CSI fingerprint database optimization method and system based on depth and width learning

Abstract

The invention provides a CSI fingerprint database optimization method and system based on depth and width learning, and belongs to the technical field of CSI fingerprint data identification. A fine-grained CSI fingerprint database constructed based on confidence coefficients calculated based on DeepBLS and GAN is combined with a k-nearest neighbor algorithm to carry out fingerprint matching, and the position of a user is estimated. The method comprises the following steps: collecting original CSI fingerprint data at a sampling point to fit a multivariate Gaussian model; using a path planning algorithm to collect additional original CSI data along the optimal path; extracting features by adopting a DeepBLS network structure, and generating CSI data distribution values of residual sampling points; the accuracy of the predicted fingerprints is improved through the GAN, and a fine-grained fingerprint database is constructed by using the confidence coefficient, so that the positioning accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of CSI fingerprint data recognition, and in particular to a CSI fingerprint database optimization method and system based on depth and width learning. Background Art

[0002] With the rapid development of 6G synaesthesia integration, location-based services are becoming an indispensable part of people's daily lives. Studies have shown that people spend about 80% of their time indoors, which further highlights the importance of indoor positioning technology. Among the many existing technologies, location fingerprint technology has attracted much attention due to its simple implementation, efficient measurement process and economic cost-effectiveness. Location fingerprint technology is divided into two key stages: offline data collection and construction and online real-time positioning. In the offline stage, through grid sampling, professionals collect and store key fingerprint data in the database, laying a solid foundation for the system. In the online stage, the system uses user real-time data to match the fingerprint database to quickly and accurately locate the user, ensuring the efficiency and accuracy of positioning. Among the many location fingerprint technologies, positioning technology based on WiFi fingerprint has become the mainstream solution due to its wide applicability. The received signal strength (RSS) and channel state information (CSI) of WiFi signals can both be used as fingerprint data. In particular, CSI, as the physical layer information of WiFi signals, provides richer and more detailed feature information than RSS. This gives CSI-based positioning technology significant advantages in accuracy and performance, providing a strong impetus for the further development and application of indoor positioning technology.

[0003] In the practical application of indoor positioning technology, accurate updating of the fingerprint database in the offline stage is crucial, which is directly related to the final performance of the positioning system. When building and updating the fingerprint database, people generally adopt crowdsourcing or prediction-based methods. Crowdsourcing strategies achieve large-scale collection of fingerprint data by combining with daily activities, effectively reducing the labor intensity of manual collection. However, the constant changes in equipment and environment will cause data heterogeneity, which not only poses a challenge to the stability and accuracy of the fingerprint database, but also brings additional complexity to data processing. Prediction-based machine learning algorithms can extract features from a large amount of CSI data, thereby significantly improving the accuracy and robustness of positioning. Although machine learning has shown great potential in the field of indoor positioning, its complex algorithms have high requirements for computing resources, especially in the feature extraction stage. At the same time, effective machine learning models rely on a large amount of historical data for training to ensure the generalization ability of the model. Especially in new environments or when data sampling is insufficient, the lack of training data will affect the accuracy of fingerprint database updates.

[0004] Location fingerprint technology has attracted much attention in indoor positioning due to its advantages in implementation, measurement and cost-effectiveness. In the offline stage, professionals measure and store fingerprint data through grid sampling; in the online stage, mobile devices send real-time fingerprint data to the positioning algorithm, and use the fingerprint database to estimate the user's location and achieve positioning. However, the construction and update of the existing offline fingerprint database has the following problems: First, when the fingerprint database is updated using crowdsourcing technology in the offline stage, it is easily affected by noise interference and incomplete coverage, which will lead to the degradation of the quality of the fingerprint database and thus affect the positioning accuracy. Secondly, fitting the Gaussian regression model requires a lot of computing resources, which is difficult to efficiently handle a large number of data update requirements, increasing the complexity and computing cost of the system. In addition, the use of complex machine learning algorithms to extract fingerprint data features in the offline stage also requires a lot of computing resources. These algorithms may not be able to adapt to the new environment, resulting in inaccurate final positioning results. Finally, when updating the fingerprint database, the existing methods often fail to strike a balance between accuracy and performance, resulting in unsatisfactory positioning results and difficulty in meeting actual application needs. Summary of the invention

