Indoor fingerprint positioning method and device for automatically designing network architecture

Abstract

The embodiment of the invention provides an indoor fingerprint positioning method and device for automatically designing a network architecture, and relates to the technical field of data processing, and the method comprises the steps: obtaining channel state information (CSI) data of a Wi-Fi signal in different indoor environments, constructing a fingerprint feature containing amplitude and phase information, inputting the fingerprint feature into a search space, and obtaining a search result; calculating a weight coefficient of each operation function in the spatial position feature extraction network; calculating a variable parameter of a spatial position feature extraction network optimization strategy, and updating the weight coefficient; constructing a spatial position feature extraction network architecture based on the operation function with the maximum weight coefficient, extracting CSI fingerprint spatial position features, and fusing the fingerprint features; calculating the matching probability between the known position feature and the position information, and determining the position of the fingerprint feature; and inputting fingerprint features at unknown coordinates into the network to obtain a positioning result. By applying the scheme provided by the embodiment of the invention, the precision and the universality of the fingerprint positioning system in different indoor scenes can be improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of data processing, and particularly to an indoor fingerprint positioning method and device for automatically designing a network architecture. Technical Background

[0002] With the rapid development of mobile devices and wireless technologies, the research on location-based services (LBSs) has gradually become a hot topic. In practical applications, outdoor positioning technologies (such as GPS) have shown excellent performance. Different from traditional outdoor positioning, the complexity of the indoor environment, such as building structures, causes problems such as multipath propagation, non-line-of-sight transmission, and signal attenuation for signals. The movement of people and devices, as well as constantly changing signal interference, make dynamic high-precision indoor positioning more challenging. Therefore, it is crucial to achieve high-precision positioning services in indoor environments. WiFi technology and fingerprint-based positioning technology, as cutting-edge choices for indoor positioning, make full use of the widely deployed wireless networks and environmental feature information, and can provide low-cost and low-complexity high-precision indoor positioning solutions.

[0003] With the remarkable performance of deep neural networks in scientific research and engineering tasks, deep learning algorithms are very popular among the public. It replaces the traditional method of manually extracting features and can automatically extract and learn features end-to-end. The reason for its remarkable success is usually due to their successful architecture design, creating an effective network structure according to specific task requirements. Now the focus of research work has gradually shifted from designing feature extraction modules to designing the overall optimal architecture of network models. However, this process usually relies on the intuition and experience of experts in this field, so there is a problem of limited universality. The manually designed network structure may be too specific to a particular task and difficult to generalize to different fields or problems. This limits the generality of the network and requires a large amount of manual intervention and adjustment.

[0004] Traditional deep learning-based indoor positioning methods carefully design the network structure of the model manually and use the model for estimating the sample positions in different environments. However, the irregular changes in space in different environments make the propagation characteristics, location information, and signal strength distribution of signals different, resulting in some obvious inadaptabilities of these methods when applied to different spatial scenarios. An indoor positioning system needs to be able to adapt to changes such as the movement of people, the switching of devices, and the change of building structures to maintain accuracy. Therefore, the related technologies face the following problems: The traditional method of manually designing network models has achieved relatively high positioning accuracy in fingerprint positioning, but its positioning accuracy varies with different indoor space structures, the adaptability of the model in different environments is low, and it is difficult to achieve high robustness of the model in different indoor space scenarios. Summary of the Invention

[0005] The objective of the embodiments of the present invention is to provide an indoor fingerprint positioning method and device for automatically designing a network architecture, so as to increase the accuracy and universality of an indoor wireless fingerprint positioning system. The specific technical solutions are as follows:

[0006] In a first aspect, the embodiments of the present invention provide an indoor fingerprint positioning method for automatically designing a network architecture, and the method includes:

[0007] Obtain channel state information (CSI) data of Wi-Fi signals in different indoor environments, construct fingerprint features including amplitude and phase information, input the CSI fingerprint features into a search space, and calculate the weight coefficients of each operation function in the spatial position feature extraction network;

[0008] Calculate the variable parameters of the optimization strategy of the spatial position feature extraction network, and update the weight coefficients of the operation functions based on this parameter;

[0009] Construct a spatial position feature extraction network architecture based on the operation function with the largest updated weight coefficient, obtain the spatial position features of the CSI fingerprint, and fuse the features with the above CSI fingerprint features to update the fingerprint features representing the position information of the reference points;

[0010] Use the known coordinate information of the reference points to train a model to predict the matching probability between the input features and the position information, and determine the position to which the fingerprint features belong based on the matching probability.

