Three-dimensional space positioning method and system based on multi-source opportunity signal deep learning

Abstract

The invention relates to a three-dimensional space positioning method and system based on multi-source opportunity signal deep learning, belongs to the field of navigation, and solves the problems that a traditional method is low in positioning precision and lacks three-dimensional positioning. Comprising the following steps: establishing a three-dimensional space coordinate system of a to-be-positioned area to obtain a plurality of three-dimensional coordinate points, simultaneously acquiring multiple types of opportunity signals from the plurality of three-dimensional coordinate points, preprocessing the multiple types of opportunity signals to obtain sample data, and forming a multiple types of opportunity signal sample set with corresponding sample labels; constructing a three-dimensional positioning deep neural network model, and training the three-dimensional positioning deep neural network model by using each type of opportunity signal sample set to obtain the three-dimensional positioning deep neural network model trained by various types of opportunity signals; and after a plurality of opportunity signals are acquired in real time at a to-be-positioned point for preprocessing, the trained three-dimensional positioning deep neural network model of the corresponding category is input to obtain a plurality of three-dimensional coordinate point positions corresponding to the opportunity signals, and the three-dimensional coordinate points are averaged to obtain a three-dimensional space positioning result.

Description

Technical Field

[0001] The present invention belongs to the field of navigation technology, and in particular relates to a three-dimensional space positioning method and system based on deep learning of multi-source opportunity signals. Background Art

[0002] Positioning is crucial in modern technology and everyday life, supporting autonomous navigation, improving safety, optimizing logistics and transportation, enhancing user experience, promoting the application of geographic information systems, and advancing scientific research and precision agriculture. Accurate positioning data is the foundation of technologies such as drones, self-driving cars, and smartphones, improving efficiency and decision-making, driving the development of various industries, and improving people's quality of life.

[0003] Global Navigation Satellite Systems (GNSS), such as the Global Positioning System (GPS), perform well in outdoor environments with strong signals and provide high-precision positioning services. However, GNSS systems also have significant limitations. First, satellite positioning systems rely on precise signal synchronization to determine position by measuring signal propagation time. Time errors in the microsecond range can lead to significant positioning deviations, so highly accurate time synchronization is required. Second, the accuracy of these systems is significantly affected in complex environments (such as indoors or in densely populated cities). In urban environments, signals are susceptible to electromagnetic interference, signal blockage, and obstruction by buildings, resulting in signal attenuation, reduced positioning accuracy and stability, which often fails to meet users' needs for accurate and real-time positioning services.

[0004] Opportunity signal positioning technology utilizes existing signals in the electromagnetic environment as auxiliary positioning signal sources, providing relatively reliable positioning information when traditional positioning sources are interfered with. Existing signal-of-opportunity positioning methods, such as triangulation and signal strength measurement, rely on signal propagation characteristics and receiver measurement data to infer the location of the signal source. However, these methods are susceptible to the complexity of signal propagation paths, multipath effects, and environmental interference, resulting in insufficient positioning accuracy and robustness.

[0005] In recent years, with the development of deep learning technology, researchers have begun applying deep learning to positioning methods based on opportunity signals. Deep learning can automatically learn rich features from various signal data and complete positioning tasks, demonstrating superior performance compared to traditional methods in many scenarios. Summary of the Invention

[0006] In view of the above analysis, the embodiments of the present invention aim to provide a three-dimensional spatial positioning method and system based on deep learning of multi-source opportunity signals, so as to solve the technical problems of low positioning accuracy and lack of three-dimensional positioning of traditional opportunity signal methods.

[0007] The purpose of the present invention is mainly achieved through the following technical solutions:

[0008] The present invention provides a three-dimensional space positioning method based on deep learning of multi-source opportunity signals, comprising the following steps:

[0009] Establishing a three-dimensional spatial coordinate system for the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collecting multiple types of opportunity signals from each three-dimensional coordinate point, preprocessing the multiple types of opportunity signals to obtain sample data, and forming a multi-type opportunity signal sample set with corresponding sample labels; wherein the sample label is the three-dimensional coordinate point of the collected signal;

[0010] Construct a 3D positioning deep neural network model, and use each type of opportunity signal sample set to train the 3D positioning deep neural network model to obtain a 3D positioning deep neural network model trained with each type of opportunity signal;

[0011] After preprocessing multiple opportunity signals in real time at the point to be positioned, the trained three-dimensional positioning deep neural network model of the corresponding category is input to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and the multiple three-dimensional coordinate point positions are averaged to obtain the final three-dimensional space positioning result.