[0005] The object of the present invention is to provide a CSI fingerprint database optimization method and system based on deep and wide learning to solve at least one technical problem existing in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a CSI fingerprint database optimization method based on deep and wide learning, comprising:

[0008] Get CSI information;

[0009] A fine-grained CSI fingerprint database constructed based on the confidence coefficient calculated by DeepBLS and GAN is used to perform fingerprint matching in combination with the k-nearest neighbor algorithm to estimate the user's location. Part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. The DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0010] Furthermore, the DeepBLS network is constructed by combining the efficiency of width learning and the high precision of deep learning; DeepBLS is used to extract features and build an initial coarse-grained fingerprint database; width learning captures more input data features by expanding the network width, enabling it to better adapt to training and test data, thereby enhancing its generalization ability; in the DeepBLS network structure, each layer consists of a single width learning block, and each width learning block includes a feature layer, an enhancement layer, and an output layer; the dataset is trained only once on each width learning block.

[0011] Furthermore, at the feature layer, for n feature maps, each map generates k nodes, thereby mapping the input available data x to n sets of feature nodes:

[0012]

[0013] Where X represents the feature matrix of the input data, and the weight W e and the bias term β e is randomly generated. is the hyperbolic tangent sigmoid nonlinear activation function; after that, all feature nodes are recorded as Z n ≡[Z1,Z2,…,Z n ].

[0014] Furthermore, the feature nodes are connected to the enhancement nodes, which improve the feature representation through nonlinear transformations. These transformations use randomly generated weights W h , bias term β h and nonlinear activation function ψ(·);

[0015] The output of the enhancement node is H m ≡[H1,H2,…,H m ],in

[0016] All feature nodes and enhanced nodes are connected by weights W m Connect to the output layer node to synthesize the extracted and enhanced features to obtain the final expected output of the current width learning block

[0017] The connection weights W in each width learning layer m are the only weights that need to be calculated, and they are calculated using Derived, solve [Z n |H m ] + The process is approximately a ridge regression problem:

[0018]

[0019] Furthermore, in the DeepBLS network structure, the kth width learning block is trained using the initial parameter setting ω0; after the training process is completed, the output of the first width learning block is recorded as f BLS (X; ω0); After obtaining the output f of the first width learning block BLS After (X; ω0), we subtract it from the true label Y to get the first training residual r:

[0020] r (k) =Yf BLS (X;ω0);

[0021] In the formula, X represents the feature matrix of the input data, and ω0 represents the model parameters;

[0022] The residual of the k-layer DeepBLS network is:

[0023]

[0024] in represents the output of the n-layer DeepBLS network, which is the cumulative sum of the outputs of the first k-1 width learning blocks;

[0025] Finally, the model's prediction result is the sum of the prediction results of each width learning block:

[0026]

[0027] Furthermore, GAN consists of a generator and a discriminator;

[0028] Generator G receives random noise z, labels y, and generates fake data Where z is random noise sampled from a normal distribution, y is the label associated with the input noise, is the fake data output by the generator; the discriminator D receives the real data x real and fake data Output discriminator loss d loss , which measures the discriminator's ability to distinguish real data from fake data:

[0029]

[0030] Where D(x real ,y) and Represent real data and false data respectively;

[0031] Generator G receives random noise z and generates fake data It outputs the generator loss g loss , measuring the ability of the generator to generate fake data similar to real data:

[0032] g loss =D(G(z,y),y),

[0033] The goal of the generator is to adjust the generated data to make it closer to the original true value distribution:

[0034]

[0035] In the formula, is the mean of the generated data, is the standard deviation of the generated data, μ x is the mean of the original data, σ x is the standard deviation of the original data.

[0036] Furthermore, the confidence coefficients calculated by DeepBLS and GAN are used to construct a fine-grained CSI fingerprint database:

[0037] highFinger=c·GAN data+(1-c)·final p redictions.

