A Deep Learning-Based NLOS Recognition Method Applicable to Complex Environments in Multiple Scenarios

Abstract

This invention discloses a deep learning-based NLOS recognition method applicable to complex environments in multiple scenarios. First, the collected UWB channel impulse response data from multiple scenarios is transformed to obtain discrete multipath impulse response data. Multipath delay power data and multipath similarity data are then calculated to construct a new sample dataset, followed by data preprocessing. Next, a dual-input feature fusion deep learning NLOS recognition model based on multipath information is constructed to complete NLOS classification in UWB positioning systems. This deep learning-based NLOS recognition method, applicable to complex environments in multiple scenarios, can solve the problems of low recognition capability, poor generalization accuracy, and poor recognition effect in complex scenarios with limited data, thereby achieving high NLOS recognition accuracy and effectively improving UWB positioning accuracy in complex environments.

Description

Technical Field

[0001] This invention relates to a method for identifying non-line-of-sight (NLOS) propagating wireless signals, specifically a deep learning NLOS identification method applicable to complex environments in multiple scenarios, belonging to the field of ultra-wideband technology. Background Technology

[0002] Positioning technology has wide applications in modern life, ranging from navigation and asset tracking to smart manufacturing. While various positioning methods exist, Ultra Wide Band (UWB) positioning technology has become a popular choice in indoor environments due to its high accuracy, low power consumption, and resistance to multipath interference. UWB positioning primarily relies on ranging technology. However, in complex environments with multiple scenarios, factors such as building structures and obstacles can cause wireless signals to propagate non-line-of-sight (NLOS). NLOS propagation can lead to ranging errors, resulting in inaccurate positioning. Therefore, accurate identification of NLOS signals is crucial in UWB positioning systems. Correct NLOS identification can significantly improve positioning accuracy and enhance the system's robustness, enabling it to adapt to various complex multi-scenario environments.

[0003] Currently, NLOS identification methods are mainly divided into channel data-based methods and channel feature-based methods. Channel data-based methods directly identify NLOS propagation using raw channel data, but their performance depends heavily on the sheer volume of data. A significant portion of this massive amount of channel data is useless or noisy, increasing data processing costs and potentially leading to overfitting. Channel feature-based methods extract features from the raw channel data and then use machine learning or various thresholding methods for classification. However, feature extraction is complex and may lose crucial information. Thresholding-based methods struggle to effectively distinguish between line-of-sight (LOS) propagation and NLOS propagation in complex environments. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a deep learning NLOS recognition method applicable to complex environments in multiple scenarios. It can effectively avoid the shortcomings of traditional channel data-based and channel feature-based methods, and achieve accurate recognition of NLOS signals. It is particularly suitable for the recognition of NLOS signals in the UWB system positioning process in complex environments in multiple scenarios.

[0005] To achieve the above objectives, this deep learning-based NLOS recognition method, applicable to complex environments in multiple scenarios, specifically includes the following steps:

[0006] Step 1: Based on the publicly available datasets of multiple scenarios collected by the UWB positioning system, collect the channel pulse response data between anchor points and tags in the UWB positioning system, and label the LOS data tags and NLOS data tags.

[0007] Step 2: Obtain CIR multipath discrete data h based on FPI. m (n);

[0008] Step 3, based on the multipath discrete data h m (n) Obtain the multipath signal delay power data TDP(x);

[0009] Step 4: First, process the multipath discrete data h m (n) Obtain multipath continuous data x(t) through linear interpolation, and then reconstruct the multipath continuous data x(t) into denoised multipath data through VMD. Then, the multipath continuous data x(t) and the denoised multipath data are calculated. The similarity features are used as multipath similarity feature data;

[0010] Step 5: For the delay power data TDP(x) and multipath similarity feature data of multipath discrete data, a multipath data feature extraction module and a multipath similarity data extraction module are constructed to extract the features of multipath information. The multipath data features and multipath similarity features are fused to form a dual-channel multipath feature fusion NLOS classification module. The multipath data feature extraction module, the multipath similarity data extraction module and the multipath feature fusion NLOS classification module together form a dual-input feature fusion deep learning NLOS recognition model based on multipath information.

