the second target domain feature data are the eigenvectors corresponding to the first M eigenvalues of .

[0180] Specifically, the improved TCA algorithm proposed by the embodiments of the present application is aimed at the case where the source domain and the target domain are in different data distributions in the Domain Adaptation problem. It is assumed that there is a high-dimensional space in which the overall probability distribution difference between the source domain and the target domain is the smallest and the intra-class distance of each class of data in the source domain is also the smallest, and the inter-class distance is the largest, so that the isomorphic data after mapping is as close as possible. The improved TCA algorithm proposed by the embodiments of the present application is actually to find a high-dimensional space in which the data collected in two different time periods are consistent in probability distribution after mapping, i.e., the traditional memory learning algorithm can be used for positioning, or the regression calculation of the position coordinates can be realized. Therefore, the second source domain feature data and the second target domain feature data obtained after dimension reduction can be used as domain-invariant feature data.

[0181] In order to measure the overall probability distribution and the size of the difference between classes, the Maximum Mean Discrepancy (MMD) and the Between-Classes Distance (BCD) and the Within-Classes Distance are introduced in the embodiments of the present application as a measurement method. The distance between the source domain and the target domain is calculated by the Maximum Mean Discrepancy MMD. Further, the inter-class distance of the class samples in the source domain is calculated by the Between-Classes Distance, and the intra-class distance of the class samples in the source domain is calculated by the Within-Classes Distance, and a multi-objective optimization problem is constructed, and the optimization goal of the multi-objective optimization problem is to minimize the MMD distance between the first source domain feature data and the first target domain feature data, and also to make the intra-class distance of the first source domain feature data as small as possible and the inter-class distance as large as possible.

[0182] S4, taking the second source domain feature data and the second target domain feature data as domain-invariant feature data, predicting the position coordinates of the to-be-positioned point by a double-layer KNN positioning algorithm according to the domain-invariant feature data and the position coordinates in the position fingerprint database.

[0183] In specific embodiments, step S4 specifically includes:

[0184] The Mahalanobis distance of the second source domain feature data and the second target domain feature data is calculated respectively, and the specific calculation formula of the Mahalanobis distance between two points is as follows:

[0185] ;

[0186] Taking the K reference fingerprint points with the smallest Mahalanobis distance between the second source domain feature data and the second target domain feature data, arranging the K reference fingerprint points in ascending order according to the calculated Mahalanobis distance, and taking the position coordinates of the reference fingerprint point with the largest similarity as a reference, calculating the Euclidean distance between the position coordinates corresponding to the remaining K-1 reference fingerprint points:

[0187] ;

[0188] Calculating the Euclidean distance of the K-1 position coordinates Taking the mean value:

[0189] ;

[0190] According to the obtained mean value, calculating the standard deviation of the K-1 position coordinates :

[0191] ;

[0192] ​According to the standard deviation obtained, the K reference fingerprint points selected are subjected to secondary screening, and reference fingerprint points greater than the standard deviation are deleted , and only reference fingerprint points less than or equal to the standard deviation are retained.

[0193] The coordinate values of the reference fingerprint points obtained after secondary screening are averaged to obtain the position coordinates of the point to be positioned.

[0194] Specifically, the second source domain feature data is subjected to double-layer nested KNN screening to obtain several reference fingerprint points with the highest similarity to the first target domain feature data, and the standard deviation is used for secondary screening. The position coordinates corresponding to the reference fingerprint points obtained after secondary screening are averaged to obtain the position coordinates of the point to be positioned.

[0195] Further reference Figure 4 , as an implementation of the method shown in the above figures, the present application provides an embodiment of an environment adaptive positioning device based on an improved TCA algorithm. The device embodiment corresponds to the method embodiment shown in Figure 2 . The device can be applied to various electronic devices.

[0196] The present application embodiment provides an environment adaptive positioning device based on an improved TCA algorithm, which comprises:

[0197] A preprocessing module 1 configured to obtain original CSI data and position coordinates of reference fingerprint points and a point to be positioned, and to preprocess the original CSI data of the reference fingerprint points and the point to be positioned to obtain CSI data.

[0198] A feature extraction module 2 configured to extract feature data in the CSI data using a 1D-CNN model, and to store the feature data of the reference fingerprint points and the corresponding position coordinates in a position fingerprint database.

[0199] A dimension reduction module 3 configured to take the feature data of the reference fingerprint points as first source domain feature data, and the feature data of the point to be positioned as first target domain feature data, and to perform data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data by an improved TCA algorithm to obtain second source domain feature data and second target domain feature data. The optimization objectives of the improved TCA algorithm include maximum mean difference distance value, intra-class distance value, and inter-class distance value.