[0011] In an embodiment of the present invention, the obtaining of the channel state information (CSI) data of Wi-Fi signals in different indoor environments, constructing the fingerprint features including amplitude and phase information, inputting the CSI fingerprint features into the search space, and calculating the weight coefficients of each operation function in the spatial position feature extraction network includes:

[0012] Continuously obtain multiple CSI data packets through Wi-Fi devices, retain the amplitude and phase information of all subcarriers and multiple links from each CSI data packet as one channel of the CSI feature map, splice the CSI features of the continuous time series to construct fingerprint features, then build the search space of the spatial position feature extraction network, input the CSI fingerprint features into the search space, then relax the search space, and relax the selection of each node operation function of the network to the softmax value of all possible operations to obtain the weight coefficients of each operation function in the spatial position feature extraction network

[0013] In an embodiment of the present invention, the calculating of the variable parameters of the optimization strategy of the spatial position feature extraction network and updating the weight coefficients of the operation functions based on this parameter includes:

[0014] Calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and calculate the minimum validation loss through gradient descent where w * is obtained by minimizing the training loss α * is obtained by minimizing and is used to update the weight coefficients of each operation function in the feature extraction network.

[0015] In an embodiment of the present invention, a spatial location feature extraction network architecture is constructed based on the operation function with the largest weight coefficient obtained by updating, the spatial location features of the CSI fingerprint are obtained, and these features are fused with the above CSI fingerprint features to update the fingerprint features representing the reference point position information, including:

[0016] Select the edge o with the largest weight coefficient of the updated operation function (i,j) , construct the model architecture of the spatial location feature extraction network, and obtain the spatial location features of the CSI fingerprint through the selected operation functions.

[0017] The obtained spatial location features are used to be fused with the above fingerprint features, and all the information of the above two features is retained here and fused in a splicing manner. The fused fingerprint features are updated to replace the original fingerprint features to represent the reference point position information.

[0018] In an embodiment of the present invention, using the known coordinate information of the reference point to train the model to predict the matching probability between the input feature and the position information, and determining the position to which the fingerprint feature belongs based on the matching probability, including:

[0019] Input the CSI data obtained from the reference point and the reference point coordinates into the searched indoor fingerprint positioning model framework. After pre-training the model, input the CSI data obtained at the unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint feature to be located output by the fingerprint positioning model.

[0020] In a second aspect, an embodiment of the present invention provides an indoor fingerprint positioning device for automatically designing a network architecture, and the device includes:

[0021] A signal processing and architecture weight calculation module: Obtain the channel state information CSI data of Wi-Fi signals in different indoor environments, construct fingerprint features including amplitude and phase information, input the search space based on the CSI fingerprint features, and calculate the weight coefficients of each operation function in the spatial location feature extraction network;

[0022] An optimization parameter calculation module: Calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and update the weight coefficients of the operation functions based on this parameter;

[0023] Spatial position feature calculation module: Based on the operation function with the largest updated weight coefficient, construct the architecture of the spatial position feature extraction network, obtain the spatial position features of the CSI fingerprint, and fuse these features with the above-mentioned CSI fingerprint features to update the fingerprint features representing the position information of the reference point;

[0024] Position estimation module: Use the known coordinate information of the reference point to train the model to predict the matching probability between the input features and the position information, and determine the position to which the fingerprint features belong based on the matching probability.

[0025] In one embodiment of the present invention, the signal processing and architecture weight calculation module is specifically used for:

[0026] Continuously obtain multiple CSI data packets through a Wi-Fi device, retain the amplitude and phase information of all subcarriers and multiple links in each CSI data packet as a channel of the CSI feature map, splice the CSI features of the continuous time series to construct fingerprint features, then build the search space of the spatial position feature extraction network, input the CSI fingerprint features into the search space, then relax the search space, and relax the selection of the operation function of each node in the network to the softmax value of all possible operations to obtain the weight coefficients of the operation functions in the spatial position feature extraction network

[0027] In one embodiment of the present invention, the optimization parameter calculation module is specifically used for:

[0028] Calculate the variable parameters of the optimization strategy of the spatial position feature extraction network, and calculate the minimum verification loss through gradient descent where w * By minimizing the training loss obtain α * By minimizing obtain, which is used to update the weight coefficients of the operation functions in the feature extraction network.