[0011] Step 6 involves transforming all channel impulse response data from the multi-scenario public dataset in Step 1 to obtain multipath delay power data TDP(x) and multipath similarity feature data, constructing a new sample dataset, and then performing data preprocessing to set up training, validation, and test sets based on the new sample dataset.

[0012] Step 7: Input the preprocessed training data into the dual-input feature fusion deep learning NLOS recognition model based on multipath information to train and validate the model. Stop training when the model converges to complete the NLOS classification in the UWB positioning system.

[0013] Furthermore, Step 2 is detailed below:

[0014] For the acquired channel impulse response data, the channel model of the acquisition device, defined by the PRF and CIR, is as follows:

[0015]

[0016] In the formula: t represents the timestamp of each value in CIR; S is the number of multipath components; a s It is the amplitude of the Sth multipath component; τ s δ(·) is the time delay of the S-th multipath component; δ(·) is the Dirac function; n(t) represents the additive white Gaussian noise present in the channel;

[0017] The acquisition device represents the acquired CIR as a complex IQ sample with a value of 1 ns, and then obtains the corresponding discrete CIR, which is represented as:

[0018]

[0019] In the formula: I n Q n These represent the real and imaginary parts of the nth complex value of the channel impulse response, acquired in units of 1 ns; P is the number of sampling points for each CIR data point.

[0020] The 150 channel impulse response data points, including the FPI and the first path signal, are taken as multipath data. The multipath channel impulse response can be expressed as:

[0021] h m (n)=h(n),n0≤t<n0+150

[0022] Where: h m (n) represents the multipath discrete data of the channel impulse response; n0 is the first path index time value in ns; h(n) represents the discrete CIR.

[0023] Furthermore, Step 3 is detailed below:

[0024] The delay power data TDP(x) of multipath discrete CIR data is specifically defined as follows:

[0025]

[0026] In the formula: L is the sampling factor; x is the data interval based on the sampling factor; n0 corresponds to h m The initial time of (n).

[0027] Furthermore, by taking a sampling factor of L = 5, the delay power data TDP(x) of the multipath discrete CIR data is obtained.

[0028] Furthermore, Step 4 is detailed below:

[0029] Step 4-1, Multipath Discrete Data h m (n) Obtain multipath continuous data x(t) through linear interpolation;

[0030] Multipath continuous data x(t) is represented as:

[0031]

[0032] In the formula: It is an interval indicator function, when t∈[t n ,t n+1 The value is 1 when the condition is met, and 0 otherwise; t n and t n+1 t is the discrete multipath CIR time point; h is the interpolation time point; m (n) and h m (n+1) represents the discrete CIR signal values ​​adjacent to the corresponding discrete time points; h m (n) represents the multipath discrete data; N is the length of the discrete multipath data.

[0033] Step 4-2: Introduce VMD to reconstruct and denoise the multipath continuous data x(t) to obtain denoised multipath data.

[0034] VMD solves for each mode u using the following optimization method. k (t) and the corresponding center frequency w k :

[0035]

[0036] In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal; w k It is the center frequency of the k-th modal signal;

[0037] Denoising multipath data obtained from denoising reconstruction Represented as:

[0038]

[0039] In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal;

[0040] Step 4-3: Calculate the multipath continuous data x(t) and the denoised multipath data. The similarity features are used as multipath similarity feature data;

[0041] Calculate multipath continuous data x(t) and denoised multipath data The similarity metrics are introduced, including CS, SES, PCC, SCC, Manhattan Distance, and Chebyshev Distance. The similarity features of multipath data are represented as follows:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] In the formula: A and B represent the multipath continuous data x(t) and the denoised multipath data, respectively. and Let A and B be the ranks, respectively.