[0012] Furthermore, the method of establishing a three-dimensional space coordinate system is as follows:

[0013] The airspace to be located is represented as a three-dimensional space, a spatial rectangular coordinate system is established, the east-west direction and the north-south direction are set as the x-axis and the y-axis respectively, the vertical height is set as the z-axis, and the multiple three-dimensional coordinate points are obtained by dividing according to predetermined intervals.

[0014] Furthermore, the steps of simultaneously collecting multiple types of opportunity signals from each three-dimensional coordinate point are as follows:

[0015] Use the UAV platform to fly to the corresponding three-dimensional coordinate point, and use the antenna carried by the platform to collect opportunity signals;

[0016] The center frequency of the acquisition is set to the frequency band where the opportunity signal is located. The acquisition bandwidth is set based on the bandwidth of the frequency band where the opportunity signal is located. The acquisition time of each coordinate point is unified, and multiple types of opportunity signals are collected multiple times at the same coordinate point.

[0017] Furthermore, the three-dimensional positioning deep neural network model includes, in sequence, an input layer, an initial convolutional layer, a first pooling layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a third transition layer, a fourth dense block, a global average pooling layer, a fully connected layer, and an output layer;

[0018] The input layer is used to receive sample data;

[0019] The initial convolutional layer is used to extract the original time domain features of the sample data;

[0020] The first pooling layer is used to downsample the original time domain features to obtain first signal features;

[0021] The first dense block, the first transition layer, the second dense block, the second transition layer, the third dense block, the third transition layer, and the fourth dense block are used to perform channel compression and feature downsampling on the first opportunity signal feature in sequence to obtain the second signal feature; wherein the first, second, and third transition layers all include 1×1 convolution operations and average pooling layers, the convolution operation is used to reduce the number of feature channels, and the average pooling is used to further reduce the size of the feature map.

[0022] Furthermore, the global average pooling layer is used to perform a global average operation on the second signal feature in the time dimension to obtain a third signal feature;

[0023] The fully connected layer is used to convert the third signal feature into a three-dimensional coordinate prediction result through a linear transformation;

[0024] The output layer is used to output the three-dimensional coordinate prediction results.

[0025] Furthermore, a separate 3D localization deep neural network model is trained for each type of opportunity signal, using the following process:

[0026] Using the training set of each type of opportunity signal sample set, a 3D positioning deep neural network model is trained;

[0027] During the training process, the AdamW optimizer is used, combined with adaptive learning rate and weight decay; the mean square error is used as the loss function;

[0028] When the accuracy on the validation set no longer improves after more than n consecutive rounds, training is stopped and the model parameters with the smallest loss on the validation set during training are saved as the trained 3D positioning deep neural network model.

[0029] Furthermore, the loss function is as follows:

[0030]

[0031] Among them, N is the number of samples in the sample set of this type of opportunity signal, (x i ,y i ,z i ) is the true three-dimensional coordinate point of the sample data, The 3D coordinate points predicted by the deep neural network model for 3D positioning.

[0032] Furthermore, the sample data obtained by preprocessing each type of opportunity signal and the corresponding sample labels are combined to form multiple opportunity signal sample sets, including:

[0033] Extracting the IQ sequence of each opportunity signal;

[0034] Performing signal interception on the IQ sequence according to a predetermined length m;

[0035] For each three-dimensional coordinate point of the acquisition location, the corresponding sample data covers various types of opportunity signals;

[0036] Each type of opportunity signal sample set is divided into a training set and a validation set according to a preset ratio.

[0037] Furthermore, categories of the opportunity signals include broadcast signals, mobile communication signals, satellite communication signals and Wi-Fi signals.

