Static indoor visible light positioning method based on wavelet transform

Abstract

The invention discloses a static indoor visible light positioning method based on wavelet transformation, and relates to the technical field of visible light communication, and the method comprises the steps: carrying out the continuous wavelet transformation of a visible light signal through a complex Morlet wavelet, converting the visible light signal in a time domain into a time-frequency domain, and obtaining wavelet sub-signals of different time sequences and frequencies; converting the wavelet sub-signals with different time sequences and frequencies into a wavelet power spectrum image, and identifying LOS signals and NLOS signals in the wavelet power spectrum image by using CNN so as to determine the position of a stationary target; in the process, a visible light signal in a time domain is converted into a time-frequency domain by using a complex Morlet wavelet, in the conversion, the visible light signal on a time domain sequence is decomposed into a time domain sequence and a frequency domain sequence, multi-scale time domain and frequency domain features in the time domain sequence and the frequency domain sequence are captured, LOS signals and NLOS signals in the multi-scale time domain and frequency domain features are identified by using a CNN, and the visible light signal in the time domain and the NLOS signals in the multi-scale time domain and frequency domain features is identified. Therefore, the target can be accurately positioned in a complex indoor environment.

Description

Technical Field

[0001] The present invention relates to the field of visible light communication technology, and in particular to a static indoor visible light positioning method, device, equipment and medium based on wavelet transform. Background Art

[0002] Indoor positioning technology is an important research direction in the field of modern location services, and has significant application value in the fields of smart buildings, Internet of Things, warehouse management, and security monitoring. However, the traditional global navigation satellite system (GNSS) is difficult to achieve high-precision positioning in indoor environments due to signal attenuation and multipath effects. Therefore, positioning technologies based on wireless signals (such as Wi-Fi, Bluetooth, etc.) have gradually emerged, but such methods still have significant deficiencies in positioning accuracy and robustness in complex indoor environments.

[0003] As an emerging technology, Visible Light Positioning (VLP) uses the light signals emitted by LED lighting equipment as the positioning source and achieves high-precision estimation of position and direction by measuring the received signal strength (RSS). Compared with traditional wireless radio frequency positioning methods, VLP technology has the advantages of no electromagnetic interference and low deployment cost.

[0004] Current indoor visible light positioning (VLP) methods, including support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), Gaussian process (GP), random forest (RF), etc., mainly identify LOS (line-of-sight link) and NLOS (non-line-of-sight link) signals through time domain features, but there are still significant deficiencies in complex environments; current methods mostly rely on single static time domain analysis and angle domain recognition, which makes it difficult to fully capture the multi-scale time-frequency characteristics of the signal, making it difficult to identify LOS and NLOS signals under the influence of multipath effects, so that it is difficult to estimate the target position and direction by measuring the strength of the received signal, resulting in poor positioning accuracy of the target in complex indoor environments. Summary of the invention

[0005] The embodiment of the present invention provides a static indoor visible light positioning method based on wavelet transform, which can solve the problem in the prior art that it is difficult to fully capture the multi-scale time-frequency characteristics of the signal, making it difficult to identify LOS signals and NLOS signals under the influence of multipath effects, and even difficult to estimate the target position and direction by measuring the strength of the received signal.

[0006] The embodiment of the present invention provides a static indoor visible light positioning method based on wavelet transform, comprising the following steps:

[0007] Build an indoor visible light positioning VLP system, where the stationary target receives visible light signals and forms a visible light signal data set;

[0008] Morlet wavelet transform uses complex Morlet wavelet as mother wavelet basis function to perform continuous wavelet transform on visible light signal, so as to convert visible light signal in time domain into visible light signal in time-frequency domain, and obtain wavelet sub-signals with different time series and frequencies;

[0009] The wavelet sub-signals of different time series and frequencies are converted into wavelet power spectrum images. The CNN convolutional neural network is used to identify the LOS signal and NLOS signal in the wavelet power spectrum image, and the position of the stationary target is determined according to the LOS signal and NLOS signal.

[0010] Preferably, the construction of an indoor visible light positioning VLP system includes:

[0011] Build an indoor visible light positioning VLP system with the size of L×W×H, where L, W, and H represent the length, width, and height of the room respectively;

[0012] The indoor visible light positioning VLP system includes Nr APs installed on the ceiling of the room, each AP faces downward, and each AP is adjacent to an LED and a PD. The LED is used for both lighting and data transmission, and the PD is used for data reception.

