Nlos signal identification method and apparatus based on environmental perception

Abstract

The application belongs to the technical field of satellite positioning and navigation, and discloses a NLOS signal identification method and device based on environment perception, which comprises the following steps: processing original measurement data of a GNSS receiver to obtain satellite observation characteristics of each satellite, wherein the satellite observation characteristics comprise carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudo-range residual error and pseudo-range rate consistency; determining a four-channel image based on the satellite observation characteristics of each satellite, inputting the four-channel image into a CNN network for feature extraction to obtain scene characteristics; fusing the scene characteristics and the satellite observation characteristics corresponding to each satellite to obtain a target feature vector corresponding to each satellite; inputting the target feature vector corresponding to each satellite into an LSTM network at a preset time step for identification to obtain a NLOS signal classification result corresponding to each satellite. The application can improve the accuracy of NLOS signal identification in GNSS navigation and positioning.

Description

Technical Field

[0001] This application relates to the technical field of satellite navigation and positioning, and in particular to a method and apparatus for NLOS signal recognition based on environmental perception. Background Technology

[0002] The latest trend in GNSS applications is the rise of smart applications in the mass market, including pedestrian navigation, mobile positioning, and autonomous driving. These applications are rapidly dominating the market. However, the inherent limitations and vulnerabilities of GNSS signals pose a significant obstacle to modern urban applications, as signals can be reflected by buildings, vehicles, and the ground. This phenomenon is known as multipath propagation. Furthermore, there is the situation where line-of-sight (LOS) signals are blocked, and only reflected signals are received, known as non-line-of-sight (NLOS) signals. NLOS signals are particularly common in densely populated urban areas.

[0003] Many researchers have proposed various techniques to mitigate the effects of multipath and NLOS signals, including antenna design, hardware-based solutions, receiver-based methods, post-receive processing techniques, and methods based on modern deep learning. Among them, (1) the scheme that relies on strengthening antenna design to alleviate multipath and NLOS identification is limited by factors such as terminal cost and power consumption, making it unsuitable for widespread consumer use; (2) the scheme that improves the structural design of receiver correlators or the form of correlation functions to alleviate multipath and NLOS identification has a large computational load and high hardware performance requirements, which limits its practical application; (3) the post-received technical processing method takes into account the user's location and can use three-dimensional building models to identify signal types and eliminate signals not observed during the positioning process. It requires detailed modeling of the city in advance, resulting in slow updates and difficulty in expansion; (4) based on modern deep learning technology, machine learning is applied to improve the performance of GNSS positioning. Various classification algorithms such as decision tree (DT), support vector machine (SVM), convolutional neural network (CNN) and adaptive neural fuzzy inference system have been used for GNSS signal identification. The above deep learning methods rely on analyzing the time characteristics of the observation data, while the feature quantities based on terminal measurement only reflect the time changes of the signal and lack the spatial distribution information of the satellite. Moreover, under noisy conditions, the observation data caused by NLOS signals exhibits nonlinear time changes, which leads to the difficulty and low accuracy of the above methods in signal type identification. Summary of the Invention

[0004] This application provides an environment-aware NLOS signal identification method and apparatus, which can improve the accuracy of NLOS signal identification in the field of GNSS navigation and positioning.

[0005] In a first aspect, embodiments of this application provide an environment-aware NLOS signal recognition method, the method comprising:

[0006] The raw measurement data from the GNSS receiver is processed to obtain the satellite observation characteristics of each satellite, which include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency.

[0007] Based on the satellite observation characteristics of each satellite, four-channel images are determined, and the four-channel images are input into a CNN network for feature extraction to obtain scene features;

[0008] The target feature vector for each satellite is obtained by fusing scene features with satellite observation features corresponding to each satellite.

[0009] The target feature vectors corresponding to each satellite are input into the LSTM network for recognition at a preset time step to obtain the NLOS signal classification results for each satellite.

[0010] Furthermore, the step of determining the four-channel image based on the satellite observation features of each satellite includes:

[0011] Based on the satellite elevation angle and satellite azimuth angle in the satellite observation characteristics of each satellite, the visible satellite and the obstructed satellite are mapped to the coordinate positions on a square image of a preset radius to obtain a satellite projection image, wherein the satellite projection image is a fourth channel image;

[0012] Based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation feature, Voronoi diagram interpolation is performed on the satellite projection image at the coordinate positions mapped by each satellite to obtain the first channel image, the second channel image, and the third channel image.