[0029] In one embodiment of the present invention, the spatial position feature calculation module is specifically used for:

[0030] Select the edge o with the largest updated operation function weight coefficient (i,j) , construct the model architecture of the spatial position feature extraction network, and obtain the spatial position features of the CSI fingerprint through the selected operation functions.

[0031] The obtained spatial position features are used to fuse with the above-mentioned fingerprint features. Here, all the information of the above two features is retained and fused in a splicing manner. The fused fingerprint features are updated to replace the original fingerprint features to represent the position information of the reference point.

[0032] In one embodiment of the present invention, the position estimation module is specifically configured to:

[0033] Input the CSI data obtained at the reference point and the reference point coordinates into the searched indoor fingerprint positioning model framework together. After pre-training the model, input the CSI data obtained at the unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint feature to be located output by the fingerprint positioning model.

[0034] Beneficial effects of the embodiments of the present invention:

[0035] The present invention provides an indoor fingerprint positioning method and device for automatically designing a network architecture, which automatically designs a network model for different indoor space environmental structures, extracts CSI spatial position features in different scenarios, and fuses them with the original CSI fingerprint features. The fused features are used as fingerprints for indoor position estimation, contributing an important technical tool to the development of indoor position services. Compared with the related technologies, through this indoor fingerprint positioning method and device, the present invention can significantly improve the accuracy and universality of fingerprint positioning in indoor scenarios with different spatial structures, and increase the feasibility of deploying the fingerprint positioning system in actual indoor scenarios. Description of the Drawings

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other embodiments based on these drawings.

[0037] Figure 1 It is a schematic flowchart of an indoor fingerprint positioning method for automatically designing a network architecture provided by an embodiment of the present invention;

[0038] Figure 2 It is a schematic structural diagram of the original fingerprint features provided by an embodiment of the present invention;

[0039] Figure 3 It is a schematic structural diagram of the search space position feature network architecture provided by an embodiment of the present invention;

[0040] Figure 4 It is a schematic diagram of the optimization strategy of the space position feature network architecture provided by an embodiment of the present invention;

[0041] Figure 5 It is a schematic structural diagram of the space position feature network provided by an embodiment of the present invention;

[0042] Figure 6 It is a schematic structural diagram of an indoor fingerprint positioning device for automatically designing a network architecture provided by an embodiment of the present invention.

[0043] Specific implementation method

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art based on the present invention belong to the scope of protection of the present invention.

[0045] An indoor positioning system needs to be able to adapt to changes such as personnel movement, device switching, and building structure changes to maintain accuracy. The irregular spatial changes in different indoor environments result in different signal propagation characteristics, position information, and signal strength distributions, causing obvious inadaptabilities when traditional deep learning-based indoor positioning methods are applied to different spatial scenarios. Therefore, the related technologies face the following problems: The method of traditional manually designed network models has achieved relatively high positioning accuracy in fingerprint positioning, but its positioning accuracy varies with different indoor space structures, and the adaptability of the model in different environments is low, making it difficult to achieve high robustness of the model in different indoor space scenarios.

[0046] To solve the above problems, the embodiments of the present invention provide an indoor fingerprint positioning method and device for automatically designing a network architecture, which will be specifically described below.

[0047] First, an indoor fingerprint positioning method for automatically designing a network architecture provided by the embodiments of the present invention will be described.

[0048] See Figure 1 , which is a schematic flowchart of an indoor fingerprint positioning method provided by the embodiments of the present invention. This method is applied to an electronic device with computing capabilities. Exemplarily, this method is applied to a computer. The above method includes the following steps S101 to S104.

[0049] Step S101: Obtain the channel state information CSI data of Wi-Fi signals in different indoor environments, construct fingerprint features including amplitude and phase information, input the CSI fingerprint features into the search space, and calculate the weight coefficients of each operation function in the spatial position feature extraction network.

