Inertial data position conversion method and system based on noise codebook and multiple physical discriminators

Abstract

The application discloses an inertial data position conversion method and system based on a noise codebook and multiple physical discriminators, and the method comprises the following steps: acquiring a target position of a target device and first inertial data; performing multi-dimensional data enhancement processing on the first inertial data to obtain second inertial data; constructing a position-adaptive first noise codebook according to the second inertial data; performing extraction operation on the second inertial data through a decoupling generator to obtain motion semantic features and position noise features; obtaining a target noise code word according to the first noise codebook and the position noise features; and obtaining target inertial data of the target position through a gated fusion mechanism according to the target noise code word, the motion semantic features and the position noise features. The application can improve the robustness of inertial data and can be widely applied to the technical field of inertial data processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of inertial data processing, and in particular to an inertial data position conversion method and system based on a noise codebook and a multi-physical discriminator. BACKGROUND

[0002] Inertial sensors can perceive the motion state of an object in real time and become one of the key technologies for achieving high-precision positioning. However, inertial sensor data is easily affected by the placement position of the device, resulting in significant noise characteristics. Traditional methods rely on a single noise model or static feature matching, which is difficult to capture the dynamic superposition of motion artifact noise and environmental coupling noise. At the same time, although most of the current generative models have certain ability in data generation, they lack explicit constraints on physical laws. This often causes the generated trajectories to have problems such as sudden changes in speed and discontinuous attitude, which cannot effectively solve the noise processing and data generation problems in inertial data position conversion, thereby reducing the accuracy and reliability of inertial sensor data. SUMMARY

[0003] Therefore, the embodiments of the present application mainly aim to provide an inertial data position conversion method and system based on a noise codebook and a multi-physical discriminator, so as to solve at least one of the problems in the prior art, and the present application can improve the robustness of inertial data.

[0004] To achieve the above-mentioned purpose, one aspect of the embodiments of the present application provides an inertial data position conversion method based on a noise codebook and a multi-physical discriminator, which comprises:

[0005] obtaining a target position of a target device and first inertial data;

[0006] performing multi-dimensional data enhancement processing on the first inertial data to obtain second inertial data;

[0007] constructing a first noise codebook adaptive to the position according to the second inertial data;

[0008] extracting the second inertial data by a decoupling generator to obtain motion semantic features and position noise features;

[0009] obtaining a target noise code word according to the first noise codebook and the position noise features;

[0010] obtaining target inertial data of the target position by a gated fusion mechanism according to the target noise code word, the motion semantic features and the position noise features.

[0011] In some embodiments, an inertial data position conversion method based on a noise codebook and a multi-physical discriminator further comprises:

[0012] The target inertial data is jointly verified by a multi-physical discriminator to obtain a verification result.

[0013] The first noise codebook is pre-trained, and the decoupling generator and the multi-physical discriminator are trained.

[0014] In some embodiments, the multi-dimensional data enhancement processing on the first inertial data to obtain second inertial data includes the following steps:

[0015] The first acceleration in the first inertial data is subjected to time domain scaling processing to generate multi-rate motion data;

[0016] A linear frequency modulation disturbance is applied to the first frequency in the first inertial data to obtain a second frequency;

[0017] A random rotation matrix is generated in a three-dimensional rotation group space to simulate the orientation state of the target device;

[0018] A noise vector of the target position is randomly sampled, and the noise vector is superimposed on the first inertial data to simulate position interference;

[0019] The second inertial data is obtained according to the multi-rate motion data, the second frequency, the orientation state, and the position interference.

[0020] In some embodiments, the first noise codebook adaptive to the position is constructed according to the second inertial data, including the following steps:

[0021] The second inertial data is subjected to wavelet packet decomposition to obtain a frequency band energy distribution matrix;

[0022] The frequency band energy distribution matrix is screened to obtain a high-discrimination feature;

[0023] The high-discrimination feature is classified by a clustering algorithm to generate the first noise codebook adaptive to the position.

[0024] In some embodiments, the extraction operation on the second inertial data by the decoupling generator to obtain motion semantic features and position noise features includes the following steps:

[0025] The motion semantic features are obtained by the motion branch of the decoupling generator using a deep residual network to perform extraction operation on the second inertial data;

[0026] The position noise features are obtained by the noise branch of the decoupling generator using a bidirectional long short-term memory network to perform extraction operation on the second inertial data.

[0027] In some embodiments, the obtaining the target noise code word according to the first noise code book and the position noise feature comprises the following steps of:

[0028] obtaining each first code word of the first noise code book;

[0029] obtaining a similarity between each first code word and the position noise feature;

[0030] selecting the first code word corresponding to the highest similarity as the target noise code word.

[0031] In some embodiments, the obtaining the target inertia data of the target position according to the target noise code word, the motion semantic feature and the position noise feature through a gated fusion mechanism comprises the following steps of:

[0032] obtaining a first sparse coding coefficient of the motion semantic feature through sparse coding;

[0033] obtaining a second sparse coding coefficient of the position noise feature according to the target noise code word;

[0034] converting the first sparse coding coefficient and the second sparse coding coefficient into a gating vector through a weight matrix and an activation function;

[0035] fusing the motion semantic feature and the position noise feature according to the gating vector to obtain the target inertia data.

