CTG neural network aided navigation method and device in GNSS weak signal scene

Abstract

The invention provides a CTG neural network aided navigation method in a GNSS weak signal scene, which comprises the following steps: synchronously acquiring inertial measurement unit IMU original data and GNSS satellite positioning data of a mobile carrier in a GNSS signal stable scene, and constructing a spatial-temporal characteristic data set containing dynamic acceleration, angular velocity and position increment; constructing a CTG network based on the data set, wherein the CTG network comprises an input layer, a feature extraction layer, a coding layer and a decoding layer which are connected in sequence; performing off-line training on the CTG network by using the spatio-temporal feature data set to obtain an optimal network weight and a model structure; iMU data are acquired in real time in a GNSS weak signal scene, and motion state implicit features are extracted through the trained CTG network and pseudo position measurement information is generated; and carrying out fusion positioning on the pseudo position measurement information and inertial navigation system (INS) data through a Kalman filter, dynamically correcting an accumulated error of inertial navigation, and outputting a continuous and smooth positioning result.

Description

Technical Field

[0001] The present invention relates to the field of navigation and positioning technology, and in particular to a CTG neural network assisted navigation method and system in a GNSS weak signal scenario. Background Art

[0002] The Global Navigation Satellite System (GNSS) provides crucial data for positioning, navigation, and timing. However, in weak signal environments such as urban canyons, indoor environments, or those severely obscured by obstructions, satellite signals can be significantly weakened by attenuation, multipath effects, and interference, resulting in reduced positioning accuracy or even failure. Furthermore, while the Inertial Navigation System (INS) can continuously provide position information even when the GNSS signal is lost, the accumulated errors and drift caused by its double-integration calculations limit the ability to achieve high-precision navigation over long periods of time. Traditional GNSS / INS fusion methods primarily rely on Kalman filtering for data fusion and error compensation. However, in weak signal scenarios, the filter struggles to accurately capture nonlinear error characteristics, resulting in unstable positioning results. Furthermore, the development of deep learning technology has made it possible to compensate for GNSS measurement deficiencies in weak signal scenarios by deeply mining the spatiotemporal features of sensor data. Therefore, improving navigation performance in weak signal scenarios using advanced feature extraction and time series modeling techniques has become a research hotspot and a challenging issue in current positioning and navigation technologies.

[0003] Currently, INS-GNSS fusion positioning technology has been widely adopted in the navigation field. Its basic principle is to use an extended Kalman filter to fuse GNSS measurement data with IMU output data to achieve real-time positioning. However, in weak GNSS signal scenarios, due to severe signal attenuation, multipath effects, and interference, GNSS data often contains significant noise and uncertainty, making it difficult for the filter to obtain accurate measurement updates, resulting in a gradual amplification of INS cumulative errors. Furthermore, while some anomaly detection and compensation methods based on rules or deep learning models can improve positioning accuracy to a certain extent, their ability to describe nonlinear motion characteristics in weak signal scenarios is limited, making it difficult to achieve precise error compensation. Furthermore, some positioning solutions that utilize other sensors such as vision and radar are limited by environmental factors (such as lighting and occlusion) and high computational resource requirements, making it difficult to ensure real-time performance and stability. Overall, existing technologies in weak signal scenarios suffer from shortcomings such as insufficient data reliability, poor model adaptability, and high computational complexity.

[0004] Therefore, there is an urgent need for an innovative technical solution that can fully exploit multi-source spatiotemporal information to dynamically correct INS accumulated errors in GNSS weak signal scenarios. Summary of the Invention

[0005] The present invention provides a CTG neural network assisted navigation method and system in GNSS weak signal scenarios, which is used to overcome the problem that the existing technology fails to fully mine multi-source spatiotemporal information, resulting in unsatisfactory positioning effects in GNSS weak signal scenarios.