[0050] Specifically, multiple CSI data packets are continuously obtained through a Wi-Fi device. The amplitude and phase information of all subcarriers and multiple links in each CSI data packet are retained as one channel of the CSI feature map. The CSI features of the continuous time series are concatenated to construct fingerprint features. Then, the search space of the spatial location feature extraction network is built, and the CSI fingerprint features are input into the search space. The search space is relaxed, and the selection of the operation function of each node in the network is relaxed to the softmax value of all possible operations, obtaining the weight coefficients of the operation functions in the spatial location feature extraction network

[0051] Exemplarily, for M continuously collected CSI data packets, each antenna link contains N subcarriers, and the CSI value of each subcarrier is a complex number:

[0052] csim,n=csim,nexp(j∠csim,n)(1≤m≤M,1≤n≤N)

[0053] where csi m,n and j∠csi m,n are the amplitude and phase values corresponding to the nth subcarrier in the mth data packet, respectively. For the original phase that cannot be directly utilized, we use a linear transformation method to process it to obtain the calibrated phase

[0054] Therefore, the CSI amplitude data and phase matrix received at one time can be represented as matrix AMP and matrix PHA:

[0055]

[0056] T antenna links can obtain T amplitude matrices and T phase matrices. After min-max normalization, each amplitude matrix AMP and phase matrix PHA are concatenated together as one channel of the CSI feature map. To retain the multi-link and multi-carrier information of the wireless signal propagating in space, as Figure 2 shown, in the embodiments of the present invention, the above-mentioned CSI amplitude matrix AMP and phase matrix PHA are obtained from multiple continuously collected data packets, and the single-layer features of T data packets are concatenated together to obtain the CSI feature map as the original fingerprint feature CSI in . Then, the search space of the spatial location feature extraction network is built, and then the CSI fingerprint features are input into the search space. The search space is relaxed, and then the selection of the operation function of each node in the network is relaxed to the softmax value of all possible operations, obtaining the weight coefficients of the operation functions in the spatial location feature extraction network

[0057] The expressive power of the network model depends on the properties of different aggregation functions. Therefore, a search space with strong expressive power and simplicity is designed:

[0058] · Node aggregator: We selected 6 node aggregators according to popular CNN models, as shown in Table 1. Denote the node aggregator with O n . The function design of each node aggregator is shown in Table 1.

[0059] · Connection operation: We use O s for the connection operation. zero represents a special zero operation, indicating no connection between two nodes.

[0060] Table 1 Search space

[0061]

[0062] Input the above original fingerprint features into the search space. Each node x (i) in the search space is a latent representation (such as the feature map in a convolutional network). Each directed edge (i, j) is associated with an operation o (i) for a certain transformation of x (i,j) . Each intermediate node is calculated based on all its previous nodes:

[0063] x (j) = ∑ i<j o (i,j) (x (i) )

[0064] The task of the search space location feature extraction network framework is simplified to learning the operations on its edges. The possible operations between two nodes are selected in the search space. Let O be a set of candidate operations in O n , where each operation represents a certain function O(·) to be applied to x (i) . To make the search space continuous, the selection of a specific operation is relaxed to the softmax value of all possible operations. The calculation formula of the weight coefficient calculation module is as follows:

[0065]

[0066] Among them, a pair of nodes (i, j) is parameterized by a vector α n with a dimension of |O (i,j) |. The task of architecture search is simplified to learning a set of continuous variables α = {α (i,j)}, as shown in Figure 3 , to obtain the weight coefficient α (i,j) of each operation function in the spatial location feature extraction network.

[0067] Step S102: Calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and update the weight coefficients of the operation function based on these parameters.

[0068] Specifically, calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and calculate the minimum verification loss through gradient descent where, w * By minimizing the training loss obtain, α * By minimizing obtain, the coefficient α obtained by minimizing the variable parameters is used to update the weight coefficients of each operation function in the feature extraction network.

[0069] Exemplarily, after relaxation, it is necessary to jointly learn the structure α and network weights w in all hybrid operations, such as Figure 3 shown. The optimization strategy during the search process aims to optimize the verification loss through the method of gradient descent. and represent the training loss and verification loss respectively. These two losses not only depend on the structure α, but also on the weights w in the network. The goal of architecture search is to find the structure that minimizes the verification loss where, the architecture-related weight w * By minimizing the training loss obtain, the weight coefficient α * By minimizing the training loss obtain.