[0038] In a second aspect, the present invention provides a CSI fingerprint database optimization system based on deep and wide learning, comprising:

[0039] An acquisition module, used to obtain CSI information;

[0040] The matching module is used to use the fine-grained CSI fingerprint database built based on the confidence coefficient calculated by DeepBLS and GAN, combined with the k-nearest neighbor algorithm to perform fingerprint matching and estimate the user's location; wherein, part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network; wherein the DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer; GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0041] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the CSI fingerprint database optimization method based on deep and wide learning as described in the first aspect is implemented.

[0042] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the CSI fingerprint database optimization method based on deep and wide learning as described in the first aspect.

[0043] In a fifth aspect, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the CSI fingerprint database optimization method based on deep and wide learning as described in the first aspect.

[0044] The beneficial effects of the present invention are as follows: collecting original CSI fingerprint data at sampling points to fit a multivariate Gaussian model; using a path planning algorithm to collect additional original CSI data along the optimal path; using a DeepBLS network structure to extract features and generate CSI data distribution values ​​for the remaining sampling points; improving the accuracy of predicted fingerprints through GAN, and applying confidence coefficients to build a fine-grained fingerprint database to improve positioning accuracy.

[0045] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0047] Figure 1 This is an overall flow chart of the CSI fingerprint database optimization method based on deep and wide learning according to an embodiment of the present invention.

[0048] Figure 2 This is a block diagram of the GAN principle described in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0050] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0051] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0052] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0053] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. Different embodiments or examples described in this specification and features of different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

[0054] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0055] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0056] The existing technology adopts crowdsourcing methods to introduce a lot of noise and requires frequent updates, which limits its practical application. Using complex machine learning algorithms to extract CSI fingerprint data features requires a lot of computing resources. At the same time, these algorithms may not be able to adapt to new environments, resulting in potential inaccuracy in the final positioning. The present invention aims to design an innovative deep-broad learning system (DeepBLS) and generative adversarial network (Generative AdversarialNetworks, GAN) structure for CSI fingerprint database optimization, combining DeepBLS with GAN to improve the accuracy and efficiency of building and updating CSI fingerprint libraries in the offline stage. The present invention aims to improve the accuracy and efficiency of building and updating CSI fingerprint libraries in the offline stage by designing an innovative deep-broad learning and GAN structure for CSI fingerprint database optimization, combining deep learning, width learning and GAN. The present invention first collects raw CSI fingerprint data at sampling points to fit a multivariate Gaussian model. Subsequently, a path planning algorithm is used to collect additional raw CSI data along the optimal path. Next, the DeepBLS network structure is used to extract features and generate CSI data distribution values ​​for the remaining sampling points. Finally, the accuracy of predicted fingerprints is further improved by GAN, and the confidence coefficient is applied to construct a fine-grained fingerprint database. Finally, a large number of experiments are carried out in two real environments to verify its performance. Compared with several existing methods, the present invention shows excellent performance in updating the CSI fingerprint database, and the positioning accuracy is improved by an average of 46.78%.

[0057] Example 1

[0058] In this embodiment 1, a CSI fingerprint database optimization system based on deep and wide learning is first provided, including: an acquisition module, used to acquire CSI information; a matching module, used to use a fine-grained CSI fingerprint database constructed based on a confidence coefficient calculated based on DeepBLS and GAN, combined with a k-nearest neighbor algorithm to perform fingerprint matching and estimate the user's location; wherein, part of the original CSI data is collected, and a global CSI fingerprint database is preliminarily constructed using a DeepBLS network; wherein the DeepBLS network uses the residual of the previous layer, scaled by a learning rate, as the target output of the next layer; GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0059] In this embodiment, the above-mentioned system is used to implement a CSI fingerprint database optimization method based on deep and wide learning, including: obtaining CSI information; using a fine-grained CSI fingerprint database constructed based on a confidence coefficient calculated based on DeepBLS and GAN, and combining the k-nearest neighbor algorithm to perform fingerprint matching to estimate the user's location; wherein, part of the original CSI data is collected, and a global CSI fingerprint database is preliminarily constructed using a DeepBLS network; wherein the DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer; using GAN to further extract features from the original data, and performing secondary prediction and updating on the global CSI fingerprint database.

[0060] The DeepBLS network is constructed by combining the efficiency of width learning and the high precision of deep learning. DeepBLS is used to extract features and build an initial coarse-grained fingerprint database. Width learning captures more input data features by expanding the network width, enabling it to better adapt to training and test data, thereby enhancing its generalization ability. In the DeepBLS network structure, each layer consists of a single width learning block, and each width learning block includes a feature layer, an enhancement layer, and an output layer. The dataset is trained only once on each width learning block.