[0200] A positioning module 4 configured to take the second source domain feature data and the second target domain feature data as domain-invariant feature data, and to predict the position coordinates of the point to be positioned by a double-layer KNN positioning algorithm according to the domain-invariant feature data and the position coordinates in the position fingerprint database.

[0201] Reference will now be made toFigure 5 It illustrates an electronic device suitable for implementing embodiments of this application (e.g., Figure 1 A schematic diagram of the structure of a computer device 500 (shown as a server or terminal device). Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0202] like Figure 5 As shown, the computer device 500 includes a central processing unit (CPU) 501 and a graphics processing unit (GPU) 502, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 503 or programs loaded from storage section 509 into random access memory (RAM) 504. The RAM 504 also stores various programs and data required for the operation of the device 500. The CPU 501, GPU 502, ROM 503, and RAM 504 are interconnected via a bus 505. An input / output (I / O) interface 506 is also connected to the bus 505.

[0203] The following components are connected to I / O interface 506: an input section 507 including a keyboard, mouse, etc.; an output section 508 including an LCD, speakers, etc.; a storage section 509 including a hard disk, etc.; and a communication section 510 including a network interface card, such as a LAN card or modem. The communication section 510 performs communication processing via a network such as the Internet. A drive 511 may also be connected to I / O interface 506 as needed. A removable medium 512, such as a hard disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 511 as needed so that computer programs read from it can be installed into storage section 509 as needed.

[0204] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 510, and / or installed from removable medium 512. When the computer program is executed by central processing unit (CPU) 501 and graphics processing unit (GPU) 502, the functions defined in the methods of this application are performed.

[0205] Note that the computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a computer-readable medium can 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 context, a computer-readable medium can be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present context, a computer-readable signal medium can include a computer-readable program code in a baseband or propagated as carrier waves in a propagated data signal associating with a carrier wave. Such a propagated signal can take a wide variety of forms including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium that can be used to carry or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The program contained in the computer-readable medium can be transmitted in any suitable format including, but not limited to, wireless, wireline, optical fiber cable, RF, or any suitable combination thereof.

[0206] Computer program code for carrying out operations of the present application can 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 can 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 latter scenario, the remote computer can 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 can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0207] The flow and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flow and block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can 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 be executed in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by a dedicated hardware-based device, or a combination of dedicated hardware-based devices and computer instructions.

[0208] The modules involved in the embodiments of the present application can be implemented in the form of software or in the form of hardware. The described modules can also be arranged in a processor.

[0209] As another aspect, the present application also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist independently 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: acquire original CSI data and position coordinates of reference fingerprint points and points to be positioned, pre-process the original CSI data of the reference fingerprint points and the points to be positioned to obtain CSI data; extract feature data in the CSI data using a 1D-CNN model, and store the feature data of the reference fingerprint points and the corresponding position coordinates in a position fingerprint database; take the feature data of the reference fingerprint points as first source domain feature data, take the feature data of the points to be positioned as first target domain feature data, perform data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data according to an improved TCA algorithm to obtain second source domain feature data and second target domain feature data, and the optimization objectives of the improved TCA algorithm include maximum mean difference distance value, intra-class distance value and inter-class distance value; take the second source domain feature data and the second target domain feature data as domain-invariant feature data, and predict the position coordinates of the points to be positioned according to the domain-invariant feature data and the position coordinates in the position fingerprint database through a double-layer KNN positioning algorithm.

[0210] The above description is only the preferred embodiment of the present application and the explanation of the technical principles. It should be understood by those skilled in the art that the scope of the protection of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features. It should also cover other technical solutions formed by the combinations of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solutions formed by the mutual replacement of the above features and the technical features with similar functions disclosed (but not limited to) in the present application.