[0013] The indoor visible light positioning VLP system also includes mobile terminals for visible light communication, which are randomly arranged in the room. The mobile terminals are equipped with Nt PDs, and each PD has an adjacent IR-LED for data reception and transmission;

[0014] In the indoor visible light positioning VLP system, intensity modulation direct detection IM / DD is used to confirm the target position and direction; the receiving signal model at PD is expressed as:

[0015] y=λHx+n

[0016] Where: λ represents the transmission coefficient, λ=TR P η; H represents the channel matrix; Indicates the N between AP and PD r ×1 noise vector.

[0017] Preferably, the step of obtaining wavelet sub-signals of different time series and frequencies includes:

[0018] After the LED on the ceiling emits a visible light signal, the PD on the stationary target receives the visible light signal and forms a visible light signal data set;

[0019] According to the selection conditions of the mother wavelet basis function: a non-orthogonal wavelet function is needed to obtain a smooth continuous wavelet amplitude, and a complex wavelet is needed to obtain the amplitude and phase information of the time series; the complex Morlet wavelet is selected as the mother wavelet basis function, and the visible light signal is subjected to continuous wavelet transform. By adjusting the scale factor and translation factor of the mother wavelet basis function, the visible light signal in the time domain is converted into the visible light signal in the time-frequency domain, and wavelet sub-signals of different time series and frequencies are obtained;

[0020] The mother wavelet basis function of the complex Morlet wavelet is expressed as:

[0021]

[0022] Where: ω0 = 2πf c , f c represents the center frequency of the wavelet; f b represents the bandwidth of the wavelet;

[0023] When the visible light signal is set to u(t), the wavelet transform is expressed as:

[0024]

[0025] in: represents the complex conjugate of the scaled and time-shifted basis function; a represents the scale; τ represents the moving time; τ controls the translation of the wavelet function; a controls the expansion and contraction of the wavelet function and affects the generated frequency components.

[0026] Preferably, the identifying of LOS signals and NLOS signals in the wavelet power spectrum image comprises:

[0027] The wavelet sub-signals with different time series and frequencies are converted into wavelet power spectrum images through MWT, and the wavelet power spectrum images are reshaped into 128×128 grayscale images;

[0028] Two convolutional layers and two pooling layers are set in the architecture of the CNN convolutional neural network and stacked into a CNN convolutional neural network; a 128×128 grayscale image is input into the CNN convolutional neural network, convolutional layer 1 uses a 9×9 kernel with a stride of 1 and a ReLU activation function to generate 64 feature maps, and pooling layer 1 uses a 2×2 maximum pooling to compress the size of the feature map to 60×60, convolutional layer 2 uses a 3×3 kernel with a stride of 1 and a ReLU activation function to generate 128 feature maps, and pooling layer 2 uses a 2×2 maximum pooling to compress the size of the feature map to 29×29;

[0029] Feature map generated by layer l-1 It is expressed as:

[0030]

[0031] Where: f represents the ReLU activation function, ReLU(x) = ln(1+e x );M j Represents the selection of the input feature map; w represents the convolution filter connecting the l-1th layer to the lth layer; represents the bias of neuron j in layer l;

[0032] The smoothing layer converts the feature map of pooling layer 2 into a one-dimensional vector, which is connected to two fully connected layers FCL, and the two fully connected layers FCL are connected to the output layer. Two neurons are set in the output layer to identify LOS and NLOS signals, and the SoftMax activation function is used to set the recognition results of LOS and NLOS signals to P k , P k It is expressed as:

[0033]

[0034] Where: k∈{0,1}; Z represents the output of the previous fully connected layer; (…) T represents the transpose operation; w represents the weight updated by back propagation.

[0035] Preferably, determining the position of the stationary target comprises:

[0036] When using intensity modulation direct detection IM / DD to confirm the target position and direction, the target position and direction are estimated by the signal strength of the receiving terminal, and the signal-to-noise ratio of the received signal is used as an indicator to measure the signal strength. The signal-to-noise ratio is expressed as:

[0037]

[0038] Where: P elec Indicates the electrical power of the transmitted signal; Related to the channel matrix H, the ρ vector depends on the position and direction, that is, (x, y, z, α, β, γ); (x, y, z) represents the position of the target in the coordinate system; (α, β, γ) represents the rotation angle of the target;

[0039] This correlation is then used to determine the instantaneous position and direction of the target.