[0070] Regarding the network architecture problem of the search space location feature extraction network as a two-level optimization problem, α is the upper-level variable, w is the lower-level variable, and the calculation formula of the structure α is as follows:

[0071] min α

[0072] s.t.

[0073] Finally, the structure coefficient α and network weight w in the variable parameters of the optimization strategy for the spatial location feature extraction network will be obtained * , such as Figure 4 shown, and update the weight coefficients of each operation function in the feature extraction network with the structure coefficient α.

[0074] Step S103: Based on the operation function with the largest updated weight coefficient, construct the spatial location feature extraction network architecture, obtain the spatial location features of the CSI fingerprint, and fuse this feature with the above CSI fingerprint features to update the fingerprint features representing the reference point location information.

[0075] Specifically, select the edge o with the largest weight coefficient of the updated operation function (i,j) , construct the model architecture of the spatial location feature extraction network, and obtain the spatial location features of the CSI fingerprint through each selected operation function.

[0076] Exemplarily, at the end of the search, by replacing each hybrid operation with the most likely operation, its calculation formula is:[[]]END]]

[0077]

[0078] After training is completed, find the edge with the highest probability from all edges, as Figure 5 shown, to form the architecture of the spatial location feature extraction network. The spatial location features of the CSI fingerprint are obtained through the finally selected convolution operation, pooling operation, etc. in the network architecture. The above-obtained CSI spatial location features and the original fingerprint features are linearly concatenated, updated to replace the original fingerprint features, and used as the fingerprint features representing the position information of the reference point.

[0079] Step S104: Use the known coordinate information of the reference point to train the model to predict the matching probability between the input feature and the position information, and determine the position to which the fingerprint feature belongs based on the matching probability.

[0080] Specifically, input the CSI data obtained from the reference point and the reference point coordinates into the indoor fingerprint positioning model framework after the search. After pre-training the model, input the CSI data obtained at the unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint feature to be located output by the fingerprint positioning model.

[0081] Exemplarily, a simple softmax classifier is used in the position estimation module to classify the input fused features. Set the size of the fused features to be equal to the number of RPs (i.e., d h = num_RP s ). In the output of the softmax classifier, the probability calculation formula for h belonging to the i-th RP is:[[]]END]]

[0082]

[0083] where h i represents the i-th eigenvalue in the fused feature vector.

[0084] To train the model end-to-end in the offline stage, cross-entropy is used as the overall loss function, and its calculation formula is:[[]]END]]

[0085]

[0086] where,[[]]END]] Let $\hat{y}$ be the predicted probability distribution and $y$ be the actual probability distribution. The Adam optimization algorithm is used to iteratively update the network parameters to minimize the loss function and complete the training of the model.

[0087] In the online positioning stage, the fingerprint features of the above CSI reference point location information are input into the trained model. The model will output the probability values of the positioning device at each RP. Select the 3 RPs with the highest probabilities, and obtain the final position estimate through the weighted centroid method. The calculation formula is:

[0088]

[0089] where $L$ is the final predicted coordinate, $p_1$, $p_2$, and $p_3$ are the three maximum probability values output by the model, and $L_1$, $L_2$, and $L_3$ are the coordinates of the RPs corresponding to these three probability values respectively.

[0090] Corresponding to the aforementioned indoor fingerprint positioning method, an embodiment of the present invention also provides an indoor fingerprint positioning device with an automatically designed network architecture.

