Training method of neural localization model, method and device for localizing target object

Abstract

The present application provides a training method for a graph neural localization model, a method and device for localizing a target object, the method comprising obtaining a first training set and a second training set, wherein the first training set comprises a plurality of first channel state undirected graphs, and the second training set comprises a plurality of second channel state undirected graphs and a location label of the target object corresponding to each second channel state undirected graph; pre-training an initial graph neural network using the plurality of first channel state undirected graphs to obtain an intermediate graph neural network; training an initial localization model using the plurality of second channel state undirected graphs and the plurality of location labels to obtain a graph neural localization model, wherein the initial localization model comprises a plurality of mutually independent intermediate graph neural networks.

Description

Technical Field

[0001] The present application relates to the field of neural network technology, and more specifically, to a training method for a graph neural localization model, a method for localizing a target object, a training device for a graph neural localization model, a device for localizing a target object, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Indoor Positioning System (IPS) aims to provide accurate location of people or objects in indoor environments where Global Positioning System (GPS) and other satellite positioning technologies lack accuracy or fail completely. IPS is an important basic task and has great value in business, military, retail, inventory tracking and other fields. However, existing indoor positioning technologies still have problems. Vision-based indoor positioning technology is easily affected by lighting conditions and has serious privacy issues; radar-based indoor positioning technology is expensive to deploy. In contrast, widely deployed commercial WiFi devices are more cost-effective and provide important ideas for indoor positioning systems. The Channel State Information (CSI) generated by commercial WiFi devices can provide detailed information about the signal propagation path, including multipath effects, scattering and fading, which makes CSI have great potential in fine-grained indoor positioning. Summary of the invention

[0003] In view of this, the present application provides a training method for a graph neural localization model, a method for locating a target object, a training device for a graph neural localization model, a device for locating a target object, an electronic device, a computer-readable storage medium, and a computer program product.

[0004] One aspect of the present application provides a method for training a graph neural localization model, comprising:

[0005] Acquire a first training set and a second training set, wherein the first training set includes a plurality of first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a position label of a target object corresponding to each of the second channel state undirected graphs;

[0006] Pre-training the initial graph neural network using the plurality of the first channel state undirected graphs to obtain an intermediate graph neural network;

[0007] The initial positioning model is trained using the multiple second channel state undirected graphs and the multiple position labels to obtain the graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

[0008] Another aspect of the present application provides a method for locating a target object, comprising:

[0009] When the transmitting end of the target object interacts with a plurality of wireless devices, obtaining a target channel state matrix generated by the plurality of wireless devices;

[0010] For each of the target channel state matrices, generating a target channel state undirected graph according to the target channel state matrix;

[0011] Input the above target channel state undirected graph into the graph neural localization model to obtain multiple initial position prediction sets;

[0012] A weighted average process is performed on the multiple initial position prediction sets to obtain the target position of the target object.

[0013] According to an embodiment of the present application, the above-mentioned initial position prediction set includes a horizontal coordinate prediction mean, a horizontal coordinate prediction variance, a vertical coordinate prediction mean and a vertical coordinate prediction variance.

[0014] Another aspect of the present application provides a training device for a graph neural localization model, comprising:

[0015] A first acquisition module is used to acquire a first training set and a second training set, wherein the first training set includes a plurality of first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a location label of a target object corresponding to each of the second channel state undirected graphs;

[0016] A pre-training module, used to pre-train the initial graph neural network using the plurality of the first channel state undirected graphs to obtain an intermediate graph neural network;

[0017] The target training module is used to train the initial positioning model using the multiple second channel state undirected graphs and the multiple position labels to obtain the graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

[0018] Another aspect of the present application provides a device for locating a target object, comprising:

[0019] A second acquisition module is used to acquire a target channel state matrix generated by a plurality of wireless devices when a transmitting end of the target object interacts with the plurality of wireless devices;

[0020] A generating module, configured to generate a target channel state undirected graph according to each of the target channel state matrices;

[0021] A prediction module, used for inputting the above-mentioned target channel state undirected graph into the graph neural localization model to obtain multiple initial position prediction sets;

[0022] The module is used to perform weighted average processing on the multiple initial position prediction sets to obtain the target position of the target object.

[0023] Another aspect of the present application provides an electronic device, comprising:

[0024] one or more processors;

[0025] a memory for storing one or more programs,

[0026] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0027] Another aspect of the present application provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the method described above when executed.

[0028] Another aspect of the present application provides a computer program product, which includes computer executable instructions, and the instructions are used to implement the method as described above when executed.

[0029] According to an embodiment of the present application, by using a first channel state undirected graph for pre-training, and then using a second channel state undirected graph and position labels to perform labeled training on the pre-trained intermediate graph neural network, a graph neural positioning model that can be used to locate the target object is obtained. By fully utilizing the high flexibility of the graph structure, a general graph neural positioning model of channel state information for actual commercial scenarios is implemented. At the same time, the model is pre-trained and learned through a large number of unlabeled first channel state undirected graphs, which improves the positioning robustness of the graph neural positioning model. An uncertainty learning strategy is introduced in the pre-training to cope with the complex and changeable environment in actual applications, further enhancing the positioning stability and reliability of the graph neural positioning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and other objects, features and advantages of the present application will become more apparent through the following description of the embodiments of the present application with reference to the accompanying drawings, in which:

[0031] Figure 1 An exemplary system architecture to which a training method of a graph neural localization model or a method for localizing a target object can be applied according to an embodiment of the present application is shown;

[0032] Figure 2 A flowchart of a method for training a graph neural localization model according to an embodiment of the present application is shown;

[0033] Figure 3 A flow chart showing a method for locating a target object according to an embodiment of the present application is shown;

[0034] Figure 4 A schematic diagram of a building-level indoor positioning scenario according to the first embodiment of the present application is shown;

[0035] Figure 5 A schematic diagram of data set points collected by building-level indoor positioning according to the first embodiment of the present application is shown;

[0036] Figure 6 A schematic diagram showing the 50th (median) and 90th (tail) percentile error scores and floor accuracy of different devices and methods according to the first embodiment of the present application is shown;

[0037] Figure 7 A schematic diagram showing points of a data set collected by building-level indoor positioning according to the second embodiment of the present application is shown;

[0038] Figure 8 A schematic diagram showing points of a data set collected by building-level indoor positioning according to the second embodiment of the present application is shown;

[0039] Fig. 9 A block diagram of a training device for a graph neural localization model according to an embodiment of the present application is shown;

[0040] Fig.10 A block diagram showing a target object positioning device according to an embodiment of the present application; and

[0041] Fig.11 A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0042] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.