[0061] At the feature layer, for n feature maps, each map generates k nodes, thereby mapping the input available data x to n sets of feature nodes:

[0062]

[0063] Where X represents the feature matrix of the input data, and the weight W e and the bias term β e is randomly generated. is the hyperbolic tangent sigmoid nonlinear activation function; after that, all feature nodes are recorded as Z n ≡[Z1,Z2,…,Z n ].

[0064] The feature nodes are connected to the enhancement nodes, which improve the feature representation through nonlinear transformations. These transformations use randomly generated weights W. h , bias term β h and nonlinear activation function ψ(·);

[0065] The output of the enhancement node is H m ≡[H1,H2,…,H m ],in

[0066] All feature nodes and enhanced nodes are connected by weights W mConnect to the output layer node to synthesize the extracted and enhanced features to obtain the final expected output of the current width learning block

[0067] The connection weights W in each width learning layer m are the only weights that need to be calculated, and they are calculated using Derived, solve [Z n |H m ] + The process is approximately a ridge regression problem:

[0068]

[0069] In the DeepBLS network structure, the kth width learning block is trained using the initial parameter setting ω0; after the training process is completed, the output of the first width learning block is recorded as f BLS (X; ω0); After obtaining the output f of the first width learning block BLS After (X; ω0), we subtract it from the true label Y to get the first training residual r:

[0070] r (k) =Yf BLS (X;ω0);

[0071] In the formula, X represents the feature matrix of the input data, and ω0 represents the model parameters;

[0072] The residual of the k-layer DeepBLS network is:

[0073]

[0074] in represents the output of the n-layer DeepBLS network, which is the cumulative sum of the outputs of the first k-1 width learning blocks;

[0075] Finally, the model's prediction result is the sum of the prediction results of each width learning block:

[0076]

[0077] GAN consists of a generator and a discriminator;

[0078] Generator G receives random noise z, labels y, and generates fake data Where z is random noise sampled from a normal distribution, y is the label associated with the input noise, is the fake data output by the generator; the discriminator D receives the real data x real and fake data Output discriminator loss d loss , which measures the discriminator's ability to distinguish real data from fake data:

[0079]

[0080] Where D(x real ,y) and Represent real data and false data respectively;

[0081] Generator G receives random noise z and generates fake data It outputs the generator loss g loss , measuring the ability of the generator to generate fake data similar to real data:

[0082] g loss =D(G(z,y),y),

[0083] The goal of the generator is to adjust the generated data to make it closer to the original true value distribution:

[0084]

[0085] In the formula, is the mean of the generated data, is the standard deviation of the generated data, μ x is the mean of the original data, σ x is the standard deviation of the original data.

[0086] Finally, the confidence coefficient calculated by DeepBLS and GAN is used to construct a fine-grained CSI fingerprint database: highFinger = c·GAN data+(1-c)·final p redictions.

[0087] Example 2

[0088] In this embodiment 2, a CSI fingerprint database optimization based on deep and wide learning is provided, and the monitoring mode of the Linux802.11n CSI tool is used to transmit and receive data to collect a small amount of raw CSI data. This method can accurately control the transmission rate of data packets, thereby saving time in the offline phase. The receiving device is a laptop computer dedicated to data collection, while all other algorithms are executed on a local server. This configuration minimizes the resources and computing power required by the device, making it suitable for large-scale localization scenarios. The algorithm is implemented in Python using PyTorch, Sklearn, and TensorFlow libraries and runs on an HP laptop equipped with an i5-8265U CPU and 8GB RAM.

[0089] The details of the method are given below. First, some raw CSI data are collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. Specifically, the DeepBLS network adopts a training method that uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. Unlike traditional deep learning networks that require iterative training, each layer of the DeepBLS network only needs to be trained once, which significantly improves the training efficiency. Next, GAN is used to further extract features from the raw data, and the global CSI fingerprint database is predicted and updated twice. The GAN model dynamically adapts to environmental changes and automatically generates high-quality CSI data that reflects the characteristics of the current environment from limited raw CSI data. Finally, during the prediction and update process, the confidence coefficients calculated by DeepBLS and GAN are used to construct a fine-grained CSI fingerprint database. In the online stage, the k-nearest neighbor algorithm is used for fingerprint matching to estimate the user's location.