[0036] To achieve the above object, another aspect of the embodiment of the present application proposes an inertia data position conversion system based on a noise code book and a multi-physical discriminator, which comprises:

[0037] a first module for obtaining a target position and first inertia data of a target device;

[0038] a second module for performing multi-dimensional data enhancement processing on the first inertia data to obtain second inertia data;

[0039] a third module for constructing a position-adaptive first noise code book according to the second inertia data;

[0040] a fourth module for performing extraction operation on the second inertia data through a decoupling generator to obtain a motion semantic feature and a position noise feature;

[0041] a fifth module for obtaining a target noise code word according to the first noise code book and the position noise feature;

[0042] A sixth module is configured to obtain target inertia data of the target position by a gating fusion mechanism according to the target noise code word, the motion semantic feature, and the position noise feature.

[0043] To achieve the above object, another aspect of the embodiment of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned inertia data position conversion method based on a noise codebook and a multi-physical discriminator when executing the computer program.

[0044] To achieve the above object, another aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program implements the above-mentioned inertia data position conversion method based on a noise codebook and a multi-physical discriminator when executed by a processor.

[0045] To achieve the above object, another aspect of the embodiment of the present application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the above-mentioned inertia data position conversion method based on a noise codebook and a multi-physical discriminator.

[0046] The embodiment of the present application at least has the following beneficial effects: the present application provides an inertia data position conversion method and system based on a noise codebook and a multi-physical discriminator, which obtains a target position and first inertia data of a target device; performs multi-dimensional data enhancement processing on the first inertia data to obtain second inertia data; constructs a position-adaptive first noise codebook according to the second inertia data; performs extraction operation on the second inertia data by a decoupling generator to obtain motion semantic features and position noise features; obtains a target noise code word according to the first noise codebook and the position noise features; and obtains target inertia data of the target position by a gating fusion mechanism according to the target noise code word, the motion semantic features, and the position noise features, which can improve the robustness of the inertia data. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0048] Figure 1This is a schematic diagram of the implementation environment for performing inertial data position conversion according to an embodiment of the present invention;

[0049] Figure 2 This is a flowchart of the inertial data position conversion method based on noise codebook and multi-physics discriminator provided in the embodiments of the present invention;

[0050] Figure 3 This is a schematic diagram illustrating the overall steps of the inertial data position conversion process based on a noise codebook and a multi-physics discriminator.

[0051] Figure 4 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0053] It should be noted that although functional modules are divided in the system diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100" and "second / S200" in the specification, claims, and the foregoing drawings may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of the embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to a determination," or "in the event of a determination."

[0054] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0056] Before providing a detailed description of the embodiments of the present invention, some of the nouns and terms involved in the embodiments of the present invention will be explained first. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0057] Inertial data is typically used to describe data in the fields of inertial measurement, inertial navigation, or related technologies. For example, in an inertial measurement unit (IMU), inertial data may include information such as acceleration and angular velocity, used to measure and describe the motion state of an object.

[0058] A decoupling generator is a component or module typically used to describe a model in decoupling representation learning or generative models. In decoupling representation learning, the role of a decoupling generator is to separate the latent generative factors of data (such as the background, texture, and shape of an image) and generate new data with specific features through independent encoding and generation processes.

[0059] A discriminator is an important component in generative adversarial networks (GANs) in machine learning, used to distinguish between real and generated data. In signal processing or communication technologies, discriminators may also be used for signal classification or recognition.

[0060] Codebooks are widely used in fields such as communication technology and signal processing. For example, in 5G networks, codebooks are used to define a set of precoding matrices, which play an important role in the context transmission of terminals with CSI-RS as the reference signal. Noise codebooks, on the other hand, are used in signal processing or communication systems to describe the codebook structure containing noise information.

[0061] Codewords are widely used in signal processing, coding theory, communication systems, and machine learning. They are a basic unit in a codebook, typically referring to a specific vector or entry used to represent a feature or pattern. Noise codewords are a specific application of codewords, specifically used to represent noise features.

[0062] It is understood that the inertial data position conversion method based on noise codebook and multi-physics discriminator provided in this embodiment of the invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet, laptop, or desktop computer, but it is not limited to these.

[0063] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided by an embodiment of the invention. (Refer to...) Figure 1 The implementation environment includes at least one terminal 101 and a server 102. Terminal 101 and server 102 can connect wirelessly or via a wired network to exchange data. Terminal 101 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Server 101 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0064] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0065] In the field of indoor positioning, inertial sensors can sense the motion state of objects in real time, making them one of the key technologies for achieving high-precision positioning. However, inertial sensor data is easily affected by the placement of the device, resulting in significant noise characteristics. For example, the high-frequency jitter noise of a handheld device and the low-frequency swaying noise of a backpack, which is cushioned, have different spectral characteristics.

[0066] Traditional methods rely on single noise models (such as Gaussian white noise models) or static feature matching, making it difficult to capture the dynamic superposition of motion artifact noise (such as non-rigid motion noise generated by the coupling of human movement and equipment placement) and environmental coupling noise (such as mechanical coupling between equipment and the external environment, such as sensor disturbances caused by clothing friction, backpack shaking, etc.). Furthermore, while most current generative models possess some data generation capabilities, they lack explicit constraints on physical laws. Therefore, existing technologies have not solved the problem of dynamic noise modeling, and generative models lack explicit physical constraints. This often results in generated trajectories exhibiting abrupt velocity changes and attitude discontinuities, failing to effectively address noise processing and data generation issues in inertial data position conversion, thus reducing the accuracy and reliability of indoor positioning.