[0043] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0044] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0045] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0046] In an indoor environment, commercial WiFi routers are deployed as receiving ends (Access Point, AP), that is, wireless devices, and the antenna spacing and antenna type of the receiving end. The WiFi routers are controlled by the central processor and AC (radio access controller) cluster to measure the channel state information (CSI) while ensuring normal communication. The channel state information of the user terminal device is represented as follows after being received by the receiving end: Let the receiving signal matrix of the receiving end be , its relationship with the transmitter input signal can be described by the following matrix equation: ,in: is the physical channel matrix, represents the number of receiving antennas, and is the cyclic shift matrix and spatial mapping matrix at the transmitter, is the transmit signal matrix.

[0047] The channel state information (CSI) matrix measured by the receiving end It is expressed as: ,in is the matrix product at the transmitter. Based on this model, the CSI matrix reported by the receiver is Represented as a three-dimensional matrix with dimensions , where each Quantity Corresponding to one subcarrier.

[0048] In order to explore the positioning scenario in a complex real indoor environment. In this configuration, the user device (such as a mobile phone, PC or laptop) acts as the transmitter and constitutes a collection , and the access points (APs) installed throughout the facility serve as the receiving end, forming a collection .

[0049] For the sake of illustration, a basic scenario is defined in which the transmitter In a specific time window Multiple receivers To interact, the time window is based on the time Centered for one second. Within , the transmitter performs channel estimation with multiple nearby APs. Each AP then transmits estimated channel state information (CSI) for each training sequence, denoted as ,in Usually indicates diversity or multiplexing mode. The total number of receivers involved is given by express.

[0050] The process of converting data collected by multiple nearby APs into a three-dimensional vector representing the location of the transmitter is called a "location event" (LocEvent), denoted as In order to maintain the reliability of the positioning results, all events involving less than three APs can be excluded, that is, events that meet the condition situation.

[0051] There are many challenges in using existing positioning methods to process the data of the above scenarios for positioning. First, traditional CSI fingerprint positioning solutions usually rely on vector encoding in Euclidean space. However, in actual commercial scenarios, this encoding method often fails due to the heterogeneity of receiving devices and differences in communication modes. Secondly, large-scale unlabeled CSI data is difficult to use in practice, and how to effectively use this data to improve positioning performance has not been fully addressed. In addition, the situation in the actual deployment environment is complex and changeable. How to maintain the robustness of the positioning solution under such conditions is still a major challenge that limits the widespread application of existing systems in actual environments.

[0052] In view of this, an embodiment of the present application provides a training method for a graph neural localization model, a method and device for localizing a target object, the method comprising obtaining a first training set and a second training set, wherein the first training set comprises a plurality of first channel state undirected graphs, and the second training set comprises a plurality of second channel state undirected graphs and location labels of target objects corresponding to each second channel state undirected graph; pre-training an initial graph neural network using the plurality of first channel state undirected graphs to obtain an intermediate graph neural network; training an initial localization model using the plurality of second channel state undirected graphs and a plurality of location labels to obtain a graph neural localization model, wherein the initial localization model comprises a plurality of mutually independent intermediate graph neural networks.

[0053] In the embodiments of this application, the collection, updating, analysis, processing, use, transmission, provision, disclosure, storage, etc. of the data involved (for example, including but not limited to user personal information) are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information and network security.

[0054] In the embodiments of the present application, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0055] Figure 1 An exemplary system architecture 100 is shown to which a training method of a graph neural localization model or a method for localizing a target object can be applied according to an embodiment of the present application. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present application can be applied, in order to help those skilled in the art understand the technical content of the present application, but it does not mean that the embodiments of the present application cannot be used in other devices, systems, environments or scenarios.

[0056] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0057] The user may use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (only for example).

[0058] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0059] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0060] It should be noted that the training method of the graph neural positioning model or the positioning method of the target object provided in the embodiment of the present application can generally be executed by the server 105. Accordingly, the training device of the graph neural positioning model or the positioning device of the target object provided in the embodiment of the present application can generally be set in the server 105. The training method of the graph neural positioning model or the positioning method of the target object provided in the embodiment of the present application can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Correspondingly, the training device of the graph neural positioning model or the positioning device of the target object provided in the embodiment of the present application can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Alternatively, the training method of the graph neural localization model or the method of locating the target object provided in the embodiment of the present application may also be performed by the first terminal device 101, the second terminal device 102 or the third terminal device 103, or may also be performed by other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103. Accordingly, the training device of the graph neural localization model or the positioning device of the target object provided in the embodiment of the present application may also be set in the first terminal device 101, the second terminal device 102 or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103.

[0061] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only . According to the implementation requirements, there can be any number of terminal devices, networks and servers.

[0062] Figure 2 A flowchart of a method for training a graph neural localization model according to an embodiment of the present application is shown.

[0063] like Figure 2 As shown, the training method of the neural localization model includes operations S201~S203.