[0049] Furthermore, Step 5 is detailed below:

[0050] The multipath data feature extraction module is based on the Attentive CNN model. For the input multipath delay power data, the multipath data feature extraction module uses convolutional layers to extract local features and enhances the ability to perceive key features by introducing an attention mechanism.

[0051] The multipath similarity feature extraction module is based on the MLP model. For the similarity relationship between multipath data, the multipath similarity feature extraction module uses a fully connected structure to extract similarity features.

[0052] The multipath feature fusion NLOS classification module is based on a DNN model. After extracting multipath data features and multipath similarity features, the multipath feature fusion NLOS classification module adopts a feature fusion strategy to integrate the two types of features in a dual-channel manner. The fused features are then fed into the DNN model for NLOS classification.

[0053] Furthermore, in Step 6, when performing data preprocessing based on the new sample dataset, the ratio of the training set, validation set, and test set is set to 7:1:2.

[0054] Furthermore, in Step 7, when the preprocessed training data is input into the deep learning NLOS recognition model based on multipath information and dual-input feature fusion for training and validating the model, the loss function of the deep learning NLOS recognition model based on multipath information and dual-input feature fusion for training is set to cross-entropy, the optimizer is Adam, and the activation function of the classification layer is Softmax. After the sample data of the training set is input into the model, features are first extracted and feature fusion is completed through forward propagation. Then, the loss between the predicted value and the true value is calculated, and the gradient is calculated and the parameters are updated through backpropagation. Training stops when the accuracy of the validation set does not improve significantly within multiple training cycles, and the model with the best performance on the validation set is saved for testing, thus completing the model training.

[0055] Compared with existing technologies, this deep learning NLOS recognition method, applicable to complex environments in multiple scenarios, first transforms the collected UWB channel impulse response data from multiple scenarios to obtain discrete multipath impulse response data. Multipath delay power data and multipath similarity data are then calculated to construct a new sample dataset, followed by data preprocessing. Next, a multipath data feature extraction module, a multipath similarity feature extraction module, and a multipath feature fusion classification module are constructed. These three modules constitute a dual-input feature fusion deep learning NLOS recognition model based on multipath information. The preprocessed training data is then input into this model to train and validate it. Training stops when the model converges, completing the NLOS classification in the UWB positioning system. Regarding the accurate recognition of NLOS signals during UWB system positioning in complex environments in multiple scenarios, the above experimental simulations demonstrate that this deep learning NLOS recognition method, applicable to complex environments in multiple scenarios, can solve the problems of low recognition capability, poor generalization accuracy, and poor recognition effect in complex scenarios with limited data, thereby achieving high NLOS recognition accuracy and effectively improving UWB positioning accuracy in complex environments in multiple scenarios. Attached Figure Description

[0056] Figure 1 This is a flowchart of the present invention;

[0057] Figure 2 This is a network architecture diagram of the dual-input feature fusion deep learning NLOS recognition model based on multipath information of the present invention;

[0058] Figure 3 This is a comparison diagram of the binary classification confusion matrix results of the present invention with support vector machine methods based on radial basis functions, support vector machine methods based on linear kernels, and convolutional bidirectional long short-term memory network models based on the original channel impulse response. Detailed Implementation

[0059] This deep learning-based NLOS recognition method, applicable to complex environments across multiple scenarios, achieves NLOS recognition by preventing key features from being overwhelmed by noise, enhancing the effective fusion of features, and fully utilizing the targeted features of multipath data.

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, this deep learning-based NLOS recognition method, applicable to complex environments in multiple scenarios, specifically includes the following steps:

[0062] Step 1: Based on the publicly available datasets of multiple scenarios collected by the UWB positioning system, collect the channel pulse response data between anchor points and tags in the UWB positioning system, and label the data to indicate whether the data is a LOS signal or an NLOS signal.