[0067] In view of this, such as Figure 2 As shown, this embodiment of the invention provides an inertial data position conversion method based on a noise codebook and a multi-physics discriminator. This method may include, but is not limited to, steps S100 to S600:

[0068] Step S100: Obtain the target position and first inertial data of the target device;

[0069] Step S200: Perform multi-dimensional data augmentation processing on the first inertial data to obtain the second inertial data;

[0070] Step S300: Construct a position-adaptive first noise codebook based on the second inertial data;

[0071] Step S400: The second inertial data is extracted using a decoupling generator to obtain motion semantic features and position noise features;

[0072] Step S500: Obtain the target noise codeword based on the first noise codebook and the position noise features;

[0073] Step S600: Based on the target noise codeword, the motion semantic features, and the position noise features, the target inertial data of the target position is obtained through a gating fusion mechanism.

[0074] In steps S100 to S600 of some embodiments, to address the problem of noise feature transfer of inertial sensor data generated by the target device under different target placement positions, a noise codebook is constructed for dynamic noise modeling, and high-fidelity inertial data conversion that conforms to physical laws is achieved by decoupling motion semantics and position noise, thereby improving the accuracy and reliability of target inertial data and significantly enhancing the robustness and environmental adaptability of the indoor positioning system.

[0075] In step S100 of some embodiments, the target locations of multiple target devices and their corresponding raw inertial sensor data are acquired. This inertial sensor data includes six-degree-of-freedom time-series data collected by accelerometers and gyroscopes. For example, the inertial measurement unit (IMU) of a mobile device collects six-degree-of-freedom time-series data of acceleration and gyroscope to obtain first inertial data, and records a target device placement location label (e.g., handheld, backpack). Through the above data collection process, it can be ensured that during the user's daily movement, the actual device placement location label and detailed information on the device's dynamic acceleration and rotational angular velocity are captured during the data collection process. The inertial data and position status are represented as follows:

[0076] x=[α X ,α Y ,α z ,ω X ,ω Y ,ω Z ]

[0077] y = [y source ]

[0078] In the formula, α X ,α Y ,α Z These represent the linear accelerations collected by the accelerometer along the X, Y, and Z axes in the world coordinate system, respectively; ω X ,ω Y ,ω Z These represent the angular velocities of the gyroscope along the X, Y, and Z axes in the world coordinate system, respectively; x represents the first inertial data; y source y represents the location state of the target device; y represents the target location information. Optionally, one-hot encoding is used to encode the location information to facilitate subsequent processing.

[0079] In some optional embodiments, taking an indoor environment as an example, this embodiment of the invention uses an inertial measurement unit (IMU) on a communication mobile device to record data on volunteers' actions such as walking straight and turning in an indoor environment using a mobile phone with built-in IMU sensors (accelerometer, gyroscope, and magnetometer). Each action is performed for at least 60 seconds to ensure sufficient data acquisition, with a sampling frequency of 200Hz. Throughout the entire data collection process, the mobile phone is placed in a fixed position, such as in a pocket, backpack, or handheld. This data collection process ensures that detailed information about the device's dynamic acceleration and rotational angular velocity can be collected in various everyday mobile phone placement scenarios.

[0080] In some embodiments, step S200 may include, but is not limited to, steps S210 to S250:

[0081] Step S210: Perform time-domain scaling on the first acceleration in the first inertial data to generate multi-rate motion data;

[0082] Step S220: Apply a linear frequency modulation perturbation to the first frequency in the first inertial data to obtain the second frequency;

[0083] Step S230: Generate a random rotation matrix in the three-dimensional rotation group space to simulate the orientation state of the target device;

[0084] Step S240: Randomly sample the noise vector at the target location, and superimpose the noise vector onto the first inertial data to simulate position interference;

[0085] Step S250: Obtain the second inertial data based on the multi-rate motion data, the second frequency, the orientation state, and the position disturbance.

[0086] In steps S210 to S250 of some embodiments, the first inertial data is subjected to multi-dimensional data augmentation processing, which may include motion state transformation based on time-domain stretching and compression and frequency-domain linear frequency modulation perturbation; orientation randomization based on random rotation in three-dimensional rotation group space; and resampling position-related noise vectors from noise codebook and injecting them into the inertial sensor data.

[0087] In steps S210 to S220 of some embodiments, a multi-rate motion pattern is generated by time-domain stretching and compression, and a linear frequency modulation perturbation is applied in the frequency domain to simulate step frequency fluctuations. For example, in terms of motion state transition, a time scaling factor α∈[0.7, 1.3] is applied to the first acceleration of the first inertial data to simulate motion data under multi-velocity conditions; a linear frequency modulation is added in the frequency domain to simulate the natural small changes in step frequency. The second frequency f′ after adding the linear frequency modulation is f′=f+Δf, where Δf is a randomly generated linear frequency modulation frequency within the range [-0.2Hz, +0.2Hz], and f is the first frequency in the first inertial data.

[0088] In step S230 of some embodiments, regarding device pose randomization, random rotation is performed in SO(3) space to simulate the state of the phone when it is facing different directions. Exemplarily, three angles θ are randomly generated. x ,θ y ,θ z Let and represent the rotation angles around the X, Y, and Z axes, respectively. Based on the generated random angles, construct rotation matrices R around the X, Y, and Z axes, respectively. x (θ x ), R y (θ y ), R z (θ z By multiplying the three rotation matrices using matrix multiplication, we can obtain the final random rotation matrix R. Here, the rotation matrix R around the X-axis is... x (θ x It has the following forms:

[0089]

[0090] Rotation matrix R around the Y-axis y (θ y It has the following forms:

[0091]

[0092] Rotation matrix R around the Z-axis z (θ z It has the following forms:

[0093]

[0094] The expression for the random rotation matrix R is:

[0095] R = R x (θ x )×R y (θ y )×R z (θ z )

[0096] A random rotation matrix represents the combined rotational effect after rotating sequentially around the X, Y, and Z axes. By generating a random rotation matrix in three-dimensional rotation space, dynamic changes in the orientation of the device can be achieved.