[0013] Furthermore, the CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer. The CNN network maps the input four-channel image into an 8x1 vector, which is an 8-dimensional scene feature vector.

[0014] Furthermore, the LSTM network sequentially comprises two LSTM layers, two fully connected hidden layers, and one output layer. The first LSTM layer has a feature dimension of 128, the second LSTM layer has a feature dimension of 64, the first fully connected hidden layer has a feature dimension of 16, and the second fully connected hidden layer has a feature dimension of 2. The output layer uses the softmax activation function to convert the input vector into class probabilities of LOS and NLOS signals. The LSTM network employs the Adam optimizer for network model optimization.

[0015] Furthermore, after the step of inputting the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step to obtain the NLOS signal recognition results corresponding to each satellite, the method further includes:

[0016] The NLOS signal classification results corresponding to each satellite are applied to the weighted least squares method to adjust the weight of each satellite's observation data in the positioning solution.

[0017] Secondly, the present invention also provides an environment-aware NLOS signal recognition device, the device comprising:

[0018] The satellite observation feature determination module is used to process the raw measurement data of the GNSS receiver to obtain the satellite observation features of each satellite. The satellite observation features include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency.

[0019] The scene feature extraction module is used to determine four-channel images based on the satellite observation features of each satellite, and input the four-channel images into a CNN network for feature extraction to obtain scene features;

[0020] The feature fusion module is used to fuse scene features with satellite observation features corresponding to each satellite to obtain the target feature vector corresponding to each satellite.

[0021] The NLOS signal recognition module is used to input the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step, so as to obtain the NLOS signal classification results corresponding to each satellite.

[0022] Furthermore, the scene feature extraction module is also specifically used for:

[0023] Based on the satellite elevation angle and satellite azimuth angle in the satellite observation characteristics of each satellite, the visible satellite and the obstructed satellite are mapped to the coordinate positions on a square image of a preset radius to obtain a satellite projection image, wherein the satellite projection image is a fourth channel image;

[0024] Based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation feature, Voronoi diagram interpolation is performed on the satellite projection image at the coordinate positions mapped by each satellite to obtain the first channel image, the second channel image, and the third channel image.

[0025] Furthermore, the CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer. The CNN network maps the input four-channel image into an 8x1 vector, which is an 8-dimensional scene feature vector.

[0026] Furthermore, the LSTM network sequentially includes two LSTM layers, two fully connected hidden layers, and one output layer. The first LSTM layer has a feature dimension of 128, the second LSTM layer has a feature dimension of 64, the first fully connected hidden layer has a feature dimension of 16, and the second fully connected hidden layer has a feature dimension of 2. The output layer uses the softmax activation function to convert the input vector into class probabilities of LOS and NLOS signals. The LSTM network uses the Adam optimizer to optimize the network model.

[0027] Furthermore, after the step of inputting the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step to obtain the NLOS signal recognition results corresponding to each satellite, the method further includes:

[0028] The NLOS signal classification results corresponding to each satellite are applied to the weighted least squares method to adjust the weight of each satellite's observation data in the positioning solution.

[0029] Thirdly, embodiments of this application also provide an environment-aware NLOS signal recognition device, the environment-aware NLOS signal recognition device including a processor, a memory, and an environment-aware NLOS signal recognition program stored in the memory and executable by the processor, wherein when the environment-aware NLOS signal recognition program is executed by the processor, it implements the steps of the environment-aware NLOS signal recognition method as described above.

[0030] Fourthly, embodiments of this application also provide a computer-readable storage medium storing an environment-aware NLOS signal recognition program, wherein when the environment-aware NLOS signal recognition program is executed by a processor, it implements the steps of the environment-aware NLOS signal recognition method as described above.