[0038] The present invention also discloses a three-dimensional spatial positioning system based on deep learning of multi-source opportunity signals, comprising a data acquisition and preprocessing module, a model training module and a real-time positioning module;

[0039] The data acquisition and preprocessing module is used to establish a three-dimensional spatial coordinate system of the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collect multiple types of opportunity signals from the multiple three-dimensional coordinate points, preprocess the multiple types of opportunity signals to obtain sample data, and form multiple types of opportunity signal sample sets with corresponding sample labels; wherein the sample labels are the three-dimensional coordinate points of the collected signals;

[0040] The model training module is used to build a three-dimensional positioning deep neural network model, and use each type of opportunity signal sample set to train the three-dimensional positioning deep neural network model to obtain a three-dimensional positioning deep neural network model trained with each type of opportunity signal;

[0041] The real-time positioning module obtains multiple opportunity signals in real time at the point to be positioned and performs preprocessing, then inputs a trained three-dimensional positioning deep neural network model of the corresponding category to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and averages the multiple three-dimensional coordinate point positions to obtain the final three-dimensional space positioning result.

[0042] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0043] (1) Aiming at the problem of three-dimensional spatial positioning, the present invention proposes a deep learning three-dimensional spatial positioning method based on multi-source opportunity signals, which is suitable for autonomous positioning in a specific three-dimensional space;

[0044] (2) The present invention utilizes opportunity signals as a positioning source, which can provide reliable positioning information when traditional positioning sources are interfered with, while reducing the system's dependence on GPS and other external signals, thereby significantly improving the robustness and availability of the positioning system;

[0045] (3) The present invention combines multiple opportunistic signal sources (such as broadcast signals, mobile communication signals, satellite communication signals, etc.), can fully utilize the characteristics of different signals, and improve positioning accuracy and robustness;

[0046] (4) The present invention directly implements positioning through deep learning based on IQ signals, without the need to extract features such as channel state information, and has no special requirements for the signal protocol of the opportunity signal source, thus having wider applicability;

[0047] (5) With the help of the powerful feature extraction capability of deep learning, the present invention can automatically learn key features from complex signal data, capture environmental changes and signal characteristics more comprehensively and accurately, and thus improve positioning accuracy and stability in complex environments.

[0048] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like parts throughout the drawings.

[0050] Figure 1 This is a flow chart of a three-dimensional spatial positioning method based on deep learning of multi-source opportunity signals in an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of constructing a positioning three-dimensional model in an embodiment of the present invention;

[0052] Figure 3 Schematic diagram of the specific network structure of the three-dimensional positioning deep neural network model constructed in an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of a three-dimensional spatial positioning system module based on deep learning of multi-source opportunity signals in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0055] Current deep learning-based signal-of-opportunity localization methods are primarily designed for specific signals of opportunity and are difficult to generalize to other sources of signals of opportunity. Therefore, further breakthroughs are needed in leveraging multiple signal-of-opportunity sources for deep learning to further improve positioning accuracy. Furthermore, current research has largely focused on two-dimensional positioning, with little involvement in three-dimensional positioning, necessitating urgent innovation.

[0056] Example 1:

[0057] Opportunity signals in this invention refer to existing electromagnetic signals in the environment that are not specifically designed for three-dimensional positioning. These include, but are not limited to, broadcast signals, mobile communication signals, satellite communication signals, and Wi-Fi signals. While these signals serve their original functions, such as communication or broadcasting, their propagation characteristics (such as signal strength, phase, and multipath) can be collected and used to assist in positioning, thereby enhancing the performance of the positioning system.

[0058] The categories of opportunity signals include broadcast signals, mobile communication signals, satellite communication signals and Wi-Fi signals.

[0059] For example, a broadcast signal, such as an FM (Frequency Modulation) or DAB (Digital Audio Broadcasting) signal;

[0060] Mobile communication signals, such as 4G (4th Generation, fourth generation mobile communication technology) / 5G (5th Generation, fifth generation mobile communication technology) base station signals;

[0061] Satellite communication signals, such as the Beidou / GPS (Global Positioning System) L5 band. The L5 band is a key frequency band used for civilian high-precision positioning in the GPS and Beidou systems. It offers high accuracy and strong anti-interference capabilities, making it widely used in various applications requiring high-precision positioning.