[0064] In operation S201, a first training set and a second training set are obtained, wherein the first training set includes a plurality of first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a location label of a target object corresponding to each second channel state undirected graph;

[0065] In operation S202, the initial graph neural network is pre-trained using a plurality of first channel state undirected graphs to obtain an intermediate graph neural network;

[0066] In operation S203, an initial positioning model is trained using multiple second channel state undirected graphs and multiple position labels to obtain a graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

[0067] According to an embodiment of the present application, the target object may be a user, a terminal such as a mobile phone, a pet, etc.

[0068] According to an embodiment of the present application, the channel state undirected graph includes multiple nodes and edges between different nodes, the nodes are constructed based on the attribute information in the channel state information matrix of the wireless device, and the edges are constructed based on the amplitude, phase difference or channel impulse response (CIR) of the receiving antenna of the wireless device.

[0069] According to an embodiment of the present application, the initial graph neural network can be any type of graph neural network (GNN), such as a graph convolutional network (GCN), a graph auto-encoder (GAE), etc.

[0070] According to an embodiment of the present application, the initial graph neural network is first pre-trained using multiple first channel state undirected graphs to obtain a pre-trained intermediate graph neural network. Thereafter, the intermediate graph neural network is again trained with labels using the second channel state undirected graph and position labels to obtain a graph neural localization model.

[0071] According to an embodiment of the present application, by using a first channel state undirected graph for pre-training, and then using a second channel state undirected graph and position labels to perform labeled training on the pre-trained intermediate graph neural network, a graph neural positioning model that can be used to locate the target object is obtained. By fully utilizing the high flexibility of the graph structure, a general graph neural positioning model of channel state information for actual commercial scenarios is implemented. At the same time, the model is pre-trained and learned through a large number of unlabeled first channel state undirected graphs, which improves the positioning robustness of the graph neural positioning model. An uncertainty learning strategy is introduced in the pre-training to cope with the complex and changeable environment in actual applications, further enhancing the positioning stability and reliability of the graph neural positioning model.

[0072] According to an embodiment of the present application, the first training set also includes floor pseudo labels of target objects corresponding to each first channel state undirected graph.

[0073] According to an embodiment of the present application, the initial graph neural network is pre-trained using a plurality of first channel state undirected graphs to obtain an intermediate graph neural network, including:

[0074] For two first channel state undirected graphs having an associated relationship, the two first channel state undirected graphs are processed by using an initial graph neural network to obtain a first output set, wherein the first output set includes a time dimension feature and a space dimension feature corresponding to each first channel state undirected graph, and the space dimension feature includes a predicted floor and a predicted position of the first channel state undirected graph;

[0075] The two spatial dimension features are processed respectively by using a metric loss function to obtain a first loss value corresponding to each first channel state undirected graph;

[0076] The two time dimension features are processed respectively by using the contrast loss function to obtain the second loss value corresponding to the two first channel state undirected graphs;

[0077] For each first channel state undirected graph, the spatial dimension features and the floor pseudo labels are processed using the mean absolute error loss function to obtain a third loss function;

[0078] Generate a pre-training loss value according to the first loss value, the plurality of second loss values, and the plurality of third loss functions;

[0079] The network parameters of the initial graph neural network are iteratively adjusted according to the pre-training loss value to obtain a trained intermediate graph neural network.

[0080] According to an embodiment of the present application, the association relationship represents an association in time, and the two first channel state undirected graphs having an association relationship may be two first channel state undirected graphs with similar acquisition times.

[0081] According to an embodiment of the present application, the first loss value As shown in formula (1), the second loss value As shown in formula (2), the pre-training loss value As shown in formula (3):

[0082] (1)

[0083] in, is the predicted position of the target object corresponding to the i-th first channel state undirected graph, and are the real positions of the two wireless devices corresponding to the first channel state undirected graph, is a hyperparameter;

[0084]

[0085] (2)

[0086] in, and Represent two first channel state undirected graphs respectively and The time dimension characteristics;

[0087] (3)

[0088] in, and are the third loss functions of the undirected graphs corresponding to the two first channel states, and are the predicted floors corresponding to the two first channel state undirected graphs, and They are floor pseudo labels corresponding to two first channel state undirected graphs, and the floor pseudo labels are determined according to the floors of the wireless devices related to the first channel state undirected graphs.

[0089] According to an embodiment of the present application, a large amount of unlabeled data (i.e., a first channel state undirected graph) is pre-trained from the time and space dimensions. Specifically, when analyzing the unlabeled first channel state undirected graph, the timestamp information not only serves as an easily accessible resource, but also provides valuable prior information for positioning, considering the limited movement speed of the user (i.e., a target object). It can be reasonably assumed that two first channel state undirected graphs collected from the same user in a short time interval are likely to come from geographically close places, regardless of whether the user is stationary or moving. On the contrary, if the first channel state undirected graphs are collected over a long time interval, or come from devices of different users, the challenges in determining spatial similarity will increase significantly. The scheme is implemented using the following contrast loss function as shown in formula (2).

[0090] According to an embodiment of the present application, in addition to time domain information, the signals received by most wireless devices equipped with wireless functions also provide prior knowledge about the spatial dimension for positioning. In a positioning event (LocEvent), multiple access points (wireless device APs) participate in measuring the current position of the transmitter within a given time. This setting means that preliminary positioning results can be generated using the relative received power at different APs. Generally, under line-of-sight conditions, APs that record higher relative power are more likely to be close to the transmitter. Assuming that the positions of the APs are known, these positions are usually obtained through computer-aided design (CAD) drawings or marked during the collection of the first channel state undirected graph. Therefore, at the predicted position d in the output dimension, the predicted position d can be defined as being closer to the AP that receives higher signal power. The scheme is implemented using a metric learning loss function as shown in formula (1).

[0091] According to an embodiment of the present application, the floor number of the nearest AP is used as a pseudo-floor label to calculate the third loss function using spatial priors. Finally, the pre-training loss value of the pre-training process is shown in formula (3).