[0091] See Figure 6 , which is a schematic structural diagram of an indoor fingerprint positioning device with an automatically designed network architecture provided by an embodiment of the present invention. The device is applied to an electronic device with computing capabilities. The above device includes:

[0092] Signal processing and architecture weight calculation module 501: used to obtain the channel state information CSI data of Wi-Fi signals in different indoor environments, construct fingerprint features including amplitude and phase information, input the CSI fingerprint features into the search space, and calculate the weight coefficients of each operation function in the spatial position feature extraction network;

[0093] Optimization parameter calculation module 502: calculate the variable parameters of the optimization strategy of the spatial position feature extraction network, and update the weight coefficients of the operation functions based on this parameter;

[0094] Spatial position feature calculation module 503: construct a spatial position feature extraction network architecture based on the operation function with the largest updated weight coefficient, obtain the spatial position features of the CSI fingerprint, and fuse the features with the above CSI fingerprint features to update the fingerprint features representing the reference point position information;

[0095] Position prediction module 504: use the known coordinate information of the reference point to train the model to predict the matching probability between the input features and the position information, and determine the position to which the fingerprint features belong based on the matching probability.

[0096] In an embodiment of the present invention, the above signal processing and architecture weight calculation module 501 is specifically used for:

[0097] Continuously obtain multiple CSI data packets through a Wi-Fi device, retain the amplitude and phase information of all subcarriers and multiple links in each CSI data packet as a channel of the CSI feature map, splice the CSI features of the continuous time series to construct fingerprint features, then build the search space of the spatial location feature extraction network, input the CSI fingerprint features into the search space, then relax the search space, and then relax the selection of the operation function of each node in the network to the softmax value of all possible operations to obtain the weight coefficients of the operation functions in the spatial location feature extraction network

[0098] In one embodiment of the present invention, the above optimization parameter calculation module 502 is specifically used for:

[0099] Calculate the variable parameters of the optimization strategy of the spatial location feature extraction network, and calculate the minimum verification loss through gradient descent where w * is obtained by minimizing the training loss, and the coefficient α obtained by minimizing the variable parameters is used to update the weight coefficients of the operation functions in the feature extraction network.

[0100] In one embodiment of the present invention, the above spatial location feature calculation module 503 is specifically used for:

[0101] Select the edge o with the largest weight coefficient of the updated operation function (i,j) to construct the model architecture of the spatial location feature extraction network, and obtain the spatial location features of the CSI fingerprint through the selected operation functions.

[0102] The obtained spatial location features are used to fuse with the above fingerprint features. All information of the above two features is retained here and fused in a splicing manner. The fused fingerprint features are updated to replace the original fingerprint features to represent the reference point location information.

[0103] In one embodiment of the present invention, the above position prediction module 504 is specifically used for:

[0104] Input the CSI data obtained at the reference point and the reference point coordinates into the indoor fingerprint positioning model framework after the search. After pre-training the model, input the CSI data obtained at the unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint feature to be located output by the fingerprint positioning model.

Claims

1. An indoor fingerprint positioning method for automatically designing a network architecture, characterized in that, The method includes: Obtaining channel state information (CSI) data of Wi-Fi signals in different indoor environments, constructing fingerprint features including amplitude and phase information, inputting the CSI fingerprint features into a search space, and calculating the weight coefficients of each operation function in the spatial location feature extraction network; Calculating the variable parameters of the optimization strategy of the spatial location feature extraction network, and updating the weight coefficients of the operation functions based on these parameters; For the operation function with the largest obtained weight coefficient, constructing the architecture of the spatial location feature extraction network based on the operation function, obtaining the spatial location features of the CSI fingerprint, and fusing the features with the above CSI fingerprint features to update the fingerprint features representing the position information of the reference point; Using the known coordinate information of the reference point to train a model to predict the matching probability between the input features and the position information, and determining the position to which the fingerprint features belong based on the matching probability.

2. The method according to claim 1, characterized in that The obtaining of the CSI data of Wi-Fi signals in different indoor environments, constructing fingerprint features including amplitude and phase information, inputting the CSI fingerprint features into a search space, and calculating the weight coefficients of each operation function in the spatial location feature extraction network includes: Continuously obtain multiple CSI data packets through a Wi-Fi device, retain the amplitude and phase information of all subcarriers and multiple links in each CSI data packet as a channel of the CSI feature map, splice the CSI features of the continuous time series to construct fingerprint features, then build the search space of the spatial location feature extraction network, input the CSI fingerprint features into the search space, relax the search space, and then relax the selection of the operation function of each node in the network to the softmax value of all possible operations to obtain the weight coefficients of the operation functions in the spatial location feature extraction network 3. The method according to claim 1, characterized in that, The calculating of the variable parameters of the optimization strategy of the spatial location feature extraction network, and updating the weight coefficients of the operation functions based on these parameters includes: Calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and calculate the minimum validation loss through gradient descent where, w * By minimizing the training loss obtained, α * By minimizing the training loss obtained, and used to update the weight coefficients of each operation function in the feature extraction network.