[0062] A specific embodiment of the present invention discloses a three-dimensional spatial positioning method based on deep learning of multi-source opportunity signals, such as Figure 1 As shown, the following steps are included:

[0063] Step S1: Establish a three-dimensional spatial coordinate system for the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collect multiple types of opportunity signals from the multiple three-dimensional coordinate points, pre-process the multiple types of opportunity signals to obtain sample data, and form a multi-type opportunity signal sample set with corresponding sample labels; wherein the sample labels are the three-dimensional coordinate points of the collected signals;

[0064] Step S2: construct a three-dimensional positioning deep neural network model, and use each type of opportunity signal sample set to train the three-dimensional positioning deep neural network model to obtain a three-dimensional positioning deep neural network model trained with each type of opportunity signal;

[0065] Step S3: After preprocessing multiple opportunity signals in real time at the point to be positioned, the trained three-dimensional positioning deep neural network model of the corresponding category is input to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and the multiple three-dimensional coordinate point positions are averaged to obtain the final three-dimensional space positioning result.

[0066] Step S1 includes steps S11-S13.

[0067] Step S1: Establish a three-dimensional space coordinate system of the area to be located to obtain multiple three-dimensional coordinate points.

[0068] The area to be positioned refers to the specific geographical range where three-dimensional spatial positioning is required. The area is divided into multiple three-dimensional coordinate points by establishing a three-dimensional spatial coordinate system to facilitate the collection and processing of opportunity signals to achieve accurate positioning.

[0069] The way to establish a three-dimensional space coordinate system is as follows:

[0070] The airspace to be located is represented as a three-dimensional space, a spatial rectangular coordinate system is established, the east-west direction and the north-south direction are set as the x-axis and the y-axis respectively, the vertical height is set as the z-axis, and the multiple three-dimensional coordinate points are obtained by dividing according to predetermined intervals.

[0071] like Figure 2 As shown, a rectangular coordinate system with O as the origin is established to represent the positioning space into a three-dimensional space;

[0072] The points are divided into certain predetermined intervals and the three-dimensional coordinate values ​​of each point are marked.

[0073] The specific value of the predetermined interval is determined based on the actual application scenario and positioning accuracy requirements. For example, in indoor environments, the interval can be set to 0.5 meters or 1 meter; in urban environments, the interval can be set to 5 meters or 10 meters; and in open areas, the interval can be set to 10 meters or 20 meters. These example values ​​can be adjusted according to specific circumstances to achieve optimal positioning results.

[0074] Step S12: Collect multiple types of opportunity signals from each three-dimensional coordinate point simultaneously.

[0075] The steps for simultaneously collecting multiple types of opportunity signals from various three-dimensional coordinate points are as follows:

[0076] Use the UAV platform to fly to the corresponding three-dimensional coordinate point, and use the antenna carried by the platform to collect opportunity signals;

[0077] The center frequency of the acquisition is set to the frequency band where the opportunity signal is located. The acquisition bandwidth is set based on the bandwidth of the frequency band where the opportunity signal is located. The acquisition time of each coordinate point is unified, and multiple types of opportunity signals are collected multiple times at the same coordinate point.

[0078] Using a drone platform, the system is launched to the airspace to be located, i.e., the three-dimensional coordinate points mentioned above. The platform's antennas simultaneously collect multiple opportunistic signals, including broadcast signals, mobile communication signals, and satellite communication signals. The center frequency of the acquisition is set to the frequency band where the opportunity signal resides. The acquisition bandwidth is set based on the bandwidth of the opportunity signal's frequency band. The acquisition duration is standardized for each point, and multiple opportunity signal acquisitions are performed at the same three-dimensional coordinate point.

[0079] Step S13: Preprocess the multiple types of opportunity signals to obtain sample data, and form multiple types of opportunity signal sample sets with corresponding sample labels; wherein the sample labels are the three-dimensional coordinate points of the collected signals.

[0080] The sample data obtained by preprocessing each type of opportunity signal and the corresponding sample labels are combined to form multiple opportunity signal sample sets, including:

[0081] Extracting the IQ sequence of each opportunity signal;

[0082] Performing signal interception on the IQ sequence according to a predetermined length m;

[0083] For each three-dimensional coordinate point of the acquisition location, the corresponding sample data covers various types of opportunity signals;

[0084] Each type of opportunity signal sample set is divided into a training set and a validation set according to a preset ratio.

[0085] The in-phase (I) and quadrature (Q) sequences of each opportunity signal are extracted to obtain IQ sequences of multiple opportunity signals.