[0092] According to an embodiment of the present application, an initial positioning model is trained using multiple second channel state undirected graphs and multiple position labels to obtain a graph neural positioning model, including:

[0093] For each intermediate graph neural network, use the intermediate graph neural network to process each second channel state undirected graph to obtain a second output set, wherein the second output set includes a horizontal coordinate prediction value, a vertical coordinate prediction value, a floor prediction value, a horizontal coordinate variance, and a vertical coordinate variance;

[0094] Generate abscissa probability distribution information and ordinate probability distribution information according to the abscissa prediction value and the ordinate prediction value respectively;

[0095] For any probability distribution information of the horizontal coordinate probability distribution information and the vertical coordinate probability distribution information, generate a probability distribution divergence according to the probability distribution information and the coordinate label function value corresponding to the probability distribution information, wherein the position label includes the horizontal coordinate label, the vertical coordinate label and the floor label;

[0096] Generate floor loss value based on floor prediction value and floor label;

[0097] Generate a combined loss value according to the probability distribution divergence corresponding to the abscissa, the probability distribution divergence corresponding to the ordinate, and the floor loss value;

[0098] The network parameters of the intermediate graph neural network are iteratively adjusted according to the combined loss value to obtain a trained target graph neural network, wherein the graph neural localization model includes multiple target graph neural networks.

[0099] According to an embodiment of the present application, the probability distribution divergence of the horizontal axis probability distribution information As shown in formula (4), the combined loss value is shown in formula (5):

[0100] (4)

[0101] in, is the predicted value on the horizontal axis, is the mean of the predicted values ​​on the horizontal axis, is the horizontal axis variance, is the impulse response impact function value obtained based on the Dirac delta function with the horizontal axis label as the center, that is, the coordinate label function value;

[0102] (5)

[0103] in, is the probability distribution divergence of the probability distribution information on the horizontal axis, is the probability distribution divergence of the probability distribution information on the horizontal axis, is the floor loss value, is the predicted value of the floor, is the variance of the floor labels.

[0104] According to the embodiment of the present application, based on uncertainty adaptation, the present application uses graph neural network to achieve further accuracy improvement. The present application improves the traditional coordinate point prediction task so that it not only includes the mean The prediction of , including the prediction of variance σ, such as the horizontal axis variance, the vertical axis variance, thus defining the probability distribution of the horizontal axis prediction value x, the vertical axis variance. Given a Dirac delta function centered on a position label (such as the horizontal axis label x or the vertical axis label y), , , by minimizing the probability distribution through formula (4) and The KL divergence between them is the divergence of the probability distribution of the horizontal axis.

[0105] According to the embodiment of the present application, the probability distribution divergence of the ordinate is obtained based on the above principle, and the floor loss value between the floor prediction value and the floor label is calculated based on the L1 loss function. The combined loss value in this training can be calculated according to formula (5).

[0106] According to an embodiment of the present application, any one of the first channel state undirected graph and the second channel state undirected graph is generated in the following manner:

[0107] Acquire channel state matrices generated by different wireless devices, wherein the multiple channel state matrices are generated when the multiple wireless devices perform data interaction with a transmitting end of a target object within a same time window;

[0108] For each channel state matrix, construct nodes in a channel state undirected graph according to the attributes in the channel state matrix;

[0109] Based on the amplitude and receiving properties of the receiving antenna of the wireless device in the channel state matrix, edges between different nodes in the channel state undirected graph are generated, wherein the receiving properties include channel impulse response information or phase differences of different receiving antennas.

[0110] In a specific embodiment, the scenario is configured with one transmitter and two receivers (i.e., wireless devices). This scenario involves two channel state matrices, namely and , the dimension of each channel state matrix is . An undirected graph of channel states is constructed using five nodes: two sets of amplitudes and channel impulse responses (CIRs) — and , and two sets of phase differences—— and . These channel features serve as feature data for each corresponding node. To ensure consistency, we normalize the feature dimension to 245 and pad with zeros when necessary, or apply downsampling when it exceeds this dimension. In addition, other supplementary information, such as the location information of the access point (AP), the center frequency of the CSI, and the received signal strength (RSS), can also be encoded to further enrich the representation of the channel state undirected graph. Edges are formed in the graph structure by establishing connections based on the amplitude and phase difference of each receiving antenna, or by establishing connections between the amplitude and CIR of the same receiving antenna.

[0111] According to the embodiment of the present application, for more complex heterogeneous CSI data , i.e., the channel state information, is constructed based on the channel state undirected graph generation method described above. When dealing with multiple transmit antennas, they are first decomposed into pairs consisting of a transmit end and two receive ends, and then the undirected graph is constructed based on the channel state undirected graph generation method described above. Next, the amplitude nodes from the same transmit antenna are connected to complete the construction. For multiple receive antennas, the construction process is similar to the case of two antennas. After completing each After the construction, in the same positioning event LocEvent, each The amplitude nodes with the highest RSS in the channel are connected to complete the channel state undirected graph of LocEvent. 's construction.

[0112] Figure 3 A flow chart of a method for locating a target object according to an embodiment of the present application is shown.

[0113] like Figure 3 As shown, the method for locating a target object includes operations S301 to S304.

[0114] In operation S301, when a transmitting end of a target object interacts with a plurality of wireless devices, a target channel state matrix generated by the plurality of wireless devices is obtained;

[0115] In operation S302, for each target channel state matrix, a target channel state undirected graph is generated according to the target channel state matrix;

[0116] In operation S303, the target channel state undirected graph is input into a graph neural localization model to obtain a plurality of initial position prediction sets;

[0117] In operation S304, a weighted average process is performed on the multiple initial position prediction sets to obtain the target position of the target object.