[0031] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following:

[0032] This application provides an environment-aware NLOS signal identification method and apparatus that combines the temporal characteristics of satellite signals with the spatial constraints of the surrounding environment of the GNSS receiver. An LSTM network is used to process the long-term dependencies and temporal information of the combined time series data. By learning the long-term correlation between the temporal dependencies of satellite signals, the surrounding environment information of the GNSS receiver, and the NLOS signal type, the signal type is identified, thereby improving the accuracy of NLOS signal identification in GNSS navigation and positioning. Attached Figure Description

[0033] Figure 1 A flowchart illustrating an environment-aware NLOS signal recognition method provided in one embodiment of this application;

[0034] Figure 2 A flowchart illustrating the model structure of an environment-aware NLOS signal recognition method according to an embodiment of this application;

[0035] Figure 3 A flowchart illustrating the LSTM network model structure in an environment-aware NLOS signal recognition method provided in one embodiment of this application;

[0036] Figure 4 A schematic diagram of the functional modules of an environment-aware NLOS signal recognition device provided in one embodiment of this application;

[0037] Figure 5 This is a schematic diagram of the hardware structure of an environment-aware NLOS signal recognition device involved in one embodiment of this application. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] See Figure 1 This application provides an environment-aware NLOS signal recognition method, which specifically includes:

[0040] Step S10: Process the raw measurement data of the GNSS receiver to obtain the satellite observation characteristics of each satellite. The satellite observation characteristics include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency.

[0041] Step S20: Determine four-channel images based on the satellite observation characteristics of each satellite, and input the four-channel images into a CNN network for feature extraction to obtain scene features;

[0042] Step S30: Based on the scene features and the satellite observation features corresponding to each satellite, the target feature vector corresponding to each satellite is obtained by fusing the scene features and the satellite observation features corresponding to each satellite;

[0043] Step S40: Input the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step to obtain the NLOS signal classification results corresponding to each satellite.

[0044] In this embodiment, considering that in complex urban environments, the feature quantities measured by the terminal only reflect the temporal changes of satellite navigation signals (LOS / NLOS signals) and lack spatial distribution information of satellites, the observed data caused by NLOS signals exhibit nonlinear temporal changes under noisy conditions, leading to difficulties and low accuracy in identifying GNSS navigation signal types. Furthermore, the signal state (LOS / NLOS) changes of satellite signals during propagation are typically relatively slow. Therefore, this embodiment combines the temporal characteristics of satellite signals with the spatial constraints of the environment surrounding the GNSS receiver, and uses an LSTM network to process the long-term dependencies and temporal information of the combined time-series data. By learning the long-term correlation between the temporal dependencies of satellite signals, the environmental information surrounding the GNSS receiver, and the NLOS signal type, the signal type is identified. This helps predict satellite visibility closely related to the environment, thereby improving the accuracy of NLOS signal identification in the field of GNSS navigation and positioning.

[0045] Specifically, in this embodiment, the observation information (raw measurement data) received by the GNSS receiver from each satellite is first integrated. This raw measurement data is then processed to obtain the satellite observation characteristics of each satellite. The satellite observation characteristics determined in this embodiment include carrier-to-noise ratio (C / N0), satellite elevation angle E1, satellite azimuth angle Az, pseudorange residual Pr, and pseudorange rate consistency Prc. Specifically, the satellite elevation angle E1 corresponds to the altitude angle of the satellite relative to the receiver in the sky. Satellite signals are often more susceptible to obstruction by buildings, trees, or terrain at low elevation angles, resulting in NLOS signals. Therefore, the satellite elevation angle E1 is used as one of the satellite observation characteristics to identify NLOS signals. The satellite azimuth angle Az corresponds to the azimuth angle of the satellite relative to the receiver in the sky. Different azimuth angles may correspond to different propagation paths (including NLOS paths), resulting in NLOS signals. Therefore, the satellite azimuth angle Az is used as one of the satellite observation characteristics to identify NLOS signals. The carrier-to-noise ratio (C / N0) corresponds to the ratio between the carrier power and noise density of the satellite signal received by the receiver. It is an indicator of satellite signal quality; a higher C / N0 value indicates better signal quality. A low C / N0 value may indicate signal interference or attenuation, potentially related to NLOS signals. Therefore, the carrier-to-noise ratio (C / N0) is used as one of the satellite observation features for identifying NLOS signals. The pseudorange residual (Pr) is the difference between the actual received pseudorange measurement and the expected pseudorange measurement. Due to the additional propagation delay and interference caused by the NLOS path, the pseudorange residual usually increases; therefore, Pr is also used as one of the satellite observation features for identifying NLOS signals. The pseudorange rate consistency (Prc) is the consistency between the received satellite signal's pseudorange change rate and the expected pseudorange change rate. In the case of NLOS signals, due to the complexity of the propagation path, the pseudorange rate consistency may also be affected by interference or deviation; therefore, Prc is also used as one of the satellite observation features for identifying NLOS signals. These satellite observation features can appropriately represent the characteristics of satellite signals in different observation epochs, thereby further assisting in subsequent accurate NLOS signal identification and classification based on these features.