[0090] like Figure 1 The detailed description is as follows:

[0091] Step 1: Divide the positioning area into grids, first randomly collect CSI information (flight data) of a small number of sampling points in the area that conform to the normal distribution, and use these flight data to fit the multivariate Gaussian model. Then use the path planning algorithm based on A3C to build an adaptive CSI fingerprint library to collect part of the CSI information on the optimal path, and merge the two parts of CSI information to obtain usable data.

[0092] Step 2: Construct the DeepBLS network by combining the efficiency of width learning and the high accuracy of deep learning. Use DeepBLS to extract features and build an initial coarse-grained fingerprint database. Width learning captures more input data features by expanding the network width, enabling it to better adapt to training and test data, thereby enhancing its generalization ability. In the DeepBLS network structure, in order to take advantage of the fast training speed of width learning and reduce the computational complexity of deep learning, each layer consists of a single width learning block, and each width learning block includes a feature layer, an enhancement layer, and an output layer. The dataset is trained only once on each width learning block. At the feature layer, for n feature maps, each map generates k nodes, thereby mapping the input available data x to n sets of feature nodes:

[0093]

[0094] Where X represents the feature matrix of the input data, and the weight W e and the bias term β e is randomly generated. is the hyperbolic tangent sigmoid nonlinear activation function. After that, all feature nodes are recorded as Z n ≡[Z1,Z2,…,Zn ].

[0095] Step 3: Connect the feature nodes to the enhancement nodes, which improve the feature representation through nonlinear transformations. These transformations use randomly generated weights W h , bias term β h and nonlinear activation function ψ(·). The output of the enhancement node is H m ≡[H1,H2,…,H m ],in

[0096]

[0097] Step 4: Connect all feature nodes and enhanced nodes through the connection weight W m Connect to the output layer node to synthesize the features extracted and enhanced in the above steps. Therefore, the final expected output of the current width learning block is obtained

[0098]

[0099] From the above formula, we can see that the connection weight W in each width learning layer m are the only weights that need to be calculated, and they are calculated using Derived. Solve [Z n |H m ] + The process can be approximated as a ridge regression problem as follows:

[0100]

[0101] Step 5: Each layer of the deep model is designed as a wide learning block, and each layer only needs to be trained once. This architecture effectively utilizes the powerful feature extraction capabilities of deep learning while maximizing the efficiency and speed of wide learning. By continuously increasing the number of layers, the depth of the model is enhanced, thereby improving accuracy and performance.

[0102] In the DeepBLS network structure, the kth width learning block is trained using the initial parameter setting ω0 (model parameter). After the training process is completed, the output of the first width learning block is recorded as f BLS (X; ω0). After obtaining the output f of the first width learning block BLS (X; ω0), subtract it from the true label Y to get the first training residual r. In the formula, X represents the feature matrix of the input data, and ω0 represents the model parameters, as follows:

[0103] r (k) =Yf BLS (X;ω0), (5)

[0104] Step 6: To ensure gradual convergence and avoid repeated oscillations around the optimal solution, the residual of the previous width learning block is multiplied by the learning rate and used as the target output of the next width learning block. This strategy effectively guides the model to gradually approach the optimal solution. The residual of the k-layer DeepBLS network is:

[0105]

[0106] in The output of the n-layer DeepBLS network is the cumulative sum of the outputs of the first k-1 width learning blocks. Finally, the model's prediction result is the sum of the prediction results of each width learning block:

[0107]

[0108] The DeepBLS network structure extracts features from raw CSI data for prediction and update, laying the foundation for building a global CSI fingerprint database.

[0109] Step 7: GAN consists of a generator and a discriminator. The generator takes random noise as input and generates fake data, and the discriminator determines whether the input data is real or fake. Through adversarial training, the two models improve each other: the generator gets better at generating realistic fake data, and the discriminator gets better at distinguishing real data from fake data. The GAN principle diagram is as follows Figure 2 shown.