[0006] To achieve the above objectives, the present invention provides a CTG neural network assisted navigation method in a GNSS weak signal scenario, comprising the following steps: In a scenario where the GNSS signal is stable, the mobile carrier's inertial measurement unit (IMU) raw data and GNSS satellite positioning data are synchronously collected to construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity, and position increments. Constructing a CTG network based on the data set, wherein the CTG network includes an input layer, a feature extraction layer, an encoding layer, and a decoding layer connected in sequence; Using the spatiotemporal feature dataset to perform offline training on the CTG network to obtain the optimal network weights and model structure; Acquire IMU data in real time in weak GNSS signal scenarios, extract motion state implicit features through the trained CTG network, and generate pseudo position measurement information; The pseudo position measurement information is fused with the inertial navigation system INS data through the Kalman filter for positioning, the accumulated error of the inertial navigation is dynamically corrected, and a continuous and smooth positioning result is output.

[0007] Moreover, the GNSS signal stable scene is an open space scene with good satellite signals.

[0008] Moreover, when constructing the spatiotemporal feature dataset, the input features are set as the three-axis specific force and three-axis angular velocity of the IMU, and the output features are set as the true value of the position increment output by the GNSS, where the sampling frequencies of the GNSS and IMU are set synchronously.

[0009] Moreover, in the CTG network, the feature extraction layer is implemented based on a one-dimensional convolutional neural network to extract local spatiotemporal features.

[0010] Moreover, the encoding layer is improved based on the Transformer module and adopts a lightweight attention mechanism.

[0011] Moreover, the lightweight attention mechanism is implemented by approximating the dot product as the inner product of the high-dimensional mapping of the two features.

[0012] Moreover, the decoding layer includes a GRU module and a fully connected layer for capturing temporal dynamic features.

[0013] On the other hand, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the program, the CTG neural network assisted navigation method in the GNSS weak signal scenario as described above is implemented.

[0014] On the other hand, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the CTG neural network-assisted navigation method in the GNSS weak signal scenario as described above.

[0015] On the other hand, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the CTG neural network assisted navigation method in the GNSS weak signal scenario as described above.

[0016] The present invention provides a CTG neural network assisted navigation technology solution in GNSS weak signal scenarios, which fully integrates the spatiotemporal features in IMU and satellite positioning data, realizes the deep extraction of implicit information of motion state, and then generates high-quality pseudo-position measurement information. Compared with the traditional solution that relies solely on extended Kalman filtering for INS / GNSS data fusion, the existing technology is easily affected by multipath effects, uneven noise distribution and INS cumulative errors in weak signal scenarios, resulting in reduced positioning accuracy and system instability. The present invention takes advantage of deep learning adaptive modeling and fully mines multi-source spatiotemporal information to dynamically correct INS cumulative errors in GNSS weak signal scenarios, thereby significantly improving positioning continuity, smoothness and robustness, and meeting the stringent requirements for navigation accuracy and stability in weak signal scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 2 is a flow chart of a CTG neural network assisted navigation method in a GNSS weak signal scenario according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of the CTG neural network in an embodiment of the present invention; Figure 3 This is a schematic diagram of the CTG neural network assisted navigation principle embodiment of the present invention; Figure 42 is a schematic diagram of the structure of a CTG neural network assisted navigation system in a GNSS weak signal scenario according to an embodiment of the present invention; Figure 5 This is an internal structure diagram of a computer device according to an embodiment of the present invention; Figure 6 1 is a schematic diagram of the position prediction result and its error sequence in a GNSS weak signal scenario according to an embodiment of the present invention; Figure 7 2 is a schematic diagram of the combined navigation results and their error sequence in a GNSS weak signal scenario according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0020] The present invention provides a CTG neural network-assisted navigation technology in GNSS weak signal scenarios. In an open environment with good and stable satellite signals, the raw data of the inertial measurement unit (IMU) of a mobile carrier and GNSS satellite positioning data are synchronously collected to construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity and position increment. A CTG (CNN-Transformer-GRU) network is constructed based on the dataset, and the CTG network is trained offline using the dataset to obtain optimal network weights and model structure. In GNSS weak signal scenarios, IMU data is acquired in real time, and the CTG network is used to extract implicit features of the motion state and generate pseudo-position measurement information. The information is then fused with INS data through a Kalman filter for positioning, and the accumulated error of inertial navigation is dynamically corrected to obtain smooth and continuous positioning results.