[0046] After determining the satellite observation characteristics, four-channel images are generated based on these characteristics and input into a CNN network. The CNN's environmental awareness function is used to extract representations of the surrounding environment, resulting in scene features. This integrates the observation information received by the GNSS receiver from each satellite into a vector representing the environment surrounding the GNSS receiver. By fusing these scene features with the corresponding satellite observation characteristics of each satellite, a target feature vector for each satellite can be obtained. This target feature vector contains the spatial distribution information of the satellites. Different observation epochs of the same satellite can be processed to obtain corresponding target feature vectors. The target feature vectors of multiple observation epochs over a long period are input into an LSTM network according to a pre-set time step for recognition, yielding the NLOS signal classification results for each satellite. Comparative experiments showing NLOS recognition with and without scene features demonstrate a correlation between NLOS recognition and environmental information. By comparing the recognition accuracy and stability of the LSTM model in two sets of comparative experiments, it can be determined that adding scene features for NLOS recognition can improve the accuracy of GNSS NLOS signal recognition results in multiple scenarios (deep indoor, shallow indoor, semi-outdoor, open-air, etc.).

[0047] Furthermore, in one embodiment, the step of determining the four-channel image based on the satellite observation features of each satellite includes:

[0048] Based on the satellite elevation angle and satellite azimuth angle in the satellite observation characteristics of each satellite, the visible satellite and the obstructed satellite are mapped to the coordinate positions on a square image of a preset radius to obtain a satellite projection image, wherein the satellite projection image is a fourth channel image;

[0049] Based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation feature, Voronoi diagram interpolation is performed on the satellite projection image at the coordinate positions mapped by each satellite to obtain the first channel image, the second channel image, and the third channel image.

[0050] In this embodiment, to optimize the spatiotemporal information contained in GNSS measurements, the step of determining four-channel images based on the satellite observation characteristics of each satellite is specifically referred to... Figure 2 During each epoch, the Satellite Projection module maps visible satellites (white dots) and obstructed satellites (black dots) onto a square image, resulting in a satellite projection image, which serves as the fourth channel of the 4-channel image. The satellite's coordinates (x, y) on the image plane are calculated using parameters such as the image radius *r* and the elevation angle. (corresponding to satellite elevation angle E1) and azimuth angle θ (corresponding to satellite azimuth angle A2), After obtaining the satellite projection image through mapping, the Voronoi tessellation module in this implementation uses three GNSS measurements (carrier-to-noise ratio C / N0, pseudorange residual Pr, and pseudorange rate consistency Prc from the satellite observation characteristics) to interpolate the satellite projection image (the above three GNSS measurements are inserted into the coordinate positions corresponding to the projected satellite respectively), thereby obtaining the three-channel image of the other three channels.

[0051] Furthermore, in one embodiment, the CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer. The CNN network maps the input four-channel image into an 8x1 vector, where the 8x1 vector is an 8-dimensional scene feature vector.

[0052] In this embodiment, refer to Figure 2 After the four-channel image is input into the CNN network, the corresponding scene feature extraction module (Construct environment representations) then utilizes the CNN's environment awareness function to extract representations of the surrounding environment from the four-channel image. The fully connected layers of the CNN network map the feature map into an 8x1 vector, and the resulting 8x1 vector is an 8-dimensional scene feature vector Sf. (Referring to...) Figure 2 Target feature vector such as X t =[C / N 0t El t Az t ,Pr t ,Prc t ,Sf t 8×1 ] 13×1 , Figure 2 In this context, variables k and n represent the time step size of the LSTM and the number of observation epochs for the same satellite, respectively.