[0040] The embodiment of the present invention further provides a static indoor visible light positioning device based on wavelet transform, comprising:

[0041] The acquisition module is used to build an indoor visible light positioning VLP system, where the stationary target receives the visible light signal and forms a visible light signal data set;

[0042] Identification module, Morlet wavelet transform uses complex Morlet wavelet as mother wavelet basis function to perform continuous wavelet transform on visible light signal, so as to convert visible light signal in time domain into visible light signal in time-frequency domain, and obtain wavelet sub-signals with different time series and frequencies;

[0043] The positioning module is used to convert wavelet sub-signals of different time series and frequencies into wavelet power spectrum images, use CNN convolutional neural network to identify LOS signals and NLOS signals in the wavelet power spectrum images, and determine the position of stationary targets based on LOS signals and NLOS signals.

[0044] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0045] The memory is used to store computer programs;

[0046] The processor is used to implement the steps of the above-mentioned static indoor visible light positioning method based on wavelet transform when executing the computer program stored in the memory.

[0047] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the steps of the static indoor visible light positioning method based on wavelet transform as described above are implemented.

[0048] The embodiment of the present invention provides a static indoor visible light positioning method based on wavelet transform. Compared with the prior art, the beneficial effects thereof are as follows:

[0049] The present invention uses complex Morlet wavelet as mother wavelet basis function through Morlet wavelet transform to perform continuous wavelet transform on visible light signal, so as to convert visible light signal in time domain into visible light signal in time-frequency domain, and obtain wavelet sub-signals with different time series and frequencies; converts wavelet sub-signals with different time series and frequencies into wavelet power spectrum image, and uses CNN convolution neural network to identify LOS signal and NLOS signal in wavelet power spectrum image, so as to determine the position of stationary target; the process uses complex Morlet wavelet in Morlet wavelet transform to convert visible light signal in time domain into visible light signal in time-frequency domain, in the conversion, visible light signal in time domain sequence is decomposed into time domain sequence and frequency domain sequence, so as to capture multi-scale time domain and frequency domain features in time domain sequence and frequency domain sequence, and uses CNN convolution neural network to identify LOS signal and NLOS signal in multi-scale time domain and frequency domain features, so as to obtain the intensity of received signal through accurate LOS signal and NLOS signal, so as to realize accurate positioning of target in complex indoor environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of the overall process of a static indoor visible light positioning method based on wavelet transform provided by an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a typical visible light positioning system of a static indoor visible light positioning method based on wavelet transform provided in an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of a test framework of an MWT-CNN model of a static indoor visible light positioning method based on wavelet transform provided in an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of the CNN architecture in the MWT-CNN of a static indoor visible light positioning method based on wavelet transform provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0055] See also Figure 1 The embodiment of the present invention provides a static indoor visible light positioning method based on wavelet transform, specifically a LOS / NLOS identification method based on Morlet wavelet transform and convolutional neural network (MWT-CNN), aiming to improve the accuracy and robustness of indoor visible light communication (VLC) positioning system; the bandwidth of visible light signal is usually between 400THz (wavelength 780nm) and 800THz (wavelength 375nm), and the signal will experience frequency selective fading due to the multipath effect; therefore, it is of great significance to analyze the visible light signal in the time-frequency domain to identify LOS (line-of-sight link) and NLOS (non-line-of-sight link) signals.

[0056] Different from the traditional Fourier transform, the wavelet transform (WT) is a spectrum analysis method that can perform positioning in both the time domain and the frequency domain, and has multi-scale analysis capabilities. Therefore, the present invention uses the Morlet wavelet transform to process the received visible light signal, convert it into a time-frequency spectrum, and extract the time-frequency domain features of the signal from it. This characteristic of the wavelet transform makes it possible to effectively capture the time-varying characteristics of the signal and extract multi-scale key feature information. In order to further accurately extract features from the time-frequency spectrum and perform efficient classification, the present invention uses a convolutional neural network (CNN) as a deep learning module. CNN can effectively identify and distinguish LOS and NLOS signals in different environments through deep feature learning, thereby greatly improving the accuracy and robustness of the positioning system in complex environments.

[0057] Specifically:

[0058] 1. Establishment of positioning theory and data set.

[0059] 1. Environment settings.