[0021] In one embodiment, Figure 1 As shown, the present invention provides a CTG neural network assisted navigation method in a GNSS weak signal scenario, comprising the following steps: S100, in an open environment with good and stable satellite signals, synchronously collects the mobile carrier's inertial measurement unit (IMU) raw data and satellite positioning data; In an embodiment, in an open space with good and stable satellite signals, the IMU raw data and GNSS positioning data of the mobile carrier are synchronously collected by multi-source sensors; wherein the open space with good and stable satellite signals can be set according to actual application requirements, such as a playground; After determining an open scene with good and stable satellite signals, a mobile terminal capable of acquiring GNSS / INS data can be used to collect the corresponding multi-constellation GNSS measured data and IMU raw data in the open scene for subsequent modeling and analysis; in order to ensure that the subsequent neural network model has good accuracy and generalization ability, as many sampling carriers as possible in the open scene with different motion state data, including straight driving, turning, acceleration, deceleration, etc., with no restrictions on the specific sampling frequency and number of sampling times. For example, in the above-mentioned playground scene, GNSS data and IMU data can be synchronously sampled at a sampling frequency of 100Hz. The GNSS data features include the true value of the three-dimensional position of latitude, longitude and altitude, and the IMU data include the three-axis specific force and three-axis angular velocity output by the accelerometer and gyroscope of the IMU.

[0022] The preferred specific implementation scheme of this step in the embodiment is: Collect and extract GNSS and IMU data features in open areas where satellite signals are stable; the GNSS data features include the three-dimensional position output by the GNSS The IMU data features include the three-axis specific force and three-axis angular velocity output by the accelerometer and gyroscope in one epoch. .

[0023] To ensure time synchronization, the sampling frequencies of GNSS and IMU are both set to 100 Hz.

[0024] S200: Construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity and position increment: use the three-axis acceleration and three-axis angular velocity collected by the imu at the same time as input features, and the position increment of the GNSS positioning result as output features to obtain a spatiotemporal feature dataset, which is divided into a training set and a validation set according to a certain ratio.

[0025] In the embodiment, an input and output feature data set is constructed, wherein the input features are the three-axis acceleration and three-axis angular velocity output by the accelerometer and gyroscope of the imu , the output feature is the true value of the position increment .

[0026] S300: Construct a CTG (CNN-Transformer-GRU) network based on the dataset. The CTG neural network includes an input layer, a feature extraction layer, an encoding layer, and a decoding layer connected in sequence; the input data of the input layer is the three-axis acceleration and three-axis angular velocity collected by the IMU, the feature extraction layer includes a one-dimensional convolutional neural network (1D-CNN) and a maximum pooling layer, the encoding layer is implemented based on the multi-head attention mechanism of Transformer, the decoding layer includes a GRU and a fully connected layer, and finally outputs the three-dimensional position increment at that moment. The number of feature extraction layers, encoding layers, and decoding layers are all set according to the training difficulty of the spatiotemporal dataset, such as Figure 2FIG2 is a schematic diagram of the structure of a CTG network in an embodiment of the present invention; CNN module design: Use convolutional neural network (CNN) to extract local spatial features. Assume that the output of layer L is , then:

[0027] in, For input data, and Respectively The convolution kernel and bias of the layer;" " represents the convolution operation; is the ReLU activation function.

[0028] Transformer module design: Based on the local features extracted by CNN, a Transformer module is constructed to capture long-range dependencies and global context information. For any similarity function, the following general attention equation can be used to express it:

[0029] in expresses the attention output at position 𝑖, represents the query vector at the i-th position, , Refers to the key and value vector of the jth position. In the traditional attention mechanism, , so the attention weight can be regarded as the weight distribution after the dot product is smoothly normalized in the input space. And because the dot product satisfies Mercer's theorem, the dot product can be approximated as the inner product of the high-dimensional mapping of the two features by using the kernel technique. That is . At this time, the attention mechanism can be expressed as follows:

[0030] At this time, for an input sequence length of n, the computational complexity of attention is , when n>>d, its computational complexity is , while the computational complexity of the traditional transformer is still Obviously, the kernel technique can significantly reduce the computational complexity at the expense of a certain expression.

[0031] Since the similarity between two features must be positive, for the kernel function The structure of must have non-negativity. Therefore, the present invention can Constructed into a linear rectification function This is because (x)+1 only involves simple comparison and addition, which is hardware-friendly and has low latency. There are only two linear regions, and the derivation is extremely simple, which helps compiler optimization and faster gradient updates.