Claims

1. An environment adaptive positioning method based on improved TCA algorithm, characterized in that, The method comprises the following steps: S1, obtaining the original CSI data and position coordinates of the reference fingerprint points and the to-be-positioned points, and preprocessing the original CSI data of the reference fingerprint points and the to-be-positioned points to obtain CSI data; S2, extracting feature data in the CSI data by using a 1D-CNN model, and storing the feature data of the reference fingerprint points and the corresponding position coordinates in a position fingerprint database; S3, taking the feature data of the reference fingerprint points as first source domain feature data, taking the feature data of the to-be-positioned points as first target domain feature data, performing data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data by using an improved TCA algorithm to extract domain-invariant feature data, second source domain feature data and second target domain feature data, and the optimization target of the improved TCA algorithm comprises a maximum mean difference distance value, an intra-class distance value and an inter-class distance value; S4, taking the second source domain feature data and the second target domain feature data as domain-invariant feature data, and predicting the position coordinates of the to-be-positioned points by using a double-layer KNN positioning algorithm according to the domain-invariant feature data and the position coordinates in the position fingerprint database, specifically comprising: calculating Mahalanobis distances of the second source domain feature data and the second target domain feature data respectively, and the specific calculation formula of the Mahalanobis distance between two points is as follows: ； taking K reference fingerprint points with the smallest Mahalanobis distances of the second source domain feature data and the second target domain feature data, arranging the K reference fingerprint points in ascending order according to the calculated Mahalanobis distances, and taking the position coordinates of the reference fingerprint points with the largest similarity as a reference to calculate the Euclidean distances between the position coordinates of the reference fingerprint points and the position coordinates of the remaining K-1 reference fingerprint points; ； calculating the Euclidean distance of the K-1 position coordinates obtained taking the mean value: ； The standard deviation of the K-1 position coordinates is calculated from the obtained mean values :​ ； According to the standard deviation obtained, the selected K reference fingerprint points are twice screened, and the reference fingerprint points greater than the standard deviation are deleted Only the nearest neighbor reference fingerprint points less than or equal to the standard deviation are retained Only the nearest neighbor reference fingerprint points less than or equal to the standard deviation are retained calculating the average value of the coordinate values of the reference fingerprint points obtained after the secondary screening to obtain the position coordinates of the to-be-positioned points.

2. The improved TCA algorithm based environment adaptive positioning method according to claim 1, characterized in that, The preprocessing of the original CSI data of the reference fingerprint points and the to-be-positioned points in step S1 to obtain the CSI data specifically comprises: using Hample filtering to remove obvious abnormal points and outliers in the original CSI data to obtain filtered original CSI data; The filtered original CSI data is decomposed by a scale wavelet to obtain wavelet decomposition coefficients W, and the wavelet decomposition coefficients W are divided into approximate wavelet coefficients and detail wavelet coefficients ; Taking absolute values of the wavelet decomposition coefficients and arranging them from small to large, and squaring the absolute values of each wavelet decomposition coefficient to obtain an estimated vector For each element in the estimated vector , a risk vector is calculated according to the following formula: ； wherein, represents the risk vector of the i-th element, the minimum point in the risk vector is found as the risk value, and the subscript value corresponding to the minimum point of the risk vector is denoted as The threshold value is calculated according to the following formula:​ ； The detail wavelet coefficients obtained by the scale wavelet decomposition The threshold shrinkage processing is performed, if the absolute value of the detail wavelet coefficient is lower than the threshold value, it is considered to be caused by noise, and the detail wavelet coefficient is set to 0; if the detail wavelet coefficient is larger than the threshold value, the shrinkage processing is performed on it, and the specific formula is as follows: ； wherein a representative symbol function, are wavelet coefficients after wavelet threshold processing; wavelet coefficients after the wavelet threshold processing and approximation wavelet coefficients performing inverse wavelet transform to obtain the CSI data.

3. The improved TCA algorithm based environment adaptive positioning method according to claim 1, wherein, The 1D-CNN model comprises an input layer, a first composite layer, a second composite layer, a third composite layer, a first full connection layer, a second full connection layer and an output layer connected in sequence, the first composite layer comprises a first convolution layer and a first pooling layer connected in sequence, the second composite layer comprises a second convolution layer and a second pooling layer connected in sequence, and the third composite layer comprises a third convolution layer and a third pooling layer connected in sequence.

4. The environment adaptive positioning method based on improved TCA algorithm according to claim 3, characterized in that, In the first convolution layer, the convolution kernel size is 5, the step is 2, the number of convolution kernels is 16, and padding is 0; in the first pooling layer, the convolution kernel size is 5, and the step is 1; In the second convolution layer, the convolution kernel size is 5, the step is 2, the number of convolution kernels is 32, and padding is 0; in the second pooling layer, the convolution kernel size is 5, and the step is 1; In the third convolution layer, the convolution kernel size is 5, the step is 1, the number of convolution kernels is 64, and padding is 0; in the third pooling layer, the convolution kernel size is 5, and the step is 1; The number of neurons of the first full connection layer is 3264, and the number of neurons of the second full connection layer is 270.

5. The improved TCA algorithm based environment adaptive positioning method according to claim 3, wherein, The several-dimensional feature data extracted in the second full connection layer are taken as the feature data.