[0063] In this embodiment, the performance of the proposed NLOS identification method is evaluated using a publicly available dataset collected from multiple scenarios using a data acquisition device based on the DW 1000UWB chip. The NLOS and LOS channel impulse response data between anchors and tags in the UWB dataset were collected from seven different indoor environments: Office I, Office II, small apartment, small studio, kitchen with living room, bedroom, and boiler room. For each environment, 3000 NLOS channel impulse response measurements and 3000 LOS channel impulse response measurements were collected, totaling 42,000 sets of sample data.

[0064] Step 2: Obtain multipath discrete data h of the Channel Impulse Response (CIR). m (n).

[0065] The pulse repetition frequency (PRF) of the DW 1000UWB acquisition device is set to 64MHz. Each CIR raw data has 1016 sampling points. The CIR channel model is represented as follows:

[0066]

[0067] In the formula: t represents the timestamp of each value in CIR; S is the number of multipath components; a s It is the amplitude of the Sth multipath component; τ s δ(·) is the time delay of the S-th multipath component; δ(·) is the Dirac function; n(t) represents the additive white Gaussian noise (AWGN) present in the channel.

[0068] The DW 1000UWB acquisition device represents the acquired CIR as IQ samples with a complex value every 1 ns. The IQ samples include I (real part) samples and Q (imaginary part) samples, and then obtains the corresponding discrete CIR values, a total of 1016 data points. The discrete CIR is represented as follows:

[0069]

[0070] In the formula: I n Q n These represent the I (real part) and Q (imaginary part) samples of the nth complex value of the channel impulse response, collected in units of 1 ns.

[0071] For the CIR received between anchor points and tags in LOS and NLOS environments acquired in the UWB system, the First Path Index (FPI) is the position of the first path determined by the DW 1000UWB chip. The data before the FPI is a noise stage, while the data after the FPI reflects most of the useful information related to propagation characteristics. Since the length of the selected data samples is environment-dependent, an excessively long selection length will introduce useless information and noise. Therefore, 150 samples are usually sufficient. The 150 channel impulse response data points containing the FPI and the first path signal are taken as multipath data. The multipath channel impulse response can be expressed as:

[0072] h m (n)=h(n),n0≤t<n0+150

[0073] Where: h m (n) represents the multipath discrete data of the channel impulse response; n0 is the first path index time value in ns; h(n) represents the discrete CIR.

[0074] Step 3, based on the multipath discrete data h m (n) Obtain multipath signal delay power data (Time-Delay Power, TDP)TDP(x).

[0075] While typically 150 samples are needed to adequately describe multipath CIR information, multipath discrete data can be optimized to reduce the number of samples without sacrificing performance, thereby accelerating network training. This optimization is feasible for two reasons: first, there is a high correlation between adjacent CIR samples; second, the amplitude and overall shape of the CIR are crucial for channel-based NLOS identification. Optimizing multipath discrete data can incorporate TDP data, which provides a function of the average power of the multipath channel impulse response as a function of delay.

[0076] The multipath signal delay power data TDP(x) can be obtained by calculating the square of the magnitude of the multipath impulse response (i.e., |h|). m (n) 2 To obtain the multipath impulse response data effectively and reduce the input dimension of the neural network, |h is used. m (n) 2 A downsampled version, calculated using a specific downsampling factor L. The proposed training features are based on |h m (n) 2 The delay power data TDP(x) of multipath discrete CIR data, which is the average value over equal intervals, is specifically defined as follows:

[0077]

[0078] In the formula: L is the sampling factor; x is the data interval based on the sampling factor; n0 corresponds to h m The initial time of (n).

[0079] In this embodiment, the sampling factor L = 5 is set to obtain the delay power data TDP(x) of the multipath discrete CIR data.

[0080] Step 4: Calculate and obtain multipath similarity feature data to effectively utilize the multipath discrete data h obtained in Step 2. m (n).