[0032] Will Constructed into The essence of mapping to a high-dimensional space for inner product is very similar to the attention calculation of the traditional Transformer. Although it does not have the nonlinear amplification characteristics of the exponential function, it can still capture the general trend of similarity. Finally, the outputs of all attention heads are spliced together through the matrix Get the final output of the multi-head self-attention layer :

[0033] The feedforward layer then introduces nonlinear transformations to the output of the multi-head attention, thereby capturing more complex features and patterns. Finally, residual connections and layer normalization operations are performed to enhance the stability and performance of the model.

[0034] GRU module design: The output sequence of the Transformer module is further input into the Gated Recurrent Unit (GRU) module to capture temporal continuity and dynamic features. The core computation steps of the GRU include:

[0035]

[0036]

[0037]

[0038] in,

[0039] Represents the input features of the current time step; To update the gate, ⋅ is the hyperbolic tangent function, ; is the hidden state at the current moment. This module is used to further capture the temporal dynamics of motion features.

[0040] S400: Use the dataset to perform offline training on the CTG network to obtain the optimal network weights and model structure.

[0041] In this embodiment, the output of the GRU module is mapped to the predicted value of the pseudo position measurement information through the fully connected layer The actual position increment is The CTG network is trained using RMSE as the loss function and Adam as the optimizer. An early stopping mechanism is used to preset an iteration termination condition to obtain the optimal network weights and model structure. The preset iteration termination condition is preferably set to terminate when the validation set accuracy does not improve for 10 consecutive rounds. The difference between the training accuracy and the validation accuracy is within a preset range. The training accuracy and validation accuracy are obtained as follows:

[0042] in is the number of samples. This loss function is used to measure the gap between the model output and the true measurement value, and optimize the network parameters through the back propagation algorithm.

[0043] During model training, CTG-net was trained offline using the dataset constructed above. During training, the weights of the CNN, Transformer, and GRU modules were updated using the Adam optimization algorithm until the loss function converged, achieving the optimal network weights and model structure.

[0044] The parameter settings during the CTG network training process in this embodiment are shown in Table 1. The corresponding preset iteration termination condition can be understood as that the trained CTG network meets the application requirements, that is, the higher the accuracy on the training set and the validation set, the better, and the closer the two are, the better. Correspondingly, the preset accuracy standard and the preset range can be determined according to actual application requirements and are not specifically limited here. Table 1 Transformer-LSTM parameter settings

[0045] S500: Acquire IMU data in real time in a weak GNSS signal scenario, use the CTG network to extract implicit features of the motion state and generate pseudo position measurement information.

[0046] In the embodiment, IMU data is collected in real time and preprocessed, the CNN module is used to extract local spatiotemporal features, the Transformer module captures global temporal dependencies, the GRU module models time continuity, and finally, the pseudo position measurement information is obtained through full connection layer mapping.

[0047] Detect whether the scenario in which the carrier is located is a weak signal scenario. Specifically, the number of visible satellites and the average carrier-to-noise ratio can be detected by the GNSS receiver to determine: when the number of visible satellites is ≤10 and the average carrier-to-noise ratio is ≤20, it is determined to be a weak signal scenario.

[0048] In a weak signal scenario, IMU data is obtained in real time as the input of the CTG network. The pseudo position increment information is output by the CTG network model trained offline. Combined with the specific position at the previous moment, the pseudo position information at this moment is obtained. Figure 3 FIG2 is a schematic diagram showing the principle of CTG neural network assisted navigation in a weak signal scenario according to an embodiment of the present invention; S600: The steps of fusing positioning with INS data through the Kalman filter to dynamically correct the accumulated error of inertial navigation to obtain smooth and continuous positioning results include: The measurement residual is generated by comparing the pseudo position information output by the CTG network with the corresponding value of the SINS, which drives the Kalman filter to adaptively estimate the high-precision position information of the carrier.

[0049] Dynamic correction of the accumulated error of inertial navigation is achieved, thereby estimating the high-precision position information of the carrier. Figure 4 The figure shows a schematic diagram of the principle of fusion positioning in an embodiment of the present invention.