[0097] In step S240 of some embodiments, a position-related noise vector is randomly sampled from the second noise codebook and superimposed on the sensor noise data to simulate position-related interference, thus simulating a real interference scenario. For example, assume that the noise vector sampled from the second noise codebook... Adding this to the original accelerometer and gyroscope data (first-level inertial data) yields position disturbance data coupled with random noise, resulting in the following expression:

[0098]

[0099] In the formula, n noise Represents a noise vector; These represent the noise components of the accelerometer in the X, Y, and Z axes, respectively. These represent the noise components of the gyroscope in the X, Y, and Z axes, respectively; a" x ,a" y ,a" z These represent the disturbance accelerations in the X, Y, and Z axes of the accelerometer data after adding noise, respectively; ω" x ,ω" y ,ω z "These represent the disturbance angular velocities of the gyroscope data in the X, Y, and Z axes after noise has been added; [·]" T This indicates the transpose operation.

[0100] In some embodiments, the second noise codebook is used to simulate noise interference that the target device may experience at different locations in a real-world scenario. It contains base noise vectors for multiple locations, enabling the injection of location-related noise interference into the original data during the data augmentation process. By adding noise, the diversity and robustness of the data are enhanced. Further, as an optional implementation, the second noise codebook can be constructed through the following steps:

[0101] First, let the set of labels for the target location be... For each target location L i When collecting noise samples, there are N for each target location. i There are noise samples, denoted as . Where i = 1, 2, ..., M. For each noise sample Six-axis signals including accelerometer and gyroscope data, where i = 1, 2, ..., N i , recorded as

[0102] Then, from the target position L i A static segment is extracted from the original data (i.e., the first inertial data), denoted as... This signal is primarily used to reflect sensor noise and environmental interference. For motion data at the target location... via the cutoff frequency f c A 0.1Hz high-pass filter (HPF) removes low-frequency motion components (motion trend terms) while retaining high-frequency noise signals. Then we have the following expression:

[0103]

[0104] Next, wavelet packet decomposition (db4 wavelet, 5-level decomposition) is performed on the noise signal of each sensor axis to obtain the energy of 32 sub-bands for each sensor axis. The calculation formula is as follows:

[0105]

[0106] Where h[·] are low-pass filter coefficients and g[·] are high-pass filter coefficients.

[0107] For each subband j obtained above, calculate its energy value. Then, the energies of all subbands are summed to obtain the total noise intensity of a single axis. The calculation formula is as follows:

[0108]

[0109] For the same target location L i All noise samples N i Randomly divided into K i There are 1 group, each group contains 1 group. One sample, This indicates rounding down to the nearest integer. For each group... Calculate the mean noise intensity for each axis:

[0110]

[0111] Each group can generate a 6-dimensional codeword vector v k ,

[0112] By integrating all the codewords at the target positions and storing them according to the labels at the target positions, we can obtain the second noise codebook, NosieCodebook:

[0113]

[0114] Based on the target location label L t Randomly select a codeword from the corresponding codeword set. Noise intensity for each axis Generate Gaussian noise.

[0115] Similarly, a 6-dimensional noise vector can be obtained:

[0116]

[0117] In step S250 of some embodiments, second inertial data can be obtained based on multi-rate motion data, second frequency, orientation state, and position disturbance.

[0118] By performing multi-dimensional data augmentation on the raw inertial data, the diversity and richness of the data can be expanded, thus making it more closely resemble actual motion scenarios.

[0119] In step S300 of some embodiments, during the system offline phase, a noise feature codebook (i.e., a first noise codebook) is constructed for the target location (such as a backpack, handheld device, etc.) that affects the abnormal fluctuations of the inertial data sequence during the acquisition process. This is to enable accurate dynamic retrieval, matching, and fusion of noise patterns during subsequent target conversion. The first noise codebook is constructed based on the enhanced inertial data and can be used for noise feature matching when generating inertial data of the target location in the future.

[0120] In some embodiments, step S300 may include, but is not limited to, steps S310 to S330:

[0121] Step S310: Perform wavelet packet decomposition on the second inertial data to obtain the frequency band energy distribution matrix;

[0122] Step S320: Filter the frequency band energy distribution matrix to obtain high-discrimination features;

[0123] Step S330: Classify the high-discrimination features using a clustering algorithm to generate a position-adaptive first noise codebook.

[0124] In step S310 of some embodiments, wavelet packet decomposition is performed based on the enhanced data, i.e., the second inertial data, to extract multi-level frequency band energy distribution feature matrices and quantize noise spectral characteristics. Optionally, the db4 (Daubechies4) wavelet is used to achieve 5-level wavelet packet decomposition. For example, the original signal x(n) of the second inertial data is used as the low-frequency component c of the 0th level. 0,0 [n], i.e., c 0,0 [n] = x(n). For the j-th subband of the i-th layer, calculate the subband coefficient:

[0125]

[0126] Where h[·] are low-pass filter coefficients; g[·] are high-pass filter coefficients; m represents the subband index, used to distinguish different subbands (low frequency or high frequency). In multi-layer wavelet packet decomposition, m is used to represent the number of the current subband, for example, 2m represents the low-frequency subband, and 2m+1 represents the high-frequency subband; k is the subband coefficient index, used to represent the position of the subband coefficients; n is the time index, used to traverse c. i-1,0 All signal sample points of [n].