[0053] Specifically, a CNN network may include: an input layer that receives a four-channel image as input, with dimensions (dim_x, dim_y, num_channels), where dim_x and dim_y represent the width and height of the image, and num_channels represents the number of channels; a first convolutional layer that uses 16 3x3 kernels for convolution, employing ReLU activation, with an input shape of (input_shape) and 'same' padding to maintain consistent input and output feature map dimensions; a batch normalization layer that batch normalizes the output of the first convolutional layer to accelerate training; and a first pooling layer that uses 2x... The first convolutional layer uses a 2x2 pooling window for max pooling to reduce the size of the feature map. The second convolutional layer uses 32 kernels of size 3x3, with ReLU activation. A batch normalization layer normalizes the output of the second convolutional layer. The second pooling layer uses a 2x2 pooling window for max pooling. A flattening layer flattens the output feature map of the second pooling layer into a one-dimensional vector, preparing it for the subsequent fully connected layers. The first fully connected layer has 2048 neurons and uses ReLU activation. The second fully connected layer has 512 neurons and uses ReLU activation. The third fully connected layer has 8 neurons and uses ReLU activation. The third fully connected layer outputs an 8-dimensional feature vector, where each dimension represents a scene feature. This feature vector can then be used to characterize the scene at a given moment.

[0054] In this method, since NLOS signal recognition is based on the fusion of scene features and satellite observation features to obtain a target feature vector, if the length of a certain feature in the target feature vector is too long or too short, it will introduce bias and feature weight imbalance. This results in high complexity, long processing time, and high computational resource consumption, thus affecting the overall NLOS signal recognition performance. Therefore, in this embodiment, the feature map obtained after a series of processing steps on the four-channel image input is mapped to 8 dimensions. This ensures that while introducing scene features improves the accuracy of NLOS signal recognition, it does not affect the overall recognition performance or reduce recognition efficiency.

[0055] Further, in one embodiment, the LSTM network sequentially includes two LSTM layers, two fully connected hidden layers, and one output layer. The feature dimension of the first LSTM layer is 128, the feature dimension of the second LSTM layer is 64, the feature dimension of the first fully connected hidden layer is 16, and the feature dimension of the second fully connected hidden layer is 2. The output layer uses the softmax activation function to convert the input vector into the class probabilities of LOS and NLOS signals. The LSTM network uses the Adam optimizer to optimize the network model.

[0056] In this embodiment, the LSTM network referenced for NLOS signal identification in GNSS is... Figure 3 , Figure 3 In this model, T represents the number of recursive steps in the sequence. T feature vectors are input into the LSTM model, passing through two LSTM layers, two fully connected (FC) layers, and an output layer Y. This yields the probability that each feature vector corresponds to either a LOS or NLOS signal under scene space constraints. Both LSTM layers contain a predetermined number of LSTM units. The first LSTM layer has a feature dimension of 128, and the second LSTM layer has a feature dimension of 64. The predetermined number of LSTM units is determined based on a comprehensive consideration of the complexity of NLOS signal recognition, the size of the dataset, and available computational resources. Appropriate model selection, parameter tuning, and validation are also necessary to achieve optimal model performance and generalization ability. After learning the long-term correlation between the time dependence of satellite signals, the environmental information surrounding the GNSS receiver, and the NLOS signal type, two fully connected hidden (FC) layers selectively pass the feature information input to the second LSTM layer, thus performing a dimensionality transformation. The feature dimension of the first fully connected hidden layer of the two LSTM layers is 16, and the feature dimension of the second fully connected hidden (FC) layer is 2. In the output layer Y of the LSTM network, the softmax activation function is used to convert the input vector into class probabilities of LOS and NLOS signals. The softmax activation function can provide probabilistic interpretation in the LSTM model classifier, which helps to improve the performance and interpretability of the LSTM model. Furthermore, the LSTM network in this embodiment uses the Adam optimizer for network model optimization training. Compared with other optimizers (such as the SGD optimizer), it has adaptive learning rate, momentum, and adaptive regularization characteristics, achieving faster convergence speed, better generalization ability, and more stable gradient estimation to optimize the LSTM network.

[0057] Furthermore, in one embodiment, after the step of inputting the target feature vectors corresponding to each satellite into the LSTM network for identification at a preset time step to obtain the NLOS signal identification results corresponding to each satellite, the method further includes:

[0058] The NLOS signal classification results corresponding to each satellite are applied to the weighted least squares method to adjust the weight of each satellite's observation data in the positioning solution.

[0059] In this embodiment, Figure 3The classification results of the Y output layer are combined with the "weighted least squares method" to adaptively adjust the weight of each satellite observation data in the GNSS positioning solution based on the classification results of the NLOS signals corresponding to each satellite. Higher weights are assigned to more reliable satellite observations to suppress the influence of abnormal observations, thereby improving the robustness and accuracy of satellite positioning results.