[0110] Generator G receives random noise z, labels y, and generates fake data

[0111]

[0112] Where z is random noise sampled from a normal distribution, y is the label associated with the input noise, is the fake data output by the generator. The discriminator D receives the real data x real and fake data Output discriminator loss d loss , which measures the discriminator’s ability to distinguish real data from fake data:

[0113]

[0114] Where D(x real ,y) and represent real data and false data respectively.

[0115] Generator G receives random noise z and generates fake data It outputs the generator loss, which measures the generator’s ability to generate fake data that is similar to real data:

[0116] g loss =D(G(z,y),y), (10)

[0117] The goal of the generator is to minimize this loss to generate more realistic fake data. The generated data is adjusted to be closer to the original true value distribution. is the mean of the generated data, is the standard deviation of the generated data, μ x is the mean of the original data, σ x is the standard deviation of the original data:

[0118]

[0119] Step 8: During the prediction and update process, the present invention uses the confidence coefficient calculated by DeepBLS and GAN to build a fine-grained CSI fingerprint database (the confidence coefficient indicates the reliability and accuracy of the data):

[0120] highFinger=c·GAN data+(1-c)·final p redictions, (12)

[0121] By using these coefficients, the data in the CSI fingerprint database can be effectively filtered and optimized to ensure a high-precision fingerprint database. A high-precision fingerprint database can more accurately reflect the state and changes of the wireless channel, thereby improving positioning accuracy. Finally, the K nearest neighbor algorithm is used for fingerprint matching to estimate the user's location.

[0122] Example 3

[0123] This embodiment 3 provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the CSI fingerprint database optimization method based on deep and wide learning as described above is implemented. The method includes:

[0124] Get CSI information;

[0125] A fine-grained CSI fingerprint database constructed based on the confidence coefficient calculated by DeepBLS and GAN is used to perform fingerprint matching in combination with the k-nearest neighbor algorithm to estimate the user's location. Part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. The DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0126] Example 4

[0127] This embodiment 4 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, the processor calls the program instructions to execute the CSI fingerprint database optimization method based on deep and wide learning as described above, the method comprising:

[0128] Get CSI information;

[0129] A fine-grained CSI fingerprint database constructed based on the confidence coefficient calculated by DeepBLS and GAN is used to perform fingerprint matching in combination with the k-nearest neighbor algorithm to estimate the user's location. Part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. The DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0130] Example 5

[0131] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the CSI fingerprint database optimization method based on deep and wide learning as described above, and the method includes:

[0132] Get CSI information;

[0133] A fine-grained CSI fingerprint database constructed based on the confidence coefficient calculated by DeepBLS and GAN is used to perform fingerprint matching in combination with the k-nearest neighbor algorithm to estimate the user's location. Part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. The DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

[0134] In summary, the CSI fingerprint database optimization method based on deep and wide learning described in the embodiment of the present invention combines the efficiency of wide learning and the high precision of deep learning to predict and update CSI data. A path planning strategy is adopted to collect a small amount of original CSI fingerprint data, and DeepBLS is used to extract features to construct an initial coarse-grained fingerprint database. GAN is used to further extract the features of the original CSI data to generate high-quality CSI data. It can effectively predict and update global CSI data, enhance the robustness and adaptability of the system, and significantly reduce the costs associated with data collection and maintenance. The confidence coefficient is used to balance the prediction results of the two structures to construct the final fine-grained fingerprint database. A large number of experiments and comparative analysis with various existing algorithms show that the CSI data updated by the invention is very close to the distribution of the original real data, and has superior overall positioning performance, with the overall positioning performance improved by an average of 46.78%.

[0135] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative work on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A CSI fingerprint database optimization method based on deep and wide learning, characterized in that: include: Get CSI information; A fine-grained CSI fingerprint database constructed based on the confidence coefficient calculated by DeepBLS and GAN is used to perform fingerprint matching in combination with the k-nearest neighbor algorithm to estimate the user's location. Part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network. The DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer. GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

2. The CSI fingerprint database optimization method based on deep and wide learning according to claim 1 is characterized in that: The DeepBLS network is constructed by combining the efficiency of width learning and the high accuracy of deep learning; DeepBLS is used to extract features and build an initial coarse-grained fingerprint database; width learning captures more input data features by expanding the network width, enabling it to better adapt to training and test data, thereby enhancing its generalization ability; In the DeepBLS network structure, each layer consists of a single width learning block, and each width learning block includes a feature layer, an enhancement layer, and an output layer; The dataset is trained only once on each width learning block.