[0050] It should be noted that although the steps in the above flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0051] The above method fully integrates the spatiotemporal features in IMU and satellite positioning data to achieve deep extraction of implicit information about motion states, thereby generating high-quality pseudo-position measurement information. Compared with the traditional solution that relies solely on extended Kalman filtering for INS / GNSS data fusion, the existing technology is easily affected by multipath effects, uneven noise distribution, and serious effects of INS cumulative errors in weak signal scenarios, resulting in reduced positioning accuracy and system instability. The present invention utilizes the advantages of deep learning adaptive modeling, fully exploits multi-source spatiotemporal information to dynamically correct INS cumulative errors in GNSS weak signal scenarios, thereby significantly improving positioning continuity, smoothness, and robustness, and meeting the stringent requirements for navigation accuracy and stability in weak signal scenarios.

[0052] To facilitate understanding of the technical effects of the present invention, a vehicle-mounted data set collected on the campus of Wuhan University on December 21, 2024, was used for verification and analysis. This data set is 1200 seconds long. The inertial navigation data and GNSS observation data are provided by the MTi-G-710 from Xsens and the AsteRx-m3 Pro dual-antenna board from Septentrio, respectively. The IMU sampling rate is 100Hz, and the AsteRx-m3 Pro dual-antenna board outputs 10Hz observation data for the GPS / BDS / GAL three systems. The base station data is provided by the mosaic-X5 module, which has a data sampling rate of 10Hz. The high-precision SPAN-CPT6 split integrated navigation system was selected as the reference device, and its high-precision navigation parameters were projected onto the reference center of the device under test through the arm as the reference true value.

[0053] The first 600 seconds of the dataset were used to construct a training set for five neural network models. These models were trained on the publicly available PyTorch deep learning framework. To ensure fair comparison, the models were trained using the same number of epochs. The weight matrices and bias values of all networks were initialized within the same range. The network weights were then updated using ADAM with an initial learning rate of 0.0001. 20% of the samples were randomly selected as the validation set, and the remaining 80% were used for parameter training. The final results for each model were the average of metrics obtained from 10 independent experiments (using different random seeds).

[0054] A training dataset was constructed to train the CTG neural network and evaluate its training accuracy. For better evaluation, the latest neural network models were implemented for comparison: 1) TCN model; 2) GNN-GRU model; 3) AT-LSTM model; and 4) CNN-Informer model. The CTG neural network served as 5th model. Five approaches were analyzed throughout the GNSS position prediction phase and the phase of combining with INS to mitigate INS error divergence.

[0055] From the attached Figure 6 As shown in Table 2: The RMS of the north-east-ground position error of the proposed CTG neural network model in the GNSS position prediction stage is (0.78, 0.40, 0.10) m, and the RMS of the plane position error is 1.05 m, which is superior to other schemes with the minimum error value; Table 2 Statistics of position prediction error indicators in GNSS weak signal scenarios

[0056] From the attached Figure 7As shown in Table 3: The RMS of the north-east-ground position error of the proposed CTG neural network model in the integrated navigation stage is (0.68, 0.25, 0.04) m, and the RMS of the plane position error is 0.93 m, which is superior to other schemes with the minimum error value; Table 3 Statistics of integrated navigation position error indicators in GNSS weak signal scenarios

[0057] In one embodiment, Figure 4 As shown, a CTG neural network assisted navigation system in a GNSS weak signal scenario is provided, the system comprising: Dataset Acquisition Module: This module uses an inertial measurement unit (IMU) and a GNSS receiver to synchronously collect raw data from a mobile carrier in an open area with stable satellite signals. The IMU module collects triaxial acceleration and triaxial angular velocity data, while the GNSS receiver collects true position data.

[0058] Dataset construction module: performs time domain alignment processing on the collected raw data to construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity, and position increment, providing a data basis for subsequent network training.

[0059] CTG network building module: The CTG network is built based on the CNN-Transformer-GRU architecture, which specifically includes: the CNN sub-module is responsible for extracting local spatiotemporal features and capturing fine-grained local patterns in the data; the Transformer sub-module uses the self-attention mechanism to capture global dependencies and contextual information, making up for the shortcomings of traditional CNN in long-distance dependency modeling; the GRU sub-module performs temporal modeling on the sequence data output by the Transformer and extracts the dynamic continuity characteristics of the motion state.