[0060] like Figure 2 The indoor VLP system shown consists of a room with dimensions of L×W×H, where L, W and H represent the length, width and height of the room, respectively; the system is equipped with Nr APs installed on the ceiling of the room, each AP faces downward and is adjacent to 1 LED and 1 PD, the LED is used for both lighting and data transmission, and the PD is used for data reception; in addition, mobile terminals supporting visible light communication are randomly located in the room and are equipped with Nt PDs, each PD has an adjacent IR-LED for data reception and transmission, wherein the communication between the AP and the terminal is bidirectional; specifically, in the downlink, the AP uses the visible spectrum to transmit information, and the terminal receives this information through its PD, while in the uplink, the terminal's IR-LED uses the infrared spectrum to transmit information to the AP, and the transmitted signal is detected by the PD; in this mechanism, there is no interference between the downlink and uplink transmissions, and the two stages can occur simultaneously.

[0061] 2. Transmission model.

[0062] In the VLP system, intensity modulation direct detection (IM / DD) is used, and the received signal model at the PD is described as:

[0063] y=λHx+n

[0064] Where: λ represents the transmission coefficient, defined as λ = TR P η; H represents the channel matrix; Indicates the N between AP and PD r×1 noise vector, where the noise includes all possible noises, such as shot noise and thermal noise, and is assumed to be a real-valued additive white Gaussian and is independent of the transmitted signal; the variance of the noise is equal to Where N0 is the one-sided power spectral density of the noise and B is the bandwidth.

[0065] 3. Received signal strength analysis.

[0066] The research objective of the present invention is to estimate the position and direction of a receiving terminal by the signal strength of the receiving terminal. Here, the signal-to-noise ratio of the received signal is used as a measurement indicator; it is defined as follows:

[0067]

[0068] Where: P elec Indicates the electrical power of the transmitted signal; Definition It is related to the channel matrix H, and the ρ vector depends on the position and direction, that is, (x, y, z, α, β, γ). Such correlation is used to estimate the instantaneous position and direction.

[0069] 4. Dataset generation.

[0070] Prerequisites: The target is stationary in the indoor environment, and the initial position of the target is random; the user is uniformly located in the indoor environment, so the probability density function (PDF) of the terminal's 3D position is expressed by the following formula:

[0071]

[0072] Where: [0,H device ] indicates the maximum height of the terminal in an indoor environment.

[0073] The above describes the probability distribution function of the three-dimensional position. The rotation angles (α, β, γ) need to be given by statistical distribution laws. The rotation angles α, β, and γ all follow the truncated Laplace distribution, and their mean and standard deviation are:

[0074] (μ α ,σ α )=(Ω-90°,3.67°)

[0075] (μ β ,σ β )=(40.78°,2.39°)

[0076] (μ γ ,σ γ )=(-0.84°,2.21°)

[0077] Where: Ω represents the direction of movement. To be precise, Ω represents the direction of movement or the direction of movement, which is measured from the east direction in the earth's coordinate system, such as Figure 2 shown.

[0078] Assuming that the target dataset contains N data points, the process of generating the nth measurement-based data point, for n∈[[1,N]], is described in detail as:

[0079] ① Generate samples of 3D positions (x,y,z) using statistics.

[0080] ②Generate samples of motion direction angle Ω uniformly from [0°, 360°].

[0081] ③ Use the truncated Laplace distribution and its statistical norm to generate three direction angles (α, β, γ).

[0082] ④ Calculate the resulting channel matrix H.

[0083] ⑤From Uniformly generate random electrical emission power P elec ,in is the highest possible electrical transmission power.

[0084] ⑥ Calculate the corresponding SNR vector ρ.

[0085] ⑦Finally, the obtained SNR vector ρ is stored in the data set as a feature vector.

[0086] 2. Design based on wavelet transform and convolutional neural network.

[0087] 1. MWT-CNN framework.

[0088] like Figure 3 As shown in the figure, MWT-CNN is divided into an offline training process and an online testing process. In the offline training process, the labeled visible signal data is first transformed by a complex Morlet wavelet transform and unified into a 128×128 grayscale time-frequency spectrum. Further feature extraction is performed offline to train the CNN model. In the online testing process, the real-time visible light signal data is converted into a grayscale image through MWT and tested with the trained CNN model to obtain the predicted label. In theory, the more different categories of features used by the classifier, the higher the accuracy. Therefore, MWT-CNN can achieve higher performance than traditional methods that only use time domain features.

[0089] 2.Morlet wavelet transform.