6. The improved TCA algorithm based environment adaptive positioning method according to claim 1, wherein, The step S3 specifically comprises: The edge probability distribution difference of the first source domain feature data and the first target domain feature data is calculated by using the maximum mean difference MMD, and the calculation formula is as follows: ； wherein, is the first source domain feature data, is the first target domain feature data, and respectively represent a total data amount of the first source domain feature data and a total data amount of the first target domain feature data. The distance calculation formula of the within-class distance value WCD of the t-th class sample in the first source domain feature data is as follows: ； The sum of the within-class distances of all class samples in the first source domain feature data is: ； where t i and t j are two samples in the t-th class respectively. The distance calculation formula of the between-class distance value BCD of the first source domain feature data is as follows: ； representing a feature mapping for mapping the raw data into a high dimensional space; The multi-objective optimization problem is constructed by taking the maximum mean difference MMD, the within-class distance value WCD and the between-class distance value BCD as the measurement method, and the multi-objective optimization problem is represented by the following formula: ； where a, b and m are influence factors for balancing the importance of different distances, the mapping function is a domain adaptation function, is a dimension reduction matrix, is a dimension reduction vector, ; The kernel function is introduced as and the kernel matrix K: ； By , the kernel function The vector derived from this can be written as ； If the kernel matrix K is reduced to a low-dimensional latent space by a domain adaptation function K' = K * f, then the reduced kernel matrix is represented as: ； Any vector in the latent space Can be represented as a linear sum of inner products in the sample space, so that the following equation holds: ； By this equation, the reduced dimensionality kernel matrix is represented as: ； wherein is the domain adaptation matrix; The maximum mean difference MMD, the within-class distance value WCD and the between-class distance value BCD are re-expressed as functions of the following: ； For simplification, let ； Therefore Simplifies to: ； Similarly, the within-class distance value WCD is simplified to obtain the following formula: ； wherein ; Similarly, the between-class distance value BCD is simplified to obtain the following formula: ； wherein ; Adding a constraint on W to the optimization objective and data divergence as a constraint on the optimization objective, the multi-objective optimization problem is transformed into the following form: ； The extreme value is solved by using Lagrange multiplier method, the second source domain feature data and the second target domain feature data are obtained by solving the generalized eigenvalue decomposition problem, and the dimension value after dimension reduction is set as M, so that the second source domain feature data and the second target domain feature data are eigenvectors corresponding to the first M eigenvalues of .

7. An environment-adaptive positioning device based on an improved TCA algorithm, characterized in that, comprise: The pre-processing module is configured to obtain original CSI data and position coordinates of reference fingerprint points and to-be-positioned points, pre-process the original CSI data of the reference fingerprint points and to-be-positioned points, and obtain CSI data. The feature extraction module is configured to extract feature data in the CSI data by using a 1D-CNN model, and store the feature data of the reference fingerprint points and corresponding position coordinates in a position fingerprint database. The dimension reduction module is configured to take the feature data of the reference fingerprint points as first source domain feature data, take the feature data of the to-be-positioned points as first target domain feature data, perform data mapping and dimension reduction processing on the first source domain feature data and the first target domain feature data by using an improved TCA algorithm, and obtain second source domain feature data and second target domain feature data, wherein the optimization objectives of the improved TCA algorithm comprise a maximum mean difference distance value, a within-class distance value and a between-class distance value. The positioning module is configured to take the second source domain feature data and the second target domain feature data as domain-invariant feature data, predict the position coordinates of the to-be-positioned points by using a double-layer KNN positioning algorithm based on the domain-invariant feature data and the position coordinates in the position fingerprint database, and specifically comprises: The Mahalanobis distance between the second source domain feature data and the second target domain feature data is calculated, and the specific calculation formula of the Mahalanobis distance between two points is as follows: ； K reference fingerprint points with the smallest Mahalanobis distance are taken from the second source domain feature data and the second target domain feature data, and the position coordinates of the reference fingerprint points with the largest similarity are taken as references to calculate the Euclidean distance between the position coordinates of the remaining K-1 reference fingerprint points. ； calculating the Euclidean distance of the K-1 position coordinates obtained taking the mean value: ； The standard deviation of the K-1 position coordinates is calculated from the obtained mean values :​ ； According to the standard deviation obtained, the selected K reference fingerprint points are twice screened, and the reference fingerprint points greater than the standard deviation are deleted Only the nearest neighbor reference fingerprint points less than or equal to the standard deviation are retained Only the nearest neighbor reference fingerprint points less than or equal to the standard deviation are retained The position coordinates of the to-be-positioned points are obtained by averaging the coordinate values of the reference fingerprint points after the secondary screening.

8. An electronic device, comprising: one or more processors; storage means for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement a method as claimed in any of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program, which when executed by a processor, implements a method as claimed in any of claims 1-6.

Images