[0060] This application also provides an NLOS signal recognition device based on environmental perception.

[0061] Reference Figure 4 A schematic diagram of the functional modules of the first embodiment of the NLOS signal recognition device based on environmental perception.

[0062] In this embodiment, the environment-aware NLOS signal recognition device includes:

[0063] The satellite observation feature determination module 10 is used to process the raw measurement data of the GNSS receiver to obtain the satellite observation features of each satellite. The satellite observation features include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency.

[0064] The scene feature extraction module 20 is used to determine four-channel images based on the satellite observation features of each satellite, and input the four-channel images into a CNN network for feature extraction to obtain scene features;

[0065] Feature fusion module 30 is used to fuse scene features with satellite observation features corresponding to each satellite to obtain target feature vectors corresponding to each satellite;

[0066] The NLOS signal recognition module 40 is used to input the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step, so as to obtain the NLOS signal classification results corresponding to each satellite.

[0067] Furthermore, in one embodiment, the scene feature extraction module 20 is further specifically used for:

[0068] Based on the satellite elevation angle and satellite azimuth angle in the satellite observation characteristics of each satellite, the visible satellite and the obstructed satellite are mapped to the coordinate positions on a square image of a preset radius to obtain a satellite projection image, wherein the satellite projection image is a fourth channel image;

[0069] Based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation feature, Voronoi diagram interpolation is performed on the satellite projection image at the coordinate positions mapped by each satellite to obtain the first channel image, the second channel image, and the third channel image.

[0070] Furthermore, in one embodiment, the CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer. The CNN network maps the input four-channel image into an 8x1 vector, where the 8x1 vector is an 8-dimensional scene feature vector.

[0071] Further, in one embodiment, the LSTM network sequentially includes two LSTM layers, two fully connected hidden layers, and one output layer. The feature dimension of the first LSTM layer is 128, the feature dimension of the second LSTM layer is 64, the feature dimension of the first fully connected hidden layer is 16, and the feature dimension of the second fully connected hidden layer is 2. The output layer uses the softmax activation function to convert the input vector into the class probabilities of LOS and NLOS signals. The LSTM network uses the Adam optimizer to optimize the network model.

[0072] Furthermore, in one embodiment, after the step of inputting the target feature vectors corresponding to each satellite into the LSTM network for identification at a preset time step to obtain the NLOS signal identification results corresponding to each satellite, the method further includes:

[0073] The NLOS signal classification results corresponding to each satellite are applied to the weighted least squares method to adjust the weight of each satellite's observation data in the positioning solution.

[0074] The functions of each module in the above-mentioned environment-aware NLOS signal recognition device correspond to the steps in the above-mentioned environment-aware NLOS signal recognition method embodiment, and their functions and implementation processes will not be described in detail here.

[0075] This application provides an environment-aware NLOS signal recognition device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0076] Reference Figure 5 , Figure 5This is a schematic diagram of the hardware structure of the NLOS signal recognition device based on environmental awareness involved in the embodiments of this application. In this embodiment, the NLOS signal recognition device based on environmental awareness may include a processor 1001 (e.g., a Central Processing Unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize communication between these components; the user interface 1003 may include a display screen or an input unit such as a keyboard; the network interface 1004 may optionally include a standard wired interface or a wireless interface (e.g., Wireless Fidelity, Wi-Fi interface); the memory 1005 may be high-speed random access memory (RAM) or stable memory (non-volatile memory), such as a disk storage device. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001. Those skilled in the art will understand that… Figure 5 The hardware structure shown does not constitute a limitation of the invention and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0077] Continue to refer to Figure 5 , Figure 5 The memory 1005, which serves as a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and an environment-aware NLOS signal recognition program. The processor 1001 can call the environment-aware NLOS signal recognition program stored in the memory 1005 and execute the steps of the environment-aware NLOS signal recognition method provided in this embodiment.