[0090] Wavelet transform can simultaneously characterize the local characteristics of the signal in the time domain and frequency domain; wavelet analysis decomposes the signal into a series of wavelet functions, which are derived from the mother wavelet function by translation and scaling; this irregular wavelet is used to approximate the sharply changing part of the non-steady-state signal, or approximate the discrete and discontinuous signal with local characteristics, so as to truly reflect the change of the original signal on a certain time scale; the selection of the mother wavelet is particularly important, and a non-orthogonal wavelet function is required to obtain a smooth continuous wavelet amplitude; in addition, a complex wavelet is required to obtain the amplitude and phase information of the time series; the Morlet wavelet is not only non-orthogonal, but also a Gaussian-regulated complex exponential wavelet; in addition, the Morlet wavelet has an appropriate balance between time and frequency positioning; therefore, the present invention selects the complex Morlet wavelet as the mother wavelet; the wavelet basis function of the Morlet wavelet can be expressed as:

[0091]

[0092] Where: ω0 = 2πf c , f c represents the center frequency of the wavelet; f b Represents the bandwidth of the wavelet; the settings of these two parameters will affect the time resolution and frequency resolution and need to be adjusted according to the specific task.

[0093] When the signal u(t) is given, the wavelet transform is expressed as:

[0094]

[0095] in: represents the complex conjugate of the scaling and time-shifting basis function; a represents the scale; τ represents the moving time; τ controls the translation of the wavelet function, so that the wavelet can realize traversal analysis along the time axis; a controls the expansion and contraction of the wavelet function and affects the generated frequency components; through MWT, an image called the wavelet power spectrum can be obtained. The wavelet power spectrum shows the fluctuation characteristics of the time series at a certain scale and its changes over time. The wavelet power spectrum is normalized and reshaped into a 128×128 grayscale image as the input of the CNN.

[0096] 3. CNN architecture.

[0097] CNN is a multi-layer neural network with the characteristics of local connection and parameter sharing. Figure 4 shown.

[0098] The CNN input is a 128×128 grayscale image; during the training process, the input has a known label l(x), l(x)=0 indicates that the channel state is LOS, and 1 indicates NLOS; the convolution layer uses the convolution kernel to extract the features of each small part of the input and generate a feature map, and the pooling layer extracts the main features from the feature map and reduces its dimension; at the same time, the use of pooling reduces overfitting and the propagation of noise, and high-level features are extracted from the input by stacking the convolution layer and the pooling layer; in the CNN architecture designed by the present invention, the convolution layer 1 (Conv1) uses a 9×9 kernel with a stride of 1 and ReLU activation to generate 64 feature maps, and the pooling layer 1 uses 2×2 maximum pooling to compress the size of the feature map to 60×60, Conv2 uses a 3×3 kernel with a stride of 1 and ReLU activation to generate 128 feature maps, and the pooling layer 2 uses 2×2 maximum pooling to compress the size of the feature map to 29×29; in the forward propagation process of the convolution layer, the feature map generated by the l-1 layer It can be expressed as:

[0099]

[0100] Where: f represents the ReLU activation function, ReLU(x) = ln(1+e x );M j Represents the selection of the input feature map; w represents the convolution filter connecting the l-1th layer to the lth layer; represents the bias of neuron j in layer l.

[0101] The smoothing layer is used to convert the feature map of pooling layer 2 into a one-dimensional vector, which is connected to two fully connected layers (FCL) that summarize the extracted features. The number of neurons in both fully connected layers is set to 128.

[0102] FCL is connected to the output layer. In order to identify LOS / NLOS, the present invention sets two neurons in the output layer, namely H0 and H1, and applies the SoftMax activation function to interpret the result as probability Pk, which can be expressed as:

[0103]

[0104] Where: k∈{0,1}; Z represents the output of the previous fully connected layer; (…) T represents the transpose operation; w represents the weight updated by back propagation.

[0105] The decision rule for CNN to predict the label l(x) is expressed as:

[0106]

[0107] The present invention uses cross entropy loss to measure the training loss and updates the weights through back propagation; the cross entropy loss function is expressed as:

[0108]

[0109] Where: n represents the batch size; y i Represents sample y i The true category of p i,k Represents x i The probability of being predicted as class k.

[0110] The method based on Morlet wavelet transform and convolutional neural network (MWT-CNN) proposed in the present invention combines the time-frequency analysis capability of wavelet transform and the deep feature learning capability of CNN, and can accurately distinguish LOS and NLOS signals in complex indoor environments. The present invention combines the advantages of wavelet transform (Wavelet Transform) and convolutional neural network (Convolutional Neural Network, CNN), fully explores the potential features in the optical signal through a deep learning model, and proposes an efficient three-dimensional position and direction synchronization estimation scheme. The method aims to break through the limitations of traditional positioning methods, significantly improve the positioning accuracy and system robustness in complex environments, and lay the foundation for the wide application of visible light positioning technology in smart buildings, the Internet of Things and other related fields. The simulation results show that MWT-CNN can achieve high accuracy in office scenes, industrial scenes, residential scenes and experimental scenes, reaching 100%, 99.89%, 96.10% and 98.84% respectively. And for static scenes, MWT-CNN shows higher robustness.