[0060] Offline training module: The CTG network is trained offline using the constructed spatiotemporal feature dataset, and the mean square error (MSE) is used in combination with the Adam optimization algorithm to obtain the optimal network weights and model structure.

[0061] Pseudo-position measurement generation module: This module uses a trained CTG network to process real-time IMU data, extracting implicit motion features and mapping these features into pseudo-position measurement information. This module predicts the current position increment to compensate for inertial navigation errors.

[0062] Fusion positioning module: The generated pseudo-position measurement information is fused with the inertial navigation system (INS) data through the Kalman filter for positioning, and the filtering algorithm is used to dynamically correct the INS accumulated error to output a smooth, continuous and highly accurate final positioning result.

[0063] Figure 5 The present invention provides a schematic diagram of the physical structure of an electronic device, which may include: a processor, a communications interface, a memory, and a communications bus, wherein the processor, the communications interface, and the memory communicate with each other via the communications bus. The processor may call logic instructions in the memory to execute a CTG neural network-assisted navigation method in a weak GNSS signal scenario. The method includes: synchronously collecting raw inertial measurement unit (IMU) data and satellite positioning data of a mobile carrier using multiple sensors in an open space with good and stable satellite signals to construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity, and position increments; constructing a CTG network (CNN-Transformer-GRU) based on the dataset, and using the dataset to perform offline training on the CTG network to obtain optimal network weights and model structure; acquiring IMU data in real time in a weak GNSS signal scenario, using the CTG network to extract implicit motion state features and generate pseudo-position measurement information, and fusing the information with INS data through a Kalman filter for positioning, dynamically correcting the accumulated error of inertial navigation, and thus obtaining a smooth and continuous positioning result.

[0064] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion 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 for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0065] In a possible embodiment, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the software processing part of the CTG neural network assisted navigation method in the GNSS weak signal scenario provided by the above methods.

[0066] In a possible embodiment, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the software processing portion of the CTG neural network-assisted navigation method in the GNSS weak signal scenario provided by the above-mentioned methods.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0068] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A CTG neural network assisted navigation method in a weak GNSS signal scenario, characterized by: The following processes are included: In a scenario where the GNSS signal is stable, the mobile carrier's inertial measurement unit (IMU) raw data and GNSS satellite positioning data are synchronously collected to construct a spatiotemporal feature dataset containing dynamic acceleration, angular velocity, and position increments. Constructing a CTG network based on the data set, wherein the CTG network includes an input layer, a feature extraction layer, an encoding layer, and a decoding layer connected in sequence; Using the spatiotemporal feature dataset to perform offline training on the CTG network to obtain the optimal network weights and model structure; Acquire IMU data in real time in weak GNSS signal scenarios, extract motion state implicit features through the trained CTG network, and generate pseudo position measurement information; The pseudo position measurement information is fused with the inertial navigation system INS data through the Kalman filter for positioning, the accumulated error of the inertial navigation is dynamically corrected, and a continuous and smooth positioning result is output.

2. The CTG neural network-assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: The GNSS signal stable scene is an open space scene with good satellite signals.

3. The CTG neural network-assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: When constructing the spatiotemporal feature dataset, the input features are set to the three-axis specific force and three-axis angular velocity of the IMU, and the output features are set to the true value of the position increment output by the GNSS. The sampling frequencies of the GNSS and IMU are set synchronously.

4. The CTG neural network-assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: In the CTG network, the feature extraction layer is implemented based on a one-dimensional convolutional neural network and is used to extract local spatiotemporal features.

5. The CTG neural network assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: The encoding layer is improved based on the Transformer module and adopts a lightweight attention mechanism.

6. The CTG neural network assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: The lightweight attention mechanism is implemented by approximating the dot product as the inner product of the high-dimensional mapping of two features.

7. The CTG neural network assisted navigation method in a weak GNSS signal scenario according to claim 1, characterized in that: The decoding layer includes a GRU module and a fully connected layer to capture temporal dynamic features.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the CTG neural network assisted navigation method in the GNSS weak signal scenario as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the CTG neural network assisted navigation method in a GNSS weak signal scenario as described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the CTG neural network assisted navigation method in a GNSS weak signal scenario as described in any one of claims 1 to 7 is implemented.

Images