[0118] According to an embodiment of the present application, in the process of the target object using a mobile phone to log in to the transmitter and interact with multiple wireless devices, multiple target channel state matrices within the same time window are obtained, and converted based on the channel state undirected graph generation method described above, thereby obtaining multiple target channel state undirected graphs, which are input into the trained graph neural localization model to obtain multiple initial position prediction sets, and then the multiple initial position prediction sets are weighted averaged to obtain the target position of the target object.

[0119] According to an embodiment of the present application, by using a first channel state undirected graph for pre-training, and then using a second channel state undirected graph and position labels to perform labeled training on the pre-trained intermediate graph neural network, a graph neural positioning model that can be used to locate the target object is obtained. By fully utilizing the high flexibility of the graph structure, a general graph neural positioning model of channel state information for actual commercial scenarios is implemented. At the same time, the model is pre-trained and learned through a large number of unlabeled first channel state undirected graphs, which improves the positioning robustness of the graph neural positioning model. An uncertainty learning strategy is introduced in the pre-training to cope with the complex and changeable environment in actual applications, further enhancing the positioning stability and reliability of the graph neural positioning model.

[0120] According to an embodiment of the present application, the initial position prediction set includes a horizontal coordinate prediction mean, a horizontal coordinate prediction variance, a vertical coordinate prediction mean, and a vertical coordinate prediction variance.

[0121] According to an embodiment of the present application, the target horizontal coordinate of the target position As shown in formula (6), the target ordinate of the target position is As shown in formula (7):

[0122] (6)

[0123] (7)

[0124] in, is the predicted mean on the horizontal axis, is the prediction variance on the horizontal axis, is the predicted mean value on the ordinate, is the prediction variance of the ordinate, and z is the zth target graph neural network in the graph neural localization model.

[0125] According to an embodiment of the present application, the number of target graph neural networks z can be 5. Each target graph neural network has the same structure and training data but uses a different random seed. Then, the final coordinate estimate is calculated using the weighted average formula, and the final target horizontal coordinate , target vertical coordinate As shown in formulas (6) and (7), for floor numbers, a similar weighted loss function can be used for calculation.

[0126] Figure 4 A schematic diagram of a building-level indoor positioning scenario according to the first embodiment of the present application is shown. Figure 5 A schematic diagram of data set points collected by building-level indoor positioning according to the first embodiment of the present application is shown. Figure 6 A schematic diagram showing the 50th (median) and 90th (tail) percentile error scores and floor accuracy of different devices and methods according to the first embodiment of the present application is shown.

[0127] In a first specific embodiment, using Figure 4 and Figure 5 The proposed method was verified in the building-level indoor positioning scenario shown in Figure 1. The dataset used contains about 40,000 samples (i.e., channel state matrices) collected by seven different types of collection devices, including various types of mobile phones or tablets. During data collection, these devices were used in a handheld or pocket manner.

[0128] In actual positioning environments, users usually have only a limited number of mobile devices, but they face a variety of different models of mobile phones in daily positioning services. Therefore, in order to simulate a more challenging and realistic scenario, this application proposes and verifies a performance test method of "leaving one mobile phone type". This method involves excluding a certain type of mobile phone from the training data set and using only that mobile phone in the test phase to evaluate the generalization ability of the graph neural positioning model across different device types.

[0129] Experimental results show that the present application can effectively encode heterogeneous and comprehensive channel state information (CSI) in the positioning event (LocEvent), thereby demonstrating strong coding capabilities. Figure 6 As shown in the figure, this detailed encoding faces certain challenges in terms of generalization across devices. In particular, when training from scratch, the performance of this application on some devices has declined. To address this problem, this application significantly improves the generalization ability of the model on different devices by introducing a pre-training strategy and uncertainty assessment mechanism.

[0130] Further experimental results (such as Figure 6 ) compares the 50% (median) and 90% (tail) percentile error scores, as well as the floor positioning accuracy (Acc%) of various devices and methods. The results show that the present application has significant advantages in computing speed and memory usage. Statistical analysis of all tested devices shows that the median positioning error of the present application's method is 2.17 meters and the 90% percentile error is 8.93 meters. In comparison, the median error of the vector-based method is 2.28 meters and the 90% percentile error is 12.16 meters. In addition, the floor positioning accuracy of the present application reaches 99.49%, which is higher than the 99.29% of the baseline method.

[0131] Overall, this application achieved an 18.7% improvement in mean absolute error, from 4.64 meters to 3.77 meters, while also outperforming the baseline method in terms of computational speed and memory usage. These advantages verify the efficiency and reliability of this application in complex real-world scenarios.

[0132] Figure 7 A schematic diagram showing the locations of data sets collected for building-level indoor positioning according to the second embodiment of the present application is shown. Figure 8 A schematic diagram showing the locations of data sets collected for building-level indoor positioning according to the second embodiment of the present application is shown.

[0133] In a second specific embodiment, an evaluation was conducted in a scene of approximately 4000 square meters (e.g. Figure 7 As shown in Figure 2, the data collection covers five different types of smartphones, with a total of about 30,000 samples collected. This application combines the time dimension and the spatial dimension with floor pseudo labels for experiments. Figure 8The experimental results shown show that when the training sample size is less than 60% of the total data, the time constraint can significantly improve the performance of the model, but this improvement tends to stabilize when the data size increases. In contrast, the topological constraint can continuously enhance the model performance at all sample sizes, although the marginal benefits of performance improvement gradually decrease as the data size increases. Finally, the best performance was achieved by integrating all components and architectures into the framework of this application.

[0134] Fig. 9 A block diagram of a training device for a graph neural localization model according to an embodiment of the present application is shown.

[0135] like Fig. 9 As shown, the training device 900 of the neural localization model includes a first acquisition module 910, a pre-training module 920, and a target training module 930.