[0127] In some embodiments, for the j-th subband of the i-th layer, the following subband property holds: if j is even, the subband coefficient c i,j The subband coefficients represent the low-frequency properties; if j is odd, the subband coefficients c i,j Subband coefficients representing high-frequency properties.

[0128] For the i-th layer and j-th subband c i,j [k], the subband energy E i,j The calculation formula is:

[0129]

[0130] Where, N i,j This represents the length of the coefficient sequence of the j-th subband in the i-th layer. For a signal set containing N samples, the frequency band energy distribution feature matrix E is an N×M matrix that reflects the energy distribution characteristics of the original signal in different scale frequency bands.

[0131] In steps S320 to S330 of some embodiments, a noise dictionary is learned using sparse coding, high-discrimination features are selected, and a clustering algorithm is used to form a first noise codebook with positional discrimination. For example, for the obtained frequency band energy distribution feature matrix E, Where K is the number of samples and M is the number of frequency bands, the initial noise dictionary can be constructed using random initialization or methods such as K-SVD. And optimize using sparse coding. Then for each sample E i Find the sparse coding coefficients c i Make E i ≈Dc i That is, finding a suitable initial noise dictionary D and sparse coding coefficients c. i This makes the reconstructed data features Dc i Approximate the original frequency band distribution characteristics E as closely as possible i While maintaining sparsity. Optionally, using L1 regularization to optimize the problem, the objective function for sparse coding optimization is as follows:

[0132]

[0133] Where λ is the regularization parameter. Based on the sparse coding coefficients c... i Update the noise dictionary D, iteratively perform sparse coding optimization and noise dictionary update, and adjust relevant parameters through multiple experiments to balance the accuracy of the noise dictionary and the sparsity of the sparse coding coefficients, thereby improving the quality of the noise codebook, until the convergence condition is met or the maximum number of iterations is reached, and the final noise dictionary and sparse coding coefficients can be obtained.

[0134] After sparse coding is completed, the final sparse coefficients contain feature representations of the noisy data. By analyzing these sparse coefficients, feature selection methods (such as mutual information, analysis of variance, etc.) can be used to filter out features with high discriminative power. Then, K-means clustering is used to classify the high-discriminative features, generating a codeword set {v1,...,v...}. m The first noise codebook with positional discrimination, wherein each codeword corresponds to a noise pattern.

[0135] In steps S400 to S600 of some embodiments, during the system online phase, the motion semantic features and position noise features of the data-enhanced source data (i.e., the second inertial data) are decoupled by a feature decoupling generator, and combined with the noise information of the target position, to generate target inertial data that conforms to the characteristics of the target position.

[0136] In some embodiments, step S400 may include, but is not limited to, steps S410 to S420:

[0137] Step S410: Through the motion branch of the decoupling generator, a deep residual network is used to extract the second inertial data to obtain the motion semantic features;

[0138] Step S420: Through the noise branch of the decoupling generator, a bidirectional long short-term memory network is used to extract the second inertial data to obtain the position noise features.

[0139] In steps S410 to S420 of some embodiments, a feature decoupling generator and a dual-branch encoder extract motion semantic features and position noise features respectively. Optionally, in the motion branch, a deep residual network captures core motion features such as stride frequency and stride length; in the noise branch, a bidirectional long short-term memory network extracts the temporal dependency of position-related noise. For example, the source data x of the second inertial data obtained after data augmentation... source Given an encoder with a dual-branch structure, extracting its motion semantic features and positional noise features yields the following expression:

[0140] h motion =ResNet(x source )

[0141] h noise =BiLSTM(x source )

[0142] Among them, h motion h represents the semantic feature vector of motion. noise This represents the location noise feature vector.

[0143] In some embodiments, step S500 may include, but is not limited to, steps S510 to S530:

[0144] Step S510: Obtain each first codeword of the first noise codebook;

[0145] Step S520: Obtain the similarity between each of the first codewords and the position noise features;

[0146] Step S530: Select the first codeword corresponding to the highest similarity as the target noise codeword.

[0147] In steps S510 to S530 of some embodiments, based on the target location, relevant noise patterns are dynamically queried from the constructed first noise codebook (each codeword corresponds to one noise pattern), and the optimal matching codeword is selected as the target noise codeword. For example, each first codeword v of the first noise codebook is obtained. i (i = 1, ..., m; m is the codebook size), and obtain the similarity s between each first codeword and the positional noise feature. i Then we have the following expression:

[0148]

[0149] Next, the system selects the codeword with the highest similarity as the target noise codeword v. target That is, v target =v q , where q = arg max i s i argmax is the value that makes the similarity s i The index q of the maximum value corresponds to v. q It is the target noise codeword.

[0150] In some embodiments, step S600 may include, but is not limited to, steps S610 to S640:

[0151] Step S610: Obtain the first sparse coding coefficients of the motion semantic features through sparse coding;

[0152] Step S620: Obtain the second sparse coding coefficients of the position noise feature based on the target noise codeword;

[0153] Step S630: Convert the first sparse coding coefficients and the second sparse coding coefficients into a gating vector using a weight matrix and an activation function;

[0154] Step S640: Based on the gating vector, fuse the motion semantic features and the position noise features to obtain the target inertial data.