[0111] Creative Aspects:

[0112] The present invention proposes a brand-new solution in the field of indoor visible light positioning technology, innovatively combining wavelet transform (Wavelet Transform) and convolutional neural network (CNN) to form a composite method called MWT-CNN; wavelet transform, as a multi-scale analysis tool, can effectively transform visible light signals from time domain to time-frequency domain, providing higher resolution and flexibility for signal feature extraction; in traditional methods, signals are usually processed in a single time domain or frequency domain, which makes it difficult to cope with the diversity and complexity of signals in complex dynamic environments; and the present invention, by combining convolutional neural network, further performs deep learning on the time-frequency domain features obtained from wavelet transform, which can effectively identify and distinguish different propagation modes of signals (such as LOS and NLOS).

[0113] The innovation of the present invention lies not only in the depth and breadth of signal feature extraction, but also in that it solves the bottleneck of traditional positioning technology in complex environments; especially in indoor environments with high multipath effects, occlusion and signal attenuation, the present invention performs multi-scale analysis of signals through wavelet transform, extracts important features at different scales, and combines the powerful learning ability of convolutional neural networks to further extract useful pattern information; compared with traditional positioning methods based on spectrum or time domain, MWT-CNN can more effectively process complex signal fluctuations and accurately identify multiple propagation paths, thereby improving the robustness and accuracy of the positioning system in dynamic and changing environments.

[0114] Novelty:

[0115] The core innovation of the present invention is that it combines Morlet wavelet transform with convolutional neural network (CNN) for the first time, and proposes a new method for LOS / NLOS recognition in complex indoor visible light positioning environments; the proposed method not only solves the limitations of traditional indoor positioning technology in the face of multipath effects, non-line-of-sight propagation (NLOS) and dynamic environments, but also achieves important breakthroughs in theory and application. Specifically:

[0116] ① Innovative combination of wavelet transform and convolutional neural network: The originality of the present invention lies in the effective combination of wavelet transform and convolutional neural network (CNN), and the composite method of MWT-CNN (Morlet wavelet transform and convolutional neural network) is proposed; wavelet transform, as a multi-scale analysis method, can effectively capture the time-frequency characteristics in the signal, especially has unique advantages in processing multipath signals with frequency selective fading; on the other hand, convolutional neural network can extract high-level abstract information from these time-frequency domain features through deep learning, and complete accurate classification and recognition; this combination not only optimizes the signal processing flow, but also makes up for the defect of insufficient positioning accuracy of traditional methods in multipath and non-line-of-sight environments; compared with the previous single signal analysis method based on frequency domain or time domain, the method of the present invention significantly improves the accuracy and robustness of the indoor visible light positioning system under dual processing in time and frequency domain, especially the positioning stability in complex environments.

[0117] ② Novel application of time-frequency domain signal analysis: The present invention innovatively applies time-frequency domain analysis technology (such as wavelet transform) to the processing of visible light signals, and combines it with convolutional neural networks, thereby solving the challenges faced by traditional methods when facing multipath effects and signal attenuation in complex environments; unlike traditional frequency domain analysis methods such as Fourier transform, wavelet transform can perform positioning in both time domain and frequency domain at the same time, capturing the time-varying characteristics of the signal; therefore, in dynamic environments, especially in indoor scenes with occlusion, reflection and multipath propagation, wavelet transform can provide richer signal information and enhance the adaptability and accuracy of the positioning system; by combining this advanced time-frequency domain signal processing method with the deep learning capability of convolutional neural networks, the present invention can efficiently and accurately identify and distinguish LOS and NLOS signals in indoor environments, thereby greatly improving the reliability of visible light positioning technology.

[0118] Practicality:

[0119] The indoor visible light positioning method based on the combination of Morlet wavelet transform and convolutional neural network (MWT-CNN) proposed in the present invention has extremely high practicability in practical applications, especially in positioning tasks in complex dynamic environments, can significantly improve positioning accuracy, stability and robustness, and has broad application prospects; specifically, it is embodied in:

[0120] ① High-precision positioning in complex environments: In complex indoor environments, optical signals are often affected by factors such as multipath propagation, non-line-of-sight propagation (NLOS) and dynamic changes. These factors make it difficult for traditional positioning methods to meet high-precision requirements. The present invention introduces the multi-scale analysis capability of wavelet transform and the deep feature learning of convolutional neural networks. The present invention can effectively identify and distinguish different propagation paths (such as LOS and NLOS), and can provide high-precision dynamic positioning even in complex indoor environments. This advantage makes the present invention particularly suitable for environments that require high-density target positioning, such as office spaces, industrial plants, shopping centers, warehouses, etc.