[0136] A first acquisition module 910 is used to acquire a first training set and a second training set, wherein the first training set includes a plurality of first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a location label of a target object corresponding to each second channel state undirected graph;

[0137] A pre-training module 920, configured to pre-train an initial graph neural network using a plurality of first channel state undirected graphs to obtain an intermediate graph neural network;

[0138] The target training module 930 is used to train the initial positioning model using multiple second channel state undirected graphs and multiple position labels to obtain a graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

[0139] According to an embodiment of the present application, by using the first channel state undirected graph for pre-training, and then using the second channel state undirected graph and position labels to perform labeled training on the pre-trained intermediate graph neural network, a graph neural positioning model that can be used to locate the target object is obtained. Since the high flexibility of the graph structure is fully utilized, a general graph neural positioning model of channel state information for actual commercial scenarios is realized. At the same time, the model is pre-trained and learned through a large number of unlabeled first channel state undirected graphs, which improves the positioning robustness of the graph neural positioning model. An uncertainty learning strategy is introduced in the pre-training to cope with the complex and changeable environment in actual applications, further enhancing the positioning stability and reliability of the graph neural positioning model.

[0140] According to an embodiment of the present application, the first training set also includes floor pseudo labels of target objects corresponding to each first channel state undirected graph.

[0141] According to an embodiment of the present application, the pre-training module 920 includes:

[0142] A first obtaining unit is used to process any two first channel state undirected graphs using an initial graph neural network to obtain a first output set, wherein the first output set includes time dimension features and space dimension features corresponding to each first channel state undirected graph, and the space dimension features include a predicted floor and a predicted position of the first channel state undirected graph;

[0143] A second obtaining unit is used to process the two spatial dimension features respectively by using a metric loss function to obtain a first loss value corresponding to each first channel state undirected graph;

[0144] A third obtaining unit is used to process the two time dimension features respectively using a contrast loss function to obtain second loss values ​​corresponding to the two first channel state undirected graphs;

[0145] A fourth obtaining unit is used for processing the spatial dimension features and the floor pseudo labels using the mean absolute error loss function for each first channel state undirected graph to obtain a third loss function;

[0146] A first generating unit, configured to generate a pre-training loss value according to the first loss value, the plurality of second loss values, and the plurality of third loss functions;

[0147] The fifth obtaining unit is used to iteratively adjust the network parameters of the initial graph neural network according to the pre-training loss value to obtain a trained intermediate graph neural network.

[0148] According to an embodiment of the present application, the target training module 930 includes:

[0149] A sixth obtaining unit is used for processing each second channel state undirected graph using the intermediate graph neural network for each intermediate graph neural network to obtain a second output set, wherein the second output set includes a horizontal coordinate prediction value, a vertical coordinate prediction value, a floor prediction value, a horizontal coordinate variance, and a vertical coordinate variance;

[0150] A second generating unit, used to generate abscissa probability distribution information and ordinate probability distribution information according to the abscissa prediction value and the ordinate prediction value respectively;

[0151] A third generating unit is used to generate a probability distribution divergence for any probability distribution information of the horizontal coordinate probability distribution information and the vertical coordinate probability distribution information according to the probability distribution information and the coordinate label function value corresponding to the probability distribution information, wherein the position label includes the horizontal coordinate label, the vertical coordinate label and the floor label;

[0152] A fourth generating unit, used for generating a floor loss value according to the floor prediction value and the floor label;

[0153] a fifth generating unit, configured to generate a combined loss value according to the probability distribution divergence corresponding to the abscissa, the probability distribution divergence corresponding to the ordinate, and the floor loss value;

[0154] The seventh obtaining unit is used to iteratively adjust the network parameters of the intermediate graph neural network according to the combined loss value to obtain a trained target graph neural network, wherein the graph neural localization model includes multiple target graph neural networks.

[0155] According to an embodiment of the present application, any one of the first channel state undirected graph and the second channel state undirected graph is generated in the following manner:

[0156] An acquisition unit, configured to acquire channel state matrices generated by different wireless devices, wherein the multiple channel state matrices are generated when multiple wireless devices perform data interaction with a transmitting end of a target object within a same time window;

[0157] A construction unit, for constructing nodes in a channel state undirected graph according to attributes in the channel state matrix for each channel state matrix;

[0158] The sixth generating unit is used to generate edges between different nodes in the channel state undirected graph based on the amplitude and receiving properties of the receiving antenna of the wireless device in the channel state matrix, wherein the receiving properties include channel impulse response information or phase difference of different receiving antennas.

[0159] Fig.10 A block diagram of a target object positioning device according to an embodiment of the present application is shown.

[0160] like Fig.10 As shown, the target object positioning device 1000 includes a second acquisition module 1010 , a generation module 1020 , a prediction module 1030 , and an acquisition module 1040 .

[0161] A second acquisition module 1010 is used to acquire a target channel state matrix generated by multiple wireless devices when a transmitting end of a target object interacts with multiple wireless devices;

[0162] A generating module 1020, configured to generate a target channel state undirected graph according to the target channel state matrix for each target channel state matrix;

[0163] A prediction module 1030, configured to input the target channel state undirected graph into a graph neural localization model to obtain a plurality of initial position prediction sets;

[0164] The obtaining module 1040 is used to perform weighted average processing on multiple initial position prediction sets to obtain the target position of the target object.

[0165] According to an embodiment of the present application, the initial position prediction set includes a horizontal coordinate prediction mean, a horizontal coordinate prediction variance, a vertical coordinate prediction mean, and a vertical coordinate prediction variance.