[0081] From multipath discrete data h m (n) Multipath continuous data x(t) is obtained through linear interpolation. The multipath continuous data x(t) contains environmental noise. Noise reduction and reconstruction of the multipath continuous data x(t) can be achieved by introducing Variational Mode Decomposition (VMD) to obtain denoised multipath data. Calculate multipath continuous data x(t) and denoised multipath data The similarity features are used as multipath similarity feature data. Specifically:

[0082] Step 4-1, Multipath Discrete Data h m (n) Obtain multipath continuous data x(t) through linear interpolation.

[0083] Multipath continuous data x(t) can be represented as:

[0084]

[0085] In the formula: It is an interval indicator function, when t∈[t n ,t n+1 The value is 1 when the condition is met, and 0 otherwise; t nand t n+1 t is the discrete multipath CIR time point; h is the interpolation time point; m (n) and h m (n+1) represents the discrete CIR signal values ​​adjacent to the corresponding discrete time points; h m (n) represents the multipath discrete data; N is the length of the discrete multipath data.

[0086] Step 4-2: Introduce VMD to reconstruct and denoise the multipath continuous data x(t) to obtain denoised multipath data.

[0087] VMD decomposes a continuous signal x(t) into several modes u k (t), where each mode represents a signal component in a different frequency band. The goal of VMD is to minimize the bandwidth of each mode signal. To obtain the variational mode decomposition multipath reconstructed signal, VMD solves for each mode u using the following optimization method. k (t) and the corresponding center frequency w k :

[0088]

[0089] In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal; w k It is the center frequency of the k-th modal signal.

[0090] In this embodiment, the number of modes K = 5 is set to obtain the modal signal and the center frequency of the modal signal.

[0091] Denoising multipath data obtained from denoising reconstruction Represented as:

[0092]

[0093] In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal.

[0094] Step 4-3: Calculate the multipath continuous data x(t) and the denoised multipath data. The similarity features are used as multipath similarity feature data.

[0095] Calculate multipath continuous data x(t) and denoised multipath data Similarity metrics are introduced, including Cosine Similarity (CS), Euclidean Similarity (SES), Pearson Correlation Coefficient (PCC), Spearman Correlation Coefficient (SCC), Manhattan Distance, and Chebyshev Distance. The similarity features of multipath data are represented as follows:

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] In the formula: A and B represent the multipath continuous data x(t) and the denoised multipath data, respectively. and Let A and B be the ranks, respectively.

[0103] Step 5: Construct a deep learning NLOS recognition model based on dual-input feature fusion using multipath information.

[0104] The deep learning NLOS recognition method applicable to complex environments in multiple scenarios uses a network architecture diagram of a dual-input feature fusion deep learning NLOS recognition model based on multipath information, as shown in the figure below. Figure 2 As shown, for the delay power data TDP(x) and multipath similarity feature data of multipath discrete data, a multipath data feature extraction module and a multipath similarity data extraction module are constructed to extract features of multipath information. The multipath data features and multipath similarity features are fused to form a dual-channel multipath feature fusion NLOS classification module. The multipath data feature extraction module, the multipath similarity data extraction module, and the multipath feature fusion NLOS classification module together form a dual-input feature fusion deep learning NLOS recognition model based on multipath information.

[0105] The multipath data feature extraction module in this embodiment is based on a convolutional neural network model incorporating an attention mechanism (Attentive CNN), and its parameter settings are shown in Table 1 below. For the input multipath delay power data, the multipath data feature extraction module utilizes convolutional layers to extract local features and enhances the perception of key features by introducing an attention mechanism.

[0106] Table 1 Parameter settings for the multipath data feature extraction module

[0107]

[0108] The multipath similarity feature extraction module in this embodiment is based on a multilayer perceptron (MLP) model, and its parameter settings are shown in Table 2 below. For the similarity relationships between multipath data, the multipath similarity feature extraction module uses a fully connected structure to extract similarity features.