[0121] ② High robustness and environmental adaptability: The present invention can effectively deal with the problems caused by fluctuations, occlusions and multipath effects of optical signals in indoor environments; through comprehensive processing in the time and frequency domains, the method can improve the system's adaptability to dynamic changes; in application scenarios such as real-time positioning of high-speed moving objects and personnel tracking, the MWT-CNN method can maintain high stability and reliability, solving the positioning error and instability problems of traditional methods in complex scenarios; especially in the fields of industrial automation, smart home, robot navigation, etc., it can provide accurate real-time positioning support for various dynamic applications.

[0122] ③ Applicable to a variety of indoor scenarios: Through simulation tests, the present invention has proved its excellent performance in a variety of typical indoor scenarios, including office environments, industrial scenarios, residences and laboratories; the method of the present invention is not only suitable for static target positioning, but can also effectively cope with the positioning and posture tracking of dynamic targets, and has good cross-scenario adaptability; in addition, the MWT-CNN method does not rely on expensive hardware equipment, and can be easily integrated on the basis of existing visible light communication (VLC) systems, providing an efficient and economical solution for existing indoor positioning systems; especially in large indoor environments, the system can be quickly deployed and flexibly adjusted to meet different application requirements.

[0123] The purpose of the present invention is to provide an indoor visible light positioning method based on wavelet transform and convolutional neural network (MWT-CNN), aiming to solve the problems of low positioning accuracy, poor robustness and inability to effectively process multipath effects and NLOS signals in the existing technology in complex indoor environments; by innovatively combining the multi-scale time-frequency analysis of wavelet transform with the deep feature learning of convolutional neural network, the present invention can accurately identify and distinguish LOS and NLOS signals, and improve the adaptability of the system in dynamic and complex environments; specifically, the present invention breaks through the limitations of traditional methods by introducing time-frequency domain signal processing technology, and can achieve high-precision, high-robustness real-time positioning and posture tracking in high-density and complex indoor environments. The purpose of the present invention is to provide a more accurate, stable and highly real-time indoor visible light positioning technology to meet the high requirements for positioning accuracy and real-time performance in practical applications.

[0124] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A static indoor visible light positioning method based on wavelet transform, characterized in that: The following steps are involved: Build an indoor visible light positioning VLP system, where the stationary target receives visible light signals and forms a visible light signal data set; Morlet wavelet transform uses complex Morlet wavelet as mother wavelet basis function to perform continuous wavelet transform on visible light signal, so as to convert visible light signal in time domain into visible light signal in time-frequency domain, and obtain wavelet sub-signals with different time series and frequencies; The wavelet sub-signals of different time series and frequencies are converted into wavelet power spectrum images. The CNN convolutional neural network is used to identify the LOS signal and NLOS signal in the wavelet power spectrum image, and the position of the stationary target is determined according to the LOS signal and NLOS signal.

2. The static indoor visible light positioning method based on wavelet transform according to claim 1, characterized in that: The construction of an indoor visible light positioning VLP system comprises: Build an indoor visible light positioning VLP system with the size of L×W×H, where L, W, and H represent the length, width, and height of the room respectively; The indoor visible light positioning VLP system includes Nr APs installed on the ceiling of the room, each AP faces downward, and each AP is adjacent to an LED and a PD. The LED is used for both lighting and data transmission, and the PD is used for data reception. The indoor visible light positioning VLP system also includes mobile terminals for visible light communication, which are randomly arranged in the room. The mobile terminals are equipped with Nt PDs, and each PD has an adjacent IR-LED for data reception and transmission; In the indoor visible light positioning VLP system, intensity modulation direct detection IM / DD is used to confirm the target position and direction; the receiving signal model at PD is expressed as: y=λHx+n Where: λ represents the transmission coefficient, λ=TR P η; H represents the channel matrix; Indicates the N between AP and PD r ×1 noise vector.