[0086] In-phase (I): In-phase component, representing the real part of the signal on the time axis; usually corresponds to the cosine component of the signal;

[0087] Quadrature (Q): Quadrature component, representing the imaginary part of the signal on the time axis; usually corresponds to the sinusoidal component of the signal.

[0088] For example, m is recommended to be set to 4096 sampling points. In actual applications, the value of m can be dynamically optimized and adjusted based on factors such as specific signal type (bandwidth, modulation method), hardware performance, and environmental interference.

[0089] The sample set for each signal of opportunity source was randomly divided into a training set and a validation set in an 8:2 ratio. Both sets contain the signal source's IQ sequence and sample labels corresponding to the acquisition point coordinates. This is used for subsequent training of the 3D positioning deep neural network model. During the division process, samples of the same type of signal of opportunity acquired multiple times at the same location were assigned to the same training and validation sets to ensure consistent data distribution and effective model training.

[0090] The function of step S1 is to establish a three-dimensional spatial coordinate system for the area to be located, collect multiple types of opportunity signals from multiple three-dimensional coordinate points and perform preprocessing to form a multi-type opportunity signal sample set containing sample data and corresponding three-dimensional coordinate point labels, providing a data basis for subsequent model training.

[0091] Step S2 includes steps S21-S22.

[0092] Step S21: Construct a three-dimensional positioning deep neural network model.

[0093] like Figure 3 As shown in the figure, a deep neural network model for 3D positioning is constructed. This model adopts a single-input, multiple-output approach. A separate model is built for each signal of opportunity source. These models use the IQ sequence of the acquired signal of opportunity as input data. The model performs a series of operations on the input IQ sequence. After the operations are completed, the fully connected layer (FC layer) converts the processed features into the final prediction output. This output is the coordinate point of the target location in 3D space, corresponding to the three coordinate axes of the 3D space, thereby accurately achieving 3D spatial positioning of the target location.

[0094] The three-dimensional positioning deep neural network model includes, in sequence, an input layer, an initial convolutional layer, a first pooling layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a third transition layer, a fourth dense block, a global average pooling layer, a fully connected layer, and an output layer;

[0095] The input layer is used to receive sample data;

[0096] The initial convolutional layer is used to extract the original time domain features of the sample data;

[0097] The first pooling layer is used to downsample the original time domain features to obtain first signal features;

[0098] The first dense block, the first transition layer, the second dense block, the second transition layer, the third dense block, the third transition layer, and the fourth dense block are used to sequentially perform channel compression and feature downsampling on the first opportunity signal feature to obtain the second signal feature; wherein the first, second, and third transition layers all include a 1×1 convolution operation and an average pooling layer, the convolution operation is used to reduce the number of feature channels, and the average pooling is used to further reduce the size of the feature map;

[0099] The global average pooling layer is used to perform a global average operation on the second signal feature in the time dimension to obtain a third signal feature;

[0100] The fully connected layer is used to convert the third signal feature into a three-dimensional coordinate prediction result through a linear transformation;

[0101] The output layer is used to output the three-dimensional coordinate prediction results.

[0102] Specifically, the model's input is preprocessed IQ signal sequence sample data corresponding to a signal of opportunity. This data is first fed through the input layer to the initial convolutional layer, which extracts the original time-domain features of the sample data. The initial convolutional layer uses a one-dimensional convolution kernel of size 15×1, with 64 convolution kernels. By performing local convolution operations on the time axis, it extracts local time-domain information from the sample data signal.

[0103] The first pooling layer is set after the initial convolution layer, which uses a 3×1 maximum pooling window with a step size of 2 to downsample the feature map, reduce the feature length, reduce the computational complexity, and enhance the translation invariance of the feature; the first pooling layer outputs the first signal feature;

[0104] The model is sequentially configured with the first, second, third, and fourth dense blocks (Dense Block), namely DenseBlock1 to DenseBlock4. Among them, the first to fourth dense blocks include 6, 12, 24, and 16 convolution units, respectively. Each convolution unit includes a one-dimensional convolution operation of size 3×1 and 1×1, and the number of convolution kernels is 32. Dense connections are used within each dense block, that is, the input of each layer of convolution units in the dense block is composed of the output features of all the previous convolution units, thereby achieving efficient feature reuse and multi-scale fusion;

[0105] The first transition layer is set after the first dense block; the second transition layer is set after the second dense block; and the third dense block is set after the third dense block. The first, second, and third transition layers are used for channel compression and feature downsampling. The first, second, and third transition layers have the same structure, including 1×1 convolution and average pooling operations. The convolution operation is used to reduce the number of feature channels, and the average pooling operation further reduces the size of the feature map, which helps to control the scale of model parameters and reduce the risk of overfitting.