[0155] In steps S610 to S640 of some embodiments, target noise codewords and motion semantic features, as well as position noise features, are dynamically fused through a learnable gating mechanism. For example, by using the motion semantic feature coefficients c... motion and location noise characteristic coefficient c noise Concatenate them into a vector, then multiply by the learnable weight matrix W. g The feature vectors are then processed through an activation function σ (such as sigmoid) to generate a gating vector g. Each element of the gating vector g has a value between 0 and 1, which controls the fusion ratio of the features. Next, the motion semantic feature vector and the position noise feature vector are weighted and fused using the gating vector to finally obtain the fused vector h. out This is used to characterize the target inertial data that matches the target's position characteristics. The calculation formula is as follows:

[0156] g=σ(W g [c motion ;c noise ])

[0157] h out =g⊙h motion +(1-g)⊙h noise

[0158] Where ⊙ denotes element-wise multiplication (Hadamard product), used to adjust the motion semantic feature vector and the position noise feature vector in the final fused vector h. out The percentage.

[0159] In some embodiments, an inertial data position conversion method based on a noise codebook and a multi-physics discriminator further includes: jointly verifying the target inertial data using a multi-physics discriminator to obtain a verification result. Optionally, the multi-physics discriminator verifies the generated data from three aspects: authenticity, position attributes, and physical rationality. For the authenticity discrimination branch, a temporal convolutional network is used to verify the consistency between the temporal distribution of the generated data and the real data of the target position; for the position attribute discrimination branch, a bidirectional gated recurrent unit is used to predict the position label of the generated data, and cross-entropy loss is used to constrain attribute matching; for the physical rationality discrimination branch, velocity integral verification and attitude constraint verification of the kinematic equations are used to ensure that the generated data satisfies the kinematic equations and ensures the rationality of the physical trajectory. Optionally, the kinematic equations for the physical rationality discrimination of the multi-physics discriminator can be changed according to the actual position information representation to adaptively adapt to multiple real-world scenarios.

[0160] For example, after generating pseudo data of the target location in the embodiment of the present invention, in the authenticity determination branch, for the generated pseudo data x fake A three-layer convolutional network (TCN) is used to extract temporal features. Let the output of the l-th layer TCN be... Then we have the following expression:

[0161]

[0162] The probability p of authenticity is output through a fully connected layer (FC). real , where p real If ∈[0,1], then the following expression holds:

[0163]

[0164] In the location attribute discrimination branch, attempt to process the generated pseudo-data x. fake Location label prediction results are obtained using a bidirectional GRU and a classifier Softmax. Then we have the following expression:

[0165] h GRU =BiGRU(x fake )

[0166]

[0167] Among them, h GRU This indicates the output of the bidirectional GRU.

[0168] In the physical plausibility judgment branch, the data acquisition time interval Δt and the generated pseudo data x are combined. fake Verification of velocity integration and attitude constraints is performed as follows:

[0169]

[0170] v t gen =v t-1 gen +a t gen Δt

[0171] θ t gen =θ t-1 gen +ω t gen Δt

[0172] in, The loss is due to the velocity integral; The loss is for attitude constraints; v t gen The generated pseudo-acceleration at the current moment; v t-1 gen The velocity at the previous moment; a t gen The generated pseudo-acceleration at the current moment; θ t gen θ is the pseudo-deflection angle generated at the current moment. t-1 gen ω is the deflection angle at the previous moment; t gen This is the pseudo-angular acceleration generated at the current moment; The reference attitude value is Δt; the time interval is Δt; and the length of the time series is T. Next, check x. fake Does the phone placement location label match the target location label y? target Consistent.

[0173] In some embodiments, an inertial data position conversion method based on a noise codebook and a multi-physics discriminator further includes: pre-training the first noise codebook and training the decoupling generator and the multi-physics discriminator. Exemplarily, it may include the following steps:

[0174] Step 1: Set the hyperparameters of the conversion network in the target device, where the conversion network is used for inertial data position conversion.

[0175] Step 2: During the offline training phase of the system, after obtaining the initial first noise codebook, unsupervised contrastive learning is used to further optimize the codebook and maximize the similarity of features of the same type of noise. In some optional embodiments, the following processing steps are further implemented:

[0176] Let the noise feature vector extracted from different data samples at the same location be... and (Positive sample pair), the noise feature vector extracted from data samples at different locations is and (Subsample pair)

[0177]

[0178] Where sim(·) calculates the similarity between feature vectors and is optimized using the InfoNCE loss function; N is the number of positive samples in a mini-batch; and M is the amount of data in the sub-samples. yes Positive samples (noise feature vectors from different data samples at the same location); Other positive samples τ is the negative sample (a noisy feature vector from data samples at different locations); τ is the temperature hyperparameter used to adjust the similarity score.

[0179] By minimizing noise feature vectors from the same location to be closer in the feature space, and noise feature vectors from different locations to be further apart, the noise codebook is optimized to better represent noise patterns at different locations.

[0180] Step 3: During the online training phase of the system, initialize all parameters of the generator and discriminator, and train them alternately. Further, in some optional embodiments, fix the parameters of the discriminator, input source location inertial data and target location labels into the generator for training, and generate target location data. The total loss function is:

[0181]

[0182]

[0183] Among them, combating losses The aim is to enable the data generated by the generator to deceive the discriminator; location classification loss. It is cross-entropy loss, used to constrain the location attributes of the generated data to be consistent with the target location; noise pattern matching loss. By calculating the L1 distance of the wavelet packet energy distribution characteristics (Ψ) between the generated data and the real data at the target location, it is ensured that the noise pattern of the generated data is similar to the pattern at the target location; physical constraint loss. Used to constrain the irrationality of velocity abrupt changes, attitude, and trajectory in the generated data; among them, Statistical distribution of real data from the target velocity.