[0109] Table 2 Parameter settings for the multipath similarity feature extraction module

[0110]

[0111] The multipath feature fusion NLOS classification module in this embodiment is based on a deep neural network (DNN) model, and its parameter settings are shown in Table 3 below. After extracting features from the two types of multipath information, the multipath feature fusion NLOS classification module uses a feature fusion strategy to integrate the two types of features in a dual-channel manner. The fused features are then fed into the DNN model for NLOS classification. By utilizing the strong expressive power of the DNN, the fused features are classified, thereby achieving high-precision classification of LOS and NLOS.

[0112] Table 3 Parameter settings for the multipath feature fusion NLOS classification module

[0113]

[0114]

[0115] The data processing and algorithm implementation described above in this embodiment are performed on a computer with a Win10 (x64) operating system. This computer is equipped with an i5-12600KF CPU (3.70GHz), 32.00GB RAM, and an NVIDIA RTX 4070 SUPER (12GB). Furthermore, the deep learning and machine learning models in this embodiment are implemented based on the open-source machine learning library PyTorch.

[0116] Step 6 involves transforming all channel impulse response data from the multi-scenario public dataset in Step 1 to obtain multipath delay power data TDP(x) and multipath similarity feature data including CS, SES, PCC, SCC, HD, Manhattan Distance, and Chebyshev Distance. A new sample dataset is then constructed, and data preprocessing is performed based on the new sample dataset to set up the training set, validation set, and test set.

[0117] In this example, 42,000 sets of sample data are updated to form a new sample dataset of 42,000 sets. Based on this new dataset, the ratio of training set, validation set, and test set is set to 7:1:2. 29,400 sets of samples are randomly selected from this dataset using a random seed as the training set, 4,200 sets as the validation set, and 8,400 sets as the test set. Randomly selecting these samples helps prevent overfitting of the model.

[0118] Step 7: Input the preprocessed training data into the dual-input feature fusion deep learning NLOS recognition model based on multipath information, train and validate the model, and stop training when the model converges to complete the NLOS classification in the UWB positioning system.

[0119] The embodiment uses a dual-input feature fusion deep learning NLOS recognition model based on multipath information. The loss function is cross-entropy, the optimizer is Adam, the activation function for the classification layer is Softmax, and the dropout probability for all dropout layers is set to 0.5. The training batch size is set to 200 and the epoch is set to 50. After inputting 29,400 sets of training data into the model, features are first extracted and fused through forward propagation. Then, the loss between the predicted and true values ​​is calculated, and the gradient is calculated and parameters are updated through backpropagation. An early stopping mechanism is implemented during training. Training is stopped when the accuracy on the validation set does not significantly improve over multiple training cycles to prevent overfitting. The model with the best performance on the validation set is saved for testing, thus completing the training of the NLOS classification model.

[0120] To verify the recognition accuracy of this deep learning NLOS recognition method applicable to complex environments in multiple scenarios, its classification performance was compared with that of Support Vector Machine (SVM) based on Radial Basis Function (RBF), Linear Support Vector Machine (Linear SVM) based on Linear Kernel, and Convolutional Bidirectional Long Short-Term Memory Network (CNN-BiLSTM) based on the original channel impulse response. The comparison results are shown in Table 4 below. Figure 3 As shown.