3. The CSI fingerprint database optimization method based on deep and wide learning according to claim 2 is characterized in that: At the feature layer, for n feature maps, each map generates k nodes, thereby mapping the input available data x to n sets of feature nodes: Where X represents the feature matrix of the input data, and the weight W e and the bias term β e is randomly generated. is the hyperbolic tangent sigmoid nonlinear activation function; after that, all feature nodes are recorded as Z n ≡[Z1, Z2, …, Z n ].

4. The CSI fingerprint database optimization method based on deep and wide learning according to claim 1 is characterized in that: The feature nodes are connected to the enhancement nodes, which improve the feature representation through nonlinear transformations. These transformations use randomly generated weights W. h , bias term β h and nonlinear activation function ψ(·); The output of the enhancement node is H m ≡[H1, H2, …, H m ],in All feature nodes and enhanced nodes are connected by weights W m Connect to the output layer node to synthesize the extracted and enhanced features to obtain the final expected output of the current width learning block The connection weights W in each width learning layer m are the only weights that need to be calculated, and they are calculated using Derived, solve [Z n |H m ] + The process is approximately a ridge regression problem:

5. The CSI fingerprint database optimization method based on deep and wide learning according to claim 4 is characterized in that: In the DeepBLS network structure, the kth width learning block is trained using the initial parameter setting ω0; After the training process is completed, the output of the first width learning block is recorded as f BLS (X; ω0); After obtaining the output f of the first width learning block BLS After (X; ω0), we subtract it from the true label Y to get the first training residual r: r (k) =Y-f BLS (X;ω0); In the formula, X represents the feature matrix of the input data, and ω0 represents the model parameters; The residual of the k-layer DeepBLS network is: in represents the output of the n-layer DeepBLS network, which is the cumulative sum of the outputs of the first k-1 width learning blocks; Finally, the model's prediction result is the sum of the prediction results of each width learning block:

6. The CSI fingerprint database optimization method based on deep and wide learning according to claim 1 is characterized in that: GAN consists of a generator and a discriminator; Generator G receives random noise z, labels y, and generates fake data Where z is random noise sampled from a normal distribution, y is the label associated with the input noise, is the fake data output by the generator; The discriminator D receives the real data x real and fake data Output discriminator loss d loss , which measures the discriminator's ability to distinguish real data from fake data: Where D(x real , y) and Represent real data and false data respectively; Generator G receives random noise z and generates fake data It outputs the generator loss g loss , measuring the ability of the generator to generate fake data similar to real data: g loss =D(G(z,y),y), The goal of the generator is to adjust the generated data to make it closer to the original true value distribution: In the formula, is the mean of the generated data, is the standard deviation of the generated data, μ x is the mean of the original data, σ x is the standard deviation of the original data.

7. The CSI fingerprint database optimization method based on deep and wide learning according to claim 6 is characterized in that: Use DeepBLS and GAN to calculate the confidence coefficient to build a fine-grained CSI fingerprint database: highFinger=c·GAN data+(1-c)·final p redictions。 8. A CSI fingerprint database optimization system based on deep and wide learning, characterized in that: include: An acquisition module, used to obtain CSI information; The matching module is used to use the fine-grained CSI fingerprint database built based on the confidence coefficient calculated by DeepBLS and GAN, combined with the k-nearest neighbor algorithm to perform fingerprint matching and estimate the user's location; wherein, part of the original CSI data is collected, and the global CSI fingerprint database is preliminarily constructed using the DeepBLS network; wherein the DeepBLS network uses the residual of the previous layer, scaled by the learning rate, as the target output of the next layer; GAN is used to further extract features from the original data, and the global CSI fingerprint database is secondary predicted and updated.

9. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the CSI fingerprint database optimization method based on deep and wide learning as described in any one of claims 1-7.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the CSI fingerprint database optimization method based on deep and wide learning as described in any one of claims 1-7.

Images