[0184] In some alternative embodiments, an optimizer is used to update the generator's parameters based on the generator's loss. With the generator's parameters fixed, real data and pseudo-data generated by the generator are input into the discriminator, and the discriminator's loss is calculated to update the parameters.

[0185] By introducing course learning for training, the initial stage focuses on adversarial loss, allowing the generator and discriminator to undergo preliminary adversarial training. This enables the generator to quickly learn the basic distribution and characteristics of the data, giving the generated data an initial distribution similar to real data. As training progresses, location classification loss, noise pattern matching loss, and physical rule verification loss are gradually added, making the model more attentive to the authenticity, noise patterns, and physical plausibility of the generated data, thereby improving the quality of the model-generated data.

[0186] In some embodiments, performance is evaluated using quantitative metrics such as noise pattern, trajectory drift rate, position spoofing success rate, and trajectory error. The similarity of the noise pattern is measured using the Bhattacharyya coefficient of the wavelet packet energy distribution; the trajectory drift rate refers to the rate of increase in displacement error after pre-integration processing of the generated data; the position spoofing success rate refers to the misclassification rate of the generated data by the pre-trained position classifier; and the trajectory error refers to the absolute trajectory error between the trajectory calculated by the generated data using an inertial navigation algorithm and the actual path.

[0187] In summary, as Figure 3 As shown, the overall steps of the inertial data position conversion processing procedure based on noise codebook and multi-physics discriminator in this embodiment of the invention are as follows:

[0188] Step 1: The system collects six-degree-of-freedom timing data of acceleration and gyroscope through the inertial measurement unit (IMU) of the mobile device, and records the device placement location label (such as handheld or backpack).

[0189] Step 2: Perform multi-data augmentation on the raw data collected above to expand the diversity and richness of the data and make it more relevant to the actual motion scenario.

[0190] Step 3: During the offline phase of the system, for the target location to be converted (such as a backpack, handheld device, etc.), a noise feature codebook is constructed to identify the abnormal fluctuations in the inertial data sequence during the acquisition process. This codebook is used to accurately retrieve, match, and fuse noise patterns during subsequent target conversion. During the online phase of the system, for the data-enhanced source data, the feature decoupling generator decouples the motion semantic features and position noise features, and combines them with the noise information of the target location to generate inertial data that conforms to the characteristics of the target location.

[0191] Step 4: Verify the generated data from three aspects: authenticity, location attributes, and physical plausibility, using a multi-physics discriminator.

[0192] Step 5: Train the device position transformation network using device location labels and raw inertial sensor data, including:

[0193] Step 5.1: Configure the hyperparameters for the device's network conversion;

[0194] Step 5.2: During the offline training phase of the system, after obtaining the initial noise codebook, unsupervised contrastive learning is used to further optimize the codebook and maximize the similarity of noise features of the same type.

[0195] Step 5.3: During the online training phase of the system, initialize all parameters of the generator and discriminator, and train them alternately.

[0196] This invention also provides an inertial data position conversion system based on a noise codebook and a multi-physics discriminator, which can realize the above-mentioned inertial data position conversion method based on a noise codebook and a multi-physics discriminator. The system includes:

[0197] The first module is used to acquire the target position and initial inertial data of the target device;

[0198] The second module is used to perform multi-dimensional data augmentation processing on the first inertial data to obtain the second inertial data;

[0199] The third module is used to construct a position-adaptive first noise codebook based on the second inertial data;

[0200] The fourth module is used to extract motion semantic features and position noise features from the second inertial data through a decoupling generator.

[0201] The fifth module is used to obtain the target noise codeword based on the first noise codebook and the position noise features;

[0202] The sixth module is used to obtain the target inertial data of the target position through a gating fusion mechanism based on the target noise codeword, the motion semantic features and the position noise features.

[0203] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0204] This invention also provides an electronic device, which includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned inertial data position conversion method based on a noise codebook and a multi-physics discriminator. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0205] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0206] refer to Figure 4 , Figure 4 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0207] The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0208] The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called and executed by the processor 701 to execute the inertial data position conversion method based on noise codebook and multi-physics discriminator of the present invention.

[0209] The input / output interface 703 is used to implement information input and output;

[0210] The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0211] Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704);

[0212] The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0213] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned inertial data position conversion method based on a noise codebook and a multi-physics discriminator.

[0214] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0215] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned inertial data position conversion method based on a noise codebook and a multi-physics discriminator.

[0216] In summary, the inertial data position conversion method and system based on a noise codebook and a multi-physics discriminator, as described in this invention, addresses the problem of noise feature migration in inertial sensor data generated by a target device under different placement locations. It constructs a noise codebook for dynamic noise pattern modeling and employs techniques such as decoupling motion semantics from position noise and multi-physics constraint verification to achieve feature conversion of device position-related noise and ensure the physical compliance of generated data, significantly improving the robustness and generalization ability of indoor positioning systems. Specifically, the inertial data position conversion method and system based on a noise codebook and a multi-physics discriminator of this invention has the following advantages:

[0217] 1. This invention introduces wavelet packet decomposition and sparse coding techniques to construct a noise codebook, which can accurately quantify the nonlinear mapping relationship between device placement and noise spectrum. This allows for the dynamic capture of complex noise patterns under different placement positions, such as handheld or backpack placement, improving the accuracy and flexibility of noise modeling. Compared with traditional methods that rely on a single noise model (such as Gaussian white noise) or static feature matching, this invention improves the ability to capture dynamically coupled noise patterns and enhances the quality of inertial data generated for real-world scenarios.