[0166] According to an embodiment of the present application, by using the first channel state undirected graph for pre-training, and then using the second channel state undirected graph and position labels to perform labeled training on the pre-trained intermediate graph neural network, a graph neural positioning model that can be used for positioning the target object is obtained. Since the high flexibility of the graph structure is fully utilized, a general graph neural positioning model of channel state information for actual commercial scenarios is realized. At the same time, the model is pre-trained and learned through a large number of unlabeled first channel state undirected graphs, which improves the positioning robustness of the graph neural positioning model. An uncertainty learning strategy is introduced in the pre-training to cope with the complex and changeable environment in practical applications, and the positioning stability and reliability of the graph neural positioning model are further enhanced. According to the embodiments of the present application, any multiple of the modules, sub-modules, units, and sub-units, or at least part of the functions of any multiple of them can be implemented in one module. According to the embodiments of the present application, any one or more of the modules, sub-modules, units, and sub-units can be split into multiple modules for implementation. According to the embodiments of the present application, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application-specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging circuits, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, according to the embodiments of the present application, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, which can perform corresponding functions when the computer program modules are run.

[0167] For example, any of the first acquisition module 910, the pre-training module 920, the target training module 930, or the second acquisition module 1010, the generation module 1020, the prediction module 1030, and the acquisition module 1040 can be combined in one module / unit / sub-unit for implementation, or any of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present application, at least one of the first acquisition module 910, the pre-training module 920, the target training module 930, or the second acquisition module 1010, the generation module 1020, the prediction module 1030, and the acquisition module 1040 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the first acquisition module 910, the pre-training module 920, the target training module 930, or the second acquisition module 1010, the generation module 1020, the prediction module 1030, and the acquisition module 1040 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.

[0168] It should be noted that the training device of the graph neural localization model or the positioning device of the target object in the embodiments of the present application corresponds to the training method of the graph neural localization model or the positioning method of the target object in the embodiments of the present application. The description of the training device of the graph neural localization model or the positioning device of the target object specifically refers to the training method of the graph neural localization model or the positioning method of the target object, which will not be repeated here.

[0169] Fig.11 A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present application is shown. Fig.11 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0170] like Fig.11As shown, the electronic device 1100 according to an embodiment of the present application includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage part 1108 to a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include an onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present application.

[0171] In RAM 1103, various programs and data required for the operation of electronic device 1100 are stored. Processor 1101, ROM 1102 and RAM 1103 are connected to each other via bus 1104. Processor 1101 performs various operations of the method flow according to the embodiment of the present application by executing the program in ROM 1102 and / or RAM 1103. It should be noted that the program can also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 can also perform various operations of the method flow according to the embodiment of the present application by executing the program stored in the one or more memories.

[0172] According to an embodiment of the present application, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to the bus 1104. The electronic device 1100 may further include one or more of the following components connected to the input / output (I / O) interface 1105: an input portion 1106 including a keyboard, a mouse, etc.; an output portion 1107 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 1108 including a hard disk, etc.; and a communication portion 1109 including a network interface card such as a LAN card, a modem, etc. The communication portion 1109 performs communication processing via a network such as the Internet. The drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed, so that the computer program read therefrom is installed into the storage portion 1108 as needed.

[0173] According to an embodiment of the present application, the method flow according to the embodiment of the present application can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above-mentioned functions defined in the system of the embodiment of the present application are executed. According to an embodiment of the present application, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0174] The present application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present application is implemented.

[0175] According to an embodiment of the present application, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0176] For example, according to an embodiment of the present application, the computer-readable storage medium may include the ROM 1102 and / or the RAM 1103 described above and / or one or more memories other than the ROM 1102 and the RAM 1103 .

[0177] An embodiment of the present application also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present application. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the method provided by the embodiment of the present application.

[0178] When the computer program is executed by the processor 1101, the above functions defined in the system / device of the embodiment of the present application are executed. According to the embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0179] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 1109, and / or installed from a removable medium 1111. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0180] The embodiments of the present application are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present application. Although each embodiment is described above, this does not mean that the measures in each embodiment cannot be used in combination advantageously. The present application does not depart from the scope of the present application, and those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present application.

Claims

1. A training method for a graph neural localization model, characterized in that: include: Acquire a first training set and a second training set, wherein the first training set includes a plurality of first channel state undirected graphs and a floor pseudo label of a target object corresponding to each of the first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a position label of a target object corresponding to each of the second channel state undirected graphs; For two first channel state undirected graphs having an association relationship, use an initial graph neural network to process the two first channel state undirected graphs to obtain a first output set, wherein the first output set includes a time dimension feature and a space dimension feature corresponding to each of the first channel state undirected graphs, the space dimension feature includes a predicted floor and a predicted position of the first channel state undirected graph, and the association relationship represents an association in time; Using a metric loss function to process the two spatial dimension features respectively, to obtain a first loss value corresponding to each of the first channel state undirected graphs; Using a contrast loss function to process the two time dimension features respectively, to obtain second loss values ​​corresponding to the two first channel state undirected graphs; For each of the first channel state undirected graphs, use a mean absolute error loss function to process the spatial dimension features and the floor pseudo-labels to obtain a third loss function; Generate a pre-training loss value according to the first loss value, a plurality of the second loss values, and a plurality of the third loss functions; Iteratively adjusting the network parameters of the initial graph neural network according to the pre-training loss value to obtain a trained intermediate graph neural network; The initial positioning model is trained using multiple second channel state undirected graphs and multiple position labels to obtain the graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

2. The method according to claim 1, characterized in that First loss value As shown in formula (1), the second loss value As shown in formula (2), the pre-training loss value As shown in formula (3): （1） in, is the predicted position of the target object corresponding to the i-th first channel state undirected graph, and are the real positions of the two wireless devices corresponding to the first channel state undirected graph, is a hyperparameter; （2） in, and Respectively represent the two first channel state undirected graphs and The time dimension characteristics; （3） in, and are the third loss functions of the undirected graphs corresponding to the two first channel states, and are the predicted floors corresponding to the two first channel state undirected graphs, and They are floor pseudo labels corresponding to two first channel state undirected graphs, respectively, and the floor pseudo labels are determined according to the floors of the wireless devices related to the first channel state undirected graphs.