3. The static indoor visible light positioning method based on wavelet transform according to claim 2 is characterized in that: The step of obtaining wavelet signals of different time series and frequencies includes: After the LED on the ceiling emits a visible light signal, the PD on the stationary target receives the visible light signal and forms a visible light signal data set; According to the selection conditions of the mother wavelet basis function: a non-orthogonal wavelet function is needed to obtain a smooth continuous wavelet amplitude, and a complex wavelet is needed to obtain the amplitude and phase information of the time series; the complex Morlet wavelet is selected as the mother wavelet basis function, and the visible light signal is subjected to continuous wavelet transform. By adjusting the scale factor and translation factor of the mother wavelet basis function, the visible light signal in the time domain is converted into the visible light signal in the time-frequency domain, and wavelet sub-signals of different time series and frequencies are obtained; The mother wavelet basis function of the complex Morlet wavelet is expressed as: Where: ω0 = 2πf c , f c represents the center frequency of the wavelet; f b represents the bandwidth of the wavelet; When the visible light signal is set to u(t), the wavelet transform is expressed as: in: represents the complex conjugate of the scaled and time-shifted basis function; a represents the scale; τ represents the moving time; τ controls the translation of the wavelet function; a controls the expansion and contraction of the wavelet function and affects the generated frequency components.

4. The static indoor visible light positioning method based on wavelet transform according to claim 1, characterized in that: The identifying of LOS signals and NLOS signals in the wavelet power spectrum image comprises: The wavelet sub-signals with different time series and frequencies are converted into wavelet power spectrum images through MWT, and the wavelet power spectrum images are reshaped into 128×128 grayscale images; Two convolutional layers and two pooling layers are set in the architecture of the CNN convolutional neural network and stacked into a CNN convolutional neural network; a 128×128 grayscale image is input into the CNN convolutional neural network, convolutional layer 1 uses a 9×9 kernel with a stride of 1 and a ReLU activation function to generate 64 feature maps, and pooling layer 1 uses a 2×2 maximum pooling to compress the size of the feature map to 60×60, convolutional layer 2 uses a 3×3 kernel with a stride of 1 and a ReLU activation function to generate 128 feature maps, and pooling layer 2 uses a 2×2 maximum pooling to compress the size of the feature map to 29×29; Feature map generated by layer l-1 It is expressed as: Where: f represents the ReLU activation function, ReLU(x) = ln(1+e x );M j Represents the selection of the input feature map; w represents the convolution filter connecting the l-1th layer to the lth layer; represents the bias of neuron j in layer l; The smoothing layer converts the feature map of pooling layer 2 into a one-dimensional vector, which is connected to two fully connected layers FCL, and the two fully connected layers FCL are connected to the output layer. Two neurons are set in the output layer to identify LOS and NLOS signals, and the SoftMax activation function is used to set the recognition results of LOS and NLOS signals to P k , P k It is expressed as: Where: k∈{0,1}; Z represents the output of the previous fully connected layer; (…) T represents the transpose operation; w represents the weight updated by back propagation.

5. The static indoor visible light positioning method based on wavelet transform according to claim 2 is characterized in that: Determining the position of the stationary target includes: When using intensity modulation direct detection IM / DD to confirm the target position and direction, the target position and direction are estimated by the signal strength of the receiving terminal, and the signal-to-noise ratio of the received signal is used as an indicator to measure the signal strength. The signal-to-noise ratio is expressed as: Where: P elec Indicates the electrical power of the transmitted signal; Related to the channel matrix H, the ρ vector depends on the position and direction, that is, (x, y, z, α, β, γ); (x, y, z) represents the position of the target in the coordinate system; (α, β, γ) represents the rotation angle of the target; This correlation is then used to determine the instantaneous position and direction of the target.

6. A static indoor visible light positioning device based on wavelet transform, characterized in that: include: The acquisition module is used to build an indoor visible light positioning VLP system, where the stationary target receives the visible light signal and forms a visible light signal data set; Identification module, Morlet wavelet transform uses complex Morlet wavelet as mother wavelet basis function to perform continuous wavelet transform on visible light signal, so as to convert visible light signal in time domain into visible light signal in time-frequency domain, and obtain wavelet sub-signals with different time series and frequencies; The positioning module is used to convert wavelet sub-signals of different time series and frequencies into wavelet power spectrum images, use CNN convolutional neural network to identify LOS signals and NLOS signals in the wavelet power spectrum images, and determine the position of stationary targets based on LOS signals and NLOS signals.

7. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of a static indoor visible light positioning method based on wavelet transform as described in any one of claims 1 to 5 when executing the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a static indoor visible light positioning method based on wavelet transform as claimed in any one of claims 1 to 5.

Images