[0218] 2. This invention employs a dual-branch network structure and a gating fusion mechanism, which enables fine-grained decoupling of motion semantics and noise features. This effectively avoids overfitting or underfitting of noise, improving the effectiveness of data feature extraction and fusion. Compared to traditional methods that do not explicitly separate motion semantics and noise features or use static signal processing procedures, this method improves the processing capability of motion and noise features, playing a crucial role in achieving more efficient and accurate inertial data position conversion.

[0219] 3. This invention employs a multi-physics discriminator to jointly verify data from three aspects: authenticity, location attributes, and physical plausibility. This rigorously constrains the generated data to conform to kinematic and dynamic laws, thereby avoiding problems such as sudden velocity changes and attitude discontinuities in the generated trajectory, and improving the quality and reliability of the generated data. Compared with traditional methods lacking explicit kinematic / dynamic verification, this improves the usability of the generated data in practical navigation algorithms, which is of crucial significance for applications such as high-precision indoor positioning.

[0220] 4. This invention introduces a course-based learning strategy and an unsupervised comparative learning method, which can optimize the model training process in stages. First, it optimizes the adversarial loss, then gradually adds physical constraints, while reducing dependence on labeled data. This improves the stability and generalization ability of model convergence, and enhances the efficiency and effectiveness of model training. Compared with traditional training methods, this improves the scientific rigor and rationality of model training, greatly contributing to the development of higher-quality inertial data position conversion models.

[0221] 5. This invention employs Bhattacharyya coefficients to quantify noise pattern similarity and calculates evaluation metrics such as trajectory drift rate through pre-integration, thereby providing a more reliable basis for model optimization and improvement, and enhancing the accuracy and relevance of model evaluation. Compared with traditional evaluation metrics, it improves the ability to evaluate the model's performance in inertial data processing, playing a crucial role in achieving more effective model performance verification.

[0222] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0223] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0224] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0225] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0226] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0227] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0228] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0229] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0230] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for inertial data position conversion based on a noise codebook and a multi-physics discriminator, characterized in that, Includes the following steps: Acquire the target position and initial inertial data of the target device; The first inertial data is subjected to multi-dimensional data augmentation processing to obtain the second inertial data; Based on the second inertial data, construct a position-adaptive first noise codebook; The second inertial data is extracted using a decoupling generator to obtain motion semantic features and position noise features. Based on the first noise codebook and the location noise features, the target noise codeword is obtained; Based on the target noise codeword, the motion semantic features, and the position noise features, the target inertial data of the target position is obtained through a gating fusion mechanism.

2. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, Also includes: The target inertial data is jointly verified using a multi-physics discriminator to obtain the verification results; The first noise codebook is pre-trained, and the decoupling generator and the multi-physics discriminator are trained.

3. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, The process of performing multi-dimensional data augmentation on the first inertial data to obtain the second inertial data includes the following steps: The first acceleration in the first inertial data is subjected to time-domain scaling to generate multi-rate motion data; A linear frequency modulation perturbation is applied to the first frequency in the first inertial data to obtain the second frequency; A random rotation matrix is ​​generated in a three-dimensional rotation group space to simulate the orientation state of the target device. The noise vector at the target location is randomly sampled and superimposed onto the first inertial data to simulate position interference. The second inertial data is obtained based on the multi-rate motion data, the second frequency, the orientation state, and the position disturbance.

4. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, The step of constructing a position-adaptive first noise codebook based on the second inertial data includes the following steps: Wavelet packet decomposition is performed on the second inertial data to obtain the frequency band energy distribution matrix; The frequency band energy distribution matrix is ​​filtered to obtain high-discrimination features; The high-discrimination features are classified using a clustering algorithm to generate a first noise codebook with adaptive position.

5. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, The step of extracting motion semantic features and position noise features from the second inertial data using a decoupling generator includes the following steps: The motion semantic features are obtained by extracting the second inertial data through the motion branch of the decoupling generator using a deep residual network. The location noise features are obtained by extracting the second inertial data through the noise branch of the decoupling generator using a bidirectional long short-term memory network.

6. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, The step of obtaining the target noise codeword based on the first noise codebook and the positional noise features includes the following steps: Obtain each first codeword of the first noise codebook; Obtain the similarity between each of the first codewords and the positional noise features; The first codeword corresponding to the highest similarity is selected as the target noise codeword.

7. The inertial data position conversion method based on a noise codebook and a multi-physics discriminator according to claim 1, characterized in that, The step of obtaining target inertial data of the target position based on the target noise codeword, the motion semantic features, and the position noise features through a gating fusion mechanism includes the following steps: The first sparse coding coefficients of the motion semantic features are obtained through sparse coding; Based on the target noise codeword, obtain the second sparse coding coefficients of the location noise feature; The first sparse coding coefficients and the second sparse coding coefficients are converted into a gating vector using a weight matrix and an activation function. The motion semantic features and the position noise features are fused based on the gating vector to obtain the target inertial data.

8. An inertial data position conversion system based on a noise codebook and a multi-physics discriminator, characterized in that, include: The first module is used to acquire the target position and initial inertial data of the target device; The second module is used to perform multi-dimensional data augmentation processing on the first inertial data to obtain the second inertial data; The third module is used to construct a position-adaptive first noise codebook based on the second inertial data; The fourth module is used to extract motion semantic features and position noise features from the second inertial data through a decoupling generator. The fifth module is used to obtain the target noise codeword based on the first noise codebook and the position noise features; The sixth module is used to obtain the target inertial data of the target position through a gating fusion mechanism based on the target noise codeword, the motion semantic features and the position noise features.

9. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 7.

Images