[0106] After the fourth dense block, the second signal feature is output;

[0107] A global average pooling layer is set after the fourth dense block to perform a global average operation on the feature map in the time dimension to obtain a fixed-length one-dimensional feature vector. This vector retains the global feature information while significantly reducing the number of model parameters and enhancing the generalization ability of the model. The global average pooling layer outputs the third signal feature.

[0108] The third signal feature is input to a fully connected layer, which uses a linear transformation to output a vector of length 3 to represent the three-dimensional coordinate prediction results in space. This achieves a three-dimensional positioning estimate of the corresponding position of the input IQ signal sample.

[0109] Step S22: Use each type of opportunity signal sample set to train a three-dimensional positioning deep neural network model to obtain a three-dimensional positioning deep neural network model trained with each type of opportunity signal.

[0110] For each opportunity signal source, the corresponding training set and verification set are used to train its three-dimensional positioning deep neural network model to obtain the trained model corresponding to the opportunity signal source.

[0111] Each type of opportunity signal trains a separate 3D localization deep neural network model, which is trained through the following process:

[0112] Using the training set of each type of opportunity signal sample set, a 3D positioning deep neural network model is trained;

[0113] During the training process, the AdamW optimizer is used, combined with adaptive learning rate and weight decay; the mean square error is used as the loss function;

[0114] When the accuracy on the validation set no longer improves after more than n consecutive rounds, training is stopped and the model parameters with the smallest loss on the validation set during training are saved as the trained 3D positioning deep neural network model.

[0115] For example, n is set to 2. When the accuracy on the validation set stops improving after two or more rounds, training is stopped. The model parameters with the lowest loss on the validation set during training are saved as the optimal deep neural network model. In practice, the value of n should be adjusted according to specific needs.

[0116] AdamW was used as the optimizer during training. AdamW is an optimization algorithm that combines the adaptive learning rate adjustment of the Adam optimizer with weight decay (also known as L2 regularization). A learning rate decay strategy was adopted, where the learning rate was halved after every two training rounds. This facilitated more detailed adjustment of model parameters in the later stages of training and improved model convergence accuracy. The loss function used was mean squared error.

[0117] The loss function is as follows:

[0118]

[0119] Among them, N is the number of samples in the sample set of this type of opportunity signal, (x i ,y i ,z i ) is the true three-dimensional coordinate point of the sample data, The 3D coordinate points predicted by the deep neural network model for 3D positioning.

[0120] Independent deep learning neural network models trained for different signal sources.

[0121] The purpose of step S2 is to construct a three-dimensional positioning deep neural network model for each opportunity signal source, and use the corresponding sample set for training and the validation set for verification to obtain a trained three-dimensional positioning deep neural network model, providing accurate prediction capabilities for subsequent real-time positioning.

[0122] Step S3, specifically.

[0123] After preprocessing multiple opportunity signals in real time at the point to be positioned, the trained three-dimensional positioning deep neural network model of the corresponding category is input to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and the multiple three-dimensional coordinate point positions are averaged to obtain the final three-dimensional space positioning result.

[0124] Independent deep learning neural network models are trained for different signal sources. Therefore, after completing the acquisition of the opportunity signal at the point to be positioned, the pre-processed IQ sequence is input into the corresponding trained three-dimensional positioning deep neural network model for inference according to the signal source type corresponding to the IQ sequence, so as to obtain the three-dimensional positioning coordinate point position corresponding to each signal source.

[0125] The final three-dimensional spatial positioning result is obtained by averaging the positions of multiple three-dimensional coordinate points, as follows:

[0126]

[0127] in, The position output by the trained 3D positioning depth model; For multiple The final three-dimensional spatial positioning result is obtained by averaging; M is the number of opportunity signals used.