[0121] Table 4 Comparison of classification performance of different classification methods

[0122]

[0123] The proposed deep learning NLOS recognition method, applicable to complex environments in multiple scenarios, achieves an accuracy of 88.26%, a precision of 90.32%, a recall of 85.61%, and an F1 score of 87.90%. Accuracy is crucial for providing accurate positioning information for UWB positioning systems. As shown in Table 4, the accuracy of this deep learning NLOS recognition method, applicable to complex environments in multiple scenarios, is improved by 8.67 percentage points, 9.53 percentage points, and 4.51 percentage points compared to the radial basis function-based support vector machine (RBF SVM) method, the linear kernel-based support vector machine (Linear SVM) method, and the convolutional bidirectional long short-term memory network model based on the original channel impulse response (CNN-BiLSTM) method, respectively. This deep learning-based NLOS recognition method, applicable to complex environments in multiple scenarios, also shows significant improvements across other metrics. The accuracy reaches 88.26%, with precision and recall being very close at 90.32% and 85.61% respectively. This indicates that the method can accurately identify NLOS and LOS to the greatest extent possible without excessively missing NLSO samples. The F1 score reaches 87.90%, demonstrating a good balance between precision and recall. Overall, this deep learning-based NLOS recognition method performs optimally for NLOS recognition in the UWB system.

[0124] To address the issue of accurate identification of NLOS signals during UWB system positioning in complex multi-scene environments, the above-mentioned experimental simulations verify that the deep learning NLOS identification method applicable to complex multi-scene environments can solve the problems of low recognition capability, poor generalization accuracy, and poor recognition effect in complex scenes with limited data, thereby achieving high NLOS identification accuracy and effectively improving UWB positioning accuracy in complex multi-scene environments.

Claims

1. A deep learning-based NLOS recognition method applicable to complex environments in multiple scenarios, characterized in that, Specifically, the following steps are included: Step 1: Based on the publicly available datasets of multiple scenarios collected by the UWB positioning system, collect the channel pulse response data between anchor points and tags in the UWB positioning system, and label the LOS data tags and NLOS data tags. Step 2: Obtain CIR multipath discrete data h based on FPI. m (n); Step 3, based on the multipath discrete data h m (n) Obtain the multipath signal delay power data TDP(x); Step 4: First, process the multipath discrete data h m (n) Obtain multipath continuous data x(t) through linear interpolation, and then reconstruct the multipath continuous data x(t) into denoised multipath data through VMD. Then, the multipath continuous data x(t) and the denoised multipath data are calculated. The similarity features are used as multipath similarity feature data; Step 5: For the delay power data TDP(x) and multipath similarity feature data of multipath discrete data, a multipath data feature extraction module and a multipath similarity data extraction module are constructed to extract the features of multipath information. The multipath data features and multipath similarity features are fused to form a dual-channel multipath feature fusion NLOS classification module. The multipath data feature extraction module, the multipath similarity data extraction module and the multipath feature fusion NLOS classification module together form a dual-input feature fusion deep learning NLOS recognition model based on multipath information. Step 6 involves transforming all channel impulse response data from the multi-scenario public dataset in Step 1 to obtain multipath delay power data TDP(x) and multipath similarity feature data, constructing a new sample dataset, and then performing data preprocessing to set up training, validation, and test sets based on the new sample dataset. Step 7: Input the preprocessed training data into the dual-input feature fusion deep learning NLOS recognition model based on multipath information to train and validate the model. Stop training when the model converges to complete the NLOS classification in the UWB positioning system.

2. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, Step 2 is as follows: For the acquired channel impulse response data, the channel model of the acquisition device, defined by the PRF and CIR, is as follows: In the formula: t represents the timestamp of each value in CIR; S is the number of multipath components; a s It is the amplitude of the Sth multipath component; τ s δ(·) is the time delay of the S-th multipath component; δ(·) is the Dirac function; n(t) represents the additive white Gaussian noise present in the channel; The acquisition device represents the acquired CIR as a complex IQ sample with a value of 1 ns, and then obtains the corresponding discrete CIR, which is represented as: In the formula: I n Q n These represent the real and imaginary parts of the nth complex value of the channel impulse response, acquired in units of 1 ns; P is the number of sampling points for each CIR data point. The 150 channel impulse response data points, including the FPI and the first path signal, are taken as multipath data. The multipath channel impulse response can be expressed as: h m (n)=h(n),n0≤t<n0+150 Where: h m (n) represents the multipath discrete data of the channel impulse response; n0 is the first path index time value in ns; h(n) represents the discrete CIR.

3. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, Step 3 is as follows: The delay power data TDP(x) of multipath discrete CIR data is specifically defined as follows: In the formula: L is the sampling factor; x is the data interval based on the sampling factor; n0 corresponds to h m The initial time of (n).

4. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 3, characterized in that, By setting the sampling factor L = 5, the delay power data TDP(x) of the multipath discrete CIR data is obtained.

5. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, Step 4 is as follows: Step 4-1, Multipath Discrete Data h m (n) Obtain multipath continuous data x(t) through linear interpolation; Multipath continuous data x(t) is represented as: In the formula: It is an interval indicator function, when t∈[t n ,t n+1 The value is 1 when the condition is met, and 0 otherwise; t n and t n+1 t is the discrete multipath CIR time point; h is the interpolation time point; m (n) and h m (n+1) represents the discrete CIR signal values ​​adjacent to the corresponding discrete time points; h m (n) represents the multipath discrete data; N is the length of the discrete multipath data. Step 4-2: Introduce VMD to reconstruct and denoise the multipath continuous data x(t) to obtain denoised multipath data. VMD solves for each mode u using the following optimization method. k (t) and the corresponding center frequency w k : In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal; w k It is the center frequency of the k-th modal signal; Denoising multipath data obtained from denoising reconstruction Represented as: In the formula: K is the number of modes, i.e., the number of sub-signals in the decomposition; u k (t) is the k-th modal signal; Step 4-3: Calculate the multipath continuous data x(t) and the denoised multipath data. The similarity features are used as multipath similarity feature data; Calculate multipath continuous data x(t) and denoised multipath data The similarity metrics are introduced, including CS, SES, PCC, SCC, Manhattan Distance, and Chebyshev Distance. The similarity features of multipath data are represented as follows: In the formula: A and B represent the multipath continuous data x(t) and the denoised multipath data, respectively. and Let A and B be the ranks, respectively.

6. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 5, characterized in that, In Step 4-2, solve for each mode u k (t) and the corresponding center frequency w k When the modality count is K = 5.

7. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, Step 5 is as follows: The multipath data feature extraction module is based on the Attentive CNN model. For the input multipath delay power data, the multipath data feature extraction module uses convolutional layers to extract local features and enhances the ability to perceive key features by introducing an attention mechanism. The multipath similarity feature extraction module is based on the MLP model. For the similarity relationship between multipath data, the multipath similarity feature extraction module uses a fully connected structure to extract similarity features. The multipath feature fusion NLOS classification module is based on a DNN model. After extracting multipath data features and multipath similarity features, the multipath feature fusion NLOS classification module adopts a feature fusion strategy to integrate the two types of features in a dual-channel manner. The fused features are then fed into the DNN model for NLOS classification.

8. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, In Step 6, when performing data preprocessing based on the new sample dataset, the ratio of the training set, validation set, and test set is set to 7:1:

2.

9. The deep learning NLOS recognition method applicable to complex environments in multiple scenarios according to claim 1, characterized in that, In Step 7, when the preprocessed training data is input into the deep learning NLOS recognition model based on multipath information and dual-input feature fusion for training and validating the model, the loss function of the deep learning NLOS recognition model based on multipath information and dual-input feature fusion for training is set to cross-entropy, the optimizer is Adam, and the activation function of the classification layer is Softmax. After the sample data of the training set is input into the model, features are first extracted and feature fusion is completed through forward propagation. Then, the loss between the predicted value and the true value is calculated, and the gradient is calculated and the parameters are updated through backpropagation. Training stops when the accuracy of the validation set does not improve significantly within multiple training cycles, and the model with the best performance on the validation set is saved for testing, thus completing the model training.

Images