3. The method according to claim 1, characterized in that Using a plurality of the second channel state undirected graphs and a plurality of the position tags to train an initial positioning model to obtain the graph neural positioning model includes: For each of the intermediate graph neural networks, use the intermediate graph neural network to process each of the second channel state undirected graphs to obtain a second output set, wherein the second output set includes a horizontal coordinate prediction value, a vertical coordinate prediction value, a floor prediction value, a horizontal coordinate variance, and a vertical coordinate variance; Generate abscissa probability distribution information and ordinate probability distribution information according to the abscissa prediction value and the ordinate prediction value respectively; For any probability distribution information of the horizontal coordinate probability distribution information and the vertical coordinate probability distribution information, generate a probability distribution divergence according to the probability distribution information and the coordinate label function value corresponding to the probability distribution information, wherein the position label includes a horizontal coordinate label, a vertical coordinate label and a floor label; Generate a floor loss value according to the floor prediction value and the floor label; Generate a combined loss value according to the probability distribution divergence corresponding to the abscissa, the probability distribution divergence corresponding to the ordinate, and the floor loss value; The network parameters of the intermediate graph neural network are iteratively adjusted according to the combined loss value to obtain a trained target graph neural network, wherein the graph neural localization model includes a plurality of the target graph neural networks.

4. The method according to claim 3, characterized in that The probability distribution divergence of the horizontal axis probability distribution information As shown in formula (4), the combined loss value is shown in formula (5): （4） in, is the predicted value on the horizontal axis, is the mean of the predicted values ​​on the horizontal axis, is the horizontal axis variance, is the impulse response impact function value obtained based on the Dirac delta function with the horizontal axis label as the center, that is, the coordinate label function value; （5） in, is the probability distribution divergence of the probability distribution information on the horizontal axis, is the probability distribution divergence of the probability distribution information on the horizontal axis, is the floor loss value, is the predicted value of the floor, is the variance of the floor labels.

5. The method according to claim 1, characterized in that Any one of the first channel state undirected graph and the second channel state undirected graph is generated in the following manner: Acquire channel state matrices generated by different wireless devices, wherein the plurality of channel state matrices are generated when the plurality of wireless devices perform data interaction with the transmitting end of the target object within the same time window; For each of the channel state matrices, constructing nodes in the channel state undirected graph according to attributes in the channel state matrix; Based on the amplitude and receiving properties of the receiving antenna of the wireless device in the channel state matrix, edges between different nodes in the channel state undirected graph are generated, wherein the receiving properties include channel impulse response information or phase differences of different receiving antennas.

6. A method for locating a target object, characterized in that: include In a case where a transmitting end of the target object interacts with a plurality of wireless devices, obtaining a target channel state matrix generated by the plurality of wireless devices; For each of the target channel state matrices, generating a target channel state undirected graph according to the target channel state matrix; Inputting the target channel state undirected graph into a graph neural localization model to obtain a plurality of initial position prediction sets, wherein the graph neural localization model is trained by the method of any one of claims 1 to 5; A weighted average process is performed on the multiple initial position prediction sets to obtain the target position of the target object.

7. The method according to claim 6, characterized in that The initial position prediction set includes a horizontal coordinate prediction mean, a horizontal coordinate prediction variance, a vertical coordinate prediction mean and a vertical coordinate prediction variance; Among them, the target horizontal coordinate of the target position As shown in formula (6), the target ordinate of the target position is As shown in formula (7): （6） （7） in, is the predicted mean on the horizontal axis, is the prediction variance on the horizontal axis, is the predicted mean value on the ordinate, is the prediction variance of the ordinate, and z is the zth target graph neural network in the graph neural localization model.

8. A training device for a graph neural localization model, characterized in that: include: A first acquisition module is used to acquire a first training set and a second training set, wherein the first training set includes a plurality of first channel state undirected graphs and a floor pseudo label of a target object corresponding to each of the first channel state undirected graphs, and the second training set includes a plurality of second channel state undirected graphs and a position label of a target object corresponding to each of the second channel state undirected graphs; Pre-training modules, including: A first obtaining unit is used to process any two first channel state undirected graphs using an initial graph neural network to obtain a first output set, wherein the first output set includes time dimension features and space dimension features corresponding to each first channel state undirected graph, and the space dimension features include a predicted floor and a predicted position of the first channel state undirected graph; A second obtaining unit is used to process the two spatial dimension features respectively by using a metric loss function to obtain a first loss value corresponding to each first channel state undirected graph; A third obtaining unit is used to process the two time dimension features respectively using a contrast loss function to obtain second loss values ​​corresponding to the two first channel state undirected graphs; A fourth obtaining unit is used for processing the spatial dimension features and the floor pseudo labels using the mean absolute error loss function for each first channel state undirected graph to obtain a third loss function; A first generating unit, configured to generate a pre-training loss value according to the first loss value, the plurality of second loss values, and the plurality of third loss functions; A fifth obtaining unit, used for iteratively adjusting the network parameters of the initial graph neural network according to the pre-training loss value to obtain a trained intermediate graph neural network; A target training module is used to train an initial positioning model using multiple undirected graphs of the second channel states and multiple position labels to obtain the graph neural positioning model, wherein the initial positioning model includes multiple independent intermediate graph neural networks.

9. A positioning device for a target object, characterized in that: include: A second acquisition module is used to acquire a target channel state matrix generated by a plurality of wireless devices when a transmitting end of the target object interacts with the plurality of wireless devices; A generating module, configured to generate a target channel state undirected graph according to each target channel state matrix; A prediction module, used for inputting the target channel state undirected graph into a graph neural localization model to obtain a plurality of initial position prediction sets, wherein the graph neural localization model is trained by the method according to any one of claims 1 to 5; The obtaining module is used to perform weighted average processing on the multiple initial position prediction sets to obtain the target position of the target object.

Images