[0128] The function of step S3 is to acquire and preprocess multiple types of opportunity signals in real time at the point to be positioned, input the processed signals into the corresponding trained three-dimensional positioning deep neural network model for inference, obtain the three-dimensional coordinate point position corresponding to each signal source, and finally obtain the final three-dimensional space positioning result by averaging these position coordinates.

[0129] Example 2:

[0130] Another embodiment of the present invention discloses a three-dimensional spatial positioning system based on deep learning of multi-source opportunity signals, thereby implementing the three-dimensional spatial positioning method based on deep learning of multi-source opportunity signals in Example 1. The specific implementation of each module is described in the corresponding description of Example 1.

[0131] like Figure 4 As shown, the system includes a data acquisition and preprocessing module M1, a model training module M2 and a real-time positioning module M3;

[0132] The data acquisition and preprocessing module M1 is used to establish a three-dimensional spatial coordinate system of the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collect multiple types of opportunity signals from the multiple three-dimensional coordinate points, preprocess the multiple types of opportunity signals to obtain sample data, and form a multi-type opportunity signal sample set with corresponding sample labels; wherein the sample label is the three-dimensional coordinate point of the collected signal;

[0133] The model training module M2 is used to build a three-dimensional positioning deep neural network model, and use each type of opportunity signal sample set to train the three-dimensional positioning deep neural network model to obtain a three-dimensional positioning deep neural network model trained with each type of opportunity signal;

[0134] The real-time positioning module M3 obtains multiple opportunity signals in real time at the point to be positioned and performs pre-processing, then inputs the trained three-dimensional positioning deep neural network model of the corresponding category to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and averages the multiple three-dimensional coordinate point positions to obtain the final three-dimensional space positioning result.

[0135] In summary, the three-dimensional spatial positioning method and system based on deep learning of multi-source opportunity signals according to the embodiments of the present invention have the following beneficial effects:

[0136] (1) Aiming at the problem of three-dimensional spatial positioning, the present invention proposes a deep learning three-dimensional spatial positioning method based on multi-source opportunity signals, which is suitable for autonomous positioning in a specific three-dimensional space;

[0137] (2) The present invention utilizes opportunity signals as a positioning source, which can provide reliable positioning information when traditional positioning sources are interfered with, while reducing the system's dependence on GPS and other external signals, thereby significantly improving the robustness and availability of the positioning system;

[0138] (3) The present invention combines multiple opportunistic signal sources (such as broadcast signals, mobile communication signals, satellite communication signals, etc.), can fully utilize the characteristics of different signals, and improve positioning accuracy and robustness;

[0139] (4) The present invention directly implements positioning through deep learning based on IQ signals, without the need to extract features such as channel state information, and has no special requirements for the signal protocol of the opportunity signal source, thus having wider applicability;

[0140] (5) With the help of the powerful feature extraction capability of deep learning, the present invention can automatically learn key features from complex signal data, capture environmental changes and signal characteristics more comprehensively and accurately, and thus improve positioning accuracy and stability in complex environments.

[0141] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0142] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A three-dimensional spatial positioning method based on deep learning of multi-source opportunity signals, characterized in that: include: Establishing a three-dimensional spatial coordinate system for the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collecting multiple types of opportunity signals from each three-dimensional coordinate point, preprocessing the multiple types of opportunity signals to obtain sample data, and forming a multi-type opportunity signal sample set with corresponding sample labels; wherein the sample label is the three-dimensional coordinate point of the collected signal; Construct a 3D positioning deep neural network model, and use each type of opportunity signal sample set to train the 3D positioning deep neural network model to obtain a 3D positioning deep neural network model trained with each type of opportunity signal; After preprocessing multiple opportunity signals in real time at the point to be positioned, the trained three-dimensional positioning deep neural network model of the corresponding category is input to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and the multiple three-dimensional coordinate point positions are averaged to obtain the final three-dimensional space positioning result.

2. The method according to claim 1, characterized in that The way to establish a three-dimensional space coordinate system is as follows: The airspace to be located is represented as a three-dimensional space, a spatial rectangular coordinate system is established, the east-west direction and the north-south direction are set as the x-axis and the y-axis respectively, the vertical height is set as the z-axis, and the multiple three-dimensional coordinate points are obtained by dividing according to predetermined intervals.