[0078] The method implemented when the environment-aware NLOS signal recognition program is executed can be referred to in various embodiments of the environment-aware NLOS signal recognition method of this application, and will not be repeated here.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for identifying NLOS signals based on environmental perception, characterized in that, The method includes: The raw measurement data from the GNSS receiver is processed to obtain the satellite observation characteristics of each satellite, which include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency. The four-channel image is determined based on the satellite observation characteristics of each satellite. Specifically, this includes: mapping the visible and obstructed satellites to the coordinate positions on a square image of a preset radius based on the satellite elevation angle and azimuth angle of each satellite observation characteristics, to obtain a satellite projection image, wherein the satellite projection image is the fourth channel image; performing Voronoi diagram interpolation on the satellite projection image at the coordinate positions mapped by each satellite based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation characteristics, to obtain a first channel image, a second channel image, and a third channel image; and inputting the four-channel image into a CNN network for feature extraction to obtain scene features, wherein the CNN network maps the input four-channel image into an 8x1 vector, and the 8x1 vector is an 8-dimensional scene feature vector. The target feature vector for each satellite is obtained by fusing scene features with satellite observation features corresponding to each satellite. The target feature vectors corresponding to each satellite are input into an LSTM network for identification at a preset time step to obtain the NLOS signal classification results for each satellite. The LSTM network consists of two LSTM layers, two fully connected hidden layers, and one output layer. The feature dimension of the first LSTM layer is 128, the feature dimension of the second LSTM layer is 64, the feature dimension of the first fully connected hidden layer is 16, and the feature dimension of the second fully connected hidden layer is 2. The output layer uses the softmax activation function to convert the input vector into the class probabilities of LOS and NLOS signals. The LSTM network uses the Adam optimizer to optimize the network model.

2. The method according to claim 1, characterized in that, The CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer.

3. The method according to claim 1, characterized in that, After the step of inputting the target feature vectors corresponding to each satellite into the LSTM network for recognition at a preset time step to obtain the NLOS signal recognition results corresponding to each satellite, the method further includes: The NLOS signal classification results corresponding to each satellite are applied to the weighted least squares method to adjust the weight of each satellite's observation data in the positioning solution.

4. An NLOS signal recognition device based on environmental perception, characterized in that, The device includes: The satellite observation feature determination module is used to process the raw measurement data of the GNSS receiver to obtain the satellite observation features of each satellite. The satellite observation features include carrier-to-noise ratio, satellite elevation angle, satellite azimuth angle, pseudorange residual, and pseudorange rate consistency. The scene feature extraction module is used to determine four-channel images based on the satellite observation features of each satellite. Specifically, it includes: mapping the visible satellites and obstructed satellites to the coordinate positions on a square image of a preset radius based on the satellite elevation angle and satellite azimuth angle of each satellite observation feature, to obtain a satellite projection image, wherein the satellite projection image is the fourth channel image; performing Voronoi diagram interpolation processing on the satellite projection image at the coordinate positions mapped by each satellite based on the carrier-to-noise ratio, pseudorange residual, and pseudorange rate consistency of each satellite observation feature, to obtain a first channel image, a second channel image, and a third channel image; and inputting the four-channel images into a CNN network for feature extraction to obtain scene features, wherein the CNN network maps the input four-channel images into an 8x1 vector, and the 8x1 vector is an 8-dimensional scene feature vector; The feature fusion module is used to fuse scene features with satellite observation features corresponding to each satellite to obtain the target feature vector corresponding to each satellite. The NLOS signal recognition module is used to input the target feature vectors corresponding to each satellite into an LSTM network for recognition at a preset time step, thereby obtaining the NLOS signal classification results for each satellite. The LSTM network consists of two LSTM layers, two fully connected hidden layers, and one output layer. The feature dimension of the first LSTM layer is 128, the feature dimension of the second LSTM layer is 64, the feature dimension of the first fully connected hidden layer is 16, and the feature dimension of the second fully connected hidden layer is 2. The output layer uses the softmax activation function to convert the input vector into the class probabilities of LOS and NLOS signals. The LSTM network uses the Adam optimizer to optimize the network model.

5. The apparatus according to claim 4, characterized in that, The CNN network includes a first convolutional layer, a batch normalization layer, a first pooling layer, a second convolutional layer, a batch normalization layer, a second pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer.

6. An NLOS signal recognition device based on environmental perception, characterized in that, The environment-aware NLOS signal recognition device includes a processor, a memory, and an environment-aware NLOS signal recognition program stored in the memory and executable by the processor, wherein when the environment-aware NLOS signal recognition program is executed by the processor, it implements the steps of the environment-aware NLOS signal recognition method as described in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an environment-aware NLOS signal recognition program, wherein when the environment-aware NLOS signal recognition program is executed by a processor, it implements the steps of the environment-aware NLOS signal recognition method as described in any one of claims 1 to 3.

Images