4. The method according to claim 1, characterized in that, Based on the operation function with the largest updated weight coefficient, constructing the architecture of the spatial location feature extraction network, obtaining the spatial location features of the CSI fingerprint, and fusing the features with the above CSI fingerprint features to update the fingerprint features representing the position information of the reference point includes: Select the edge with the largest weight coefficient of the updated operation function o (i,j) , construct the model architecture of the spatial location feature extraction network, and obtain the spatial location features of the CSI fingerprint through each selected operation function. The obtained spatial location features are used for fusing with the above fingerprint features, and all the information of the above two features is retained here, and they are fused in a splicing manner. The fused fingerprint features are updated to replace the original fingerprint features to represent the position information of the reference point.

5. The method according to claim 1, wherein The using of the known coordinate information of the reference point to train a model to predict the matching probability between the input features and the position information, and determining the position to which the fingerprint features belong based on the matching probability includes: Jointly inputting the CSI data obtained at the reference point and the reference point coordinates into the searched indoor fingerprint positioning model framework. After pre-training the model, inputting the CSI data obtained at the unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint features to be located output by the fingerprint positioning model.

6. An indoor fingerprint positioning device for automatically designing a network architecture, characterized in that, The device includes: A signal processing and architecture weight calculation module: used for obtaining the CSI data of Wi-Fi signals in different indoor environments, constructing fingerprint features including amplitude and phase information, inputting the CSI fingerprint features into a search space, and calculating the weight coefficients of each operation function in the spatial location feature extraction network; An optimization parameter calculation module: calculating the variable parameters of the optimization strategy of the spatial location feature extraction network, and updating the weight coefficients of the operation functions based on these parameters; A spatial location feature calculation module: based on the operation function with the largest updated weight coefficient, constructing the architecture of the spatial location feature extraction network, obtaining the spatial location features of the CSI fingerprint, and fusing the features with the above CSI fingerprint features to update the fingerprint features representing the position information of the reference point; Position Estimation Module: Train a model using the known coordinate information of reference points to predict the matching probability between input features and position information, and determine the position to which the fingerprint features belong based on the matching probability.

7. The device according to claim 6, characterized in that, The signal processing and architecture weight calculation module is specifically used for: Continuously obtain multiple CSI data packets through a Wi-Fi device, retain the amplitude and phase information of all subcarriers and multiple links in each CSI data packet as a channel of the CSI feature map, splice the CSI features of the continuous time series to construct fingerprint features, then build the search space of the spatial location feature extraction network, input the CSI fingerprint features into the search space, then relax the search space, and then relax the selection of the operation function of each node in the network to the softmax value of all possible operations to obtain the weight coefficients of the operation functions in the spatial location feature extraction network 8. The device according to claim 6, characterized in that The optimization parameter calculation module is specifically used for: Calculate the variable parameters of the optimization strategy for the spatial location feature extraction network, and calculate the minimum validation loss through gradient descent where, w * By minimizing the training loss obtained, α * By minimizing the training loss obtained, which is used to update the weight coefficients of each operation function in the feature extraction network.

9. The device according to claim 6, characterized in that The spatial position feature calculation module is specifically used for: Select the edge with the largest weight coefficient of the updated operation function o (i,j) , construct the model architecture of the spatial position feature extraction network, and obtain the spatial position features of the CSI fingerprint through each selected operation function. The obtained spatial position features are used to fuse with the above fingerprint features, retaining all the information of the above two features and fusing them in a splicing manner. The fingerprint features obtained after fusion are updated to replace the original fingerprint features to represent the position information of the reference points.

10. The device according to claim 6, characterized in that, The position estimation module is specifically used for: Input the CSI data obtained from the reference points and the reference point coordinates into the searched indoor fingerprint positioning model framework. After pre-training the model, input the CSI data obtained at unknown coordinates in the same indoor scene into the model to obtain the position information of the fingerprint features to be located output by the fingerprint positioning model.

Images