3. The method according to claim 2, characterized in that The steps for simultaneously collecting multiple types of opportunity signals from various three-dimensional coordinate points are as follows: Use the UAV platform to launch to the corresponding three-dimensional coordinate point, and use the antenna carried by the platform to collect opportunity signals; The center frequency of the acquisition is set to the frequency band where the opportunity signal is located. The acquisition bandwidth is set based on the bandwidth of the frequency band where the opportunity signal is located. The acquisition time of each coordinate point is unified, and multiple types of opportunity signals are collected multiple times at the same coordinate point.

4. The method according to claim 3, characterized in that The three-dimensional positioning deep neural network model includes, in sequence, an input layer, an initial convolutional layer, a first pooling layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a third transition layer, a fourth dense block, a global average pooling layer, a fully connected layer, and an output layer; The input layer is used to receive sample data; The initial convolutional layer is used to extract the original time domain features of the sample data; The first pooling layer is used to downsample the original time domain features to obtain first signal features; The first dense block, the first transition layer, the second dense block, the second transition layer, the third dense block, the third transition layer, and the fourth dense block are used to perform channel compression and feature downsampling on the first opportunity signal feature in sequence to obtain the second signal feature; wherein the first, second, and third transition layers all include 1×1 convolution operations and average pooling layers, the convolution operation is used to reduce the number of feature channels, and the average pooling is used to further reduce the size of the feature map.

5. The method according to claim 4, characterized in that: The global average pooling layer is used to perform a global average operation on the second signal feature in the time dimension to obtain a third signal feature; The fully connected layer is used to convert the third signal feature into a three-dimensional coordinate prediction result through a linear transformation; The output layer is used to output the three-dimensional coordinate prediction results.

6. The method according to claim 5, characterized in that Each type of opportunity signal trains a separate 3D localization deep neural network model, which is trained through the following process: Using the training set of each type of opportunity signal sample set, a 3D positioning deep neural network model is trained; During training, the AdamW optimizer is used, combined with adaptive learning rate and weight decay; The mean square error is used as the loss function; When the accuracy on the validation set no longer improves after more than n consecutive rounds, training is stopped and the model parameters with the smallest loss on the validation set during training are saved as the trained 3D positioning deep neural network model.

7. The method according to claim 6, characterized in that The loss function is as follows: Among them, N is the number of samples in the sample set of this type of opportunity signal, (x i ,y i ,z i ) is the true three-dimensional coordinate point of the sample data, The 3D coordinate points predicted by the deep neural network model for 3D positioning.

8. The method according to claim 1, characterized in that: The sample data obtained by preprocessing each type of opportunity signal and the corresponding sample labels are combined to form multiple opportunity signal sample sets, including: Extracting the IQ sequence of each opportunity signal; Performing signal interception on the IQ sequence according to a predetermined length m; For each three-dimensional coordinate point of the acquisition location, the corresponding sample data covers various types of opportunity signals; Each type of opportunity signal sample set is divided into a training set and a validation set according to a preset ratio.

9. The method according to any one of claims 1 to 8, characterized in that The categories of opportunity signals include broadcast signals, mobile communication signals, satellite communication signals and Wi-Fi signals.

10. A three-dimensional spatial positioning system based on deep learning of multi-source opportunity signals, characterized in that: Includes data acquisition and preprocessing module, model training module and real-time positioning module; The data acquisition and preprocessing module is used to establish a three-dimensional spatial coordinate system of the area to be located to obtain multiple three-dimensional coordinate points, simultaneously collect multiple types of opportunity signals from the multiple three-dimensional coordinate points, preprocess the multiple types of opportunity signals to obtain sample data, and form multiple types of opportunity signal sample sets with corresponding sample labels; wherein the sample labels are the three-dimensional coordinate points of the collected signals; The model training module is used to build a three-dimensional positioning deep neural network model, and use each type of opportunity signal sample set to train the three-dimensional positioning deep neural network model to obtain a three-dimensional positioning deep neural network model trained with each type of opportunity signal; The real-time positioning module obtains multiple opportunity signals in real time at the point to be positioned and performs preprocessing, then inputs a trained three-dimensional positioning deep neural network model of the corresponding category to obtain multiple three-dimensional coordinate point positions corresponding to the multiple opportunity signals, and averages the multiple three-dimensional coordinate point positions to obtain the final three-dimensional space positioning result.

Images