Unmanned aerial vehicle positioning method based on tensorized long short-term memory network and related device

Abstract

The invention discloses an unmanned aerial vehicle positioning method based on a tensorized long-short-term memory network and a related device, and the method comprises the steps: constructing the azimuth and pitch angle functions of a kth unmanned aerial vehicle relative to a receiving array, and carrying out the continuous detection of the azimuth angles and pitch angles of all unmanned aerial vehicles through the receiving array; constructing a three-dimensional tensor # imgabs0 # obtained when the unmanned aerial vehicle is detected for the tth time in continuous T times of detection according to the angle function, constructing a network input function # imgabs2 # during the tth time of detection on the basis of # imgabs1 #, and stacking the network input functions of the T times of detection in the fourth dimension, the obtained network input function # imgabs3 # constructs a tensorized long-short-term memory network capable of combining N previous states; and extracting a feature intermediate quantity comprising angle information of the first N state quantities from an input function # imgabs4 # by adopting a tensorized long-short-term memory network, and inputting the feature intermediate quantity into a preset full-connection layer to obtain estimated values of the azimuth angle and the pitch angle of the unmanned aerial vehicle during the (T + 1) th detection, and the method is suitable for a scene in which the unmanned aerial vehicle is continuously detected.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid inspection, and in particular to a UAV positioning method based on a tensorized long short-term memory network and related devices. Background Art

[0002] With the development of drone technology, drones have been used in power inspection and other fields. The premise of controlling drones is to obtain the drone's position information. In recent years, with the rapid development of deep learning, DOA (Direction of Arrival) estimation methods based on deep learning have been widely studied, ensuring the algorithm stability under harsh conditions such as low signal-to-noise ratio or array element defects.

[0003] However, traditional deep learning-based methods often only sample the covariance matrix of a single measurement as the network input. This input construction method is not suitable for scenarios that require continuous detection, such as drone positioning in power grids. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a UAV positioning method and related devices based on a tensorized long short-term memory network.

[0005] In a first aspect, the present invention provides a UAV positioning method based on a tensorized long short-term memory network, the number of UAVs is K, and the method comprises:

[0006] Constructing an azimuth function and a pitch angle function of the kth UAV relative to a receiving array, k=1, ..., K, wherein the receiving array continuously detects the azimuth and pitch angles of all UAVs;

[0007] According to the azimuth function and the pitch angle function, a three-dimensional tensor is constructed when the receiving array detects the drone for the tth time in T consecutive detections. t=1,…,T;

[0008] Based on three-dimensional tensor Construct the network input function for the tth detection

[0009] The network input function detected T times is stacked in the fourth dimension to obtain the network input function

[0010] Construct a tensorized long short-term memory network that can merge N previous states;

[0011] Use tensorized long short-term memory network to input function from the network Extracting the characteristic intermediate quantity including the angle information of the first N state quantities;

[0012] The feature intermediate quantity is input into a preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0013] In a second aspect, the present invention provides a UAV positioning device based on a tensorized long short-term memory network, the number of UAVs is K, and the device comprises:

[0014] An azimuth function construction module is used to construct an azimuth function and a pitch angle function of the kth UAV relative to the receiving array, k=1, ..., K, and the receiving array continuously detects the azimuth and pitch angles of all UAVs;

[0015] A three-dimensional tensor construction module is used to construct a three-dimensional tensor obtained by the receiving array in the t-th detection of the drone in T consecutive detections according to the azimuth function and the pitch angle function. t=1,…,T;

[0016] The first input function builds a module for 3D tensor-based Construct the network input function for the tth detection

[0017] The second input function building module is used to stack the network input functions of T detections in the fourth dimension to obtain the network input function

[0018] A network building module for building a tensorized long short-term memory network that can merge N previous states;

[0019] The feature intermediate quantity extraction module is used to extract the function from the network input using the tensorized long short-term memory network Extracting the characteristic intermediate quantity including the angle information of the first N state quantities;

[0020] The azimuth information estimation module is used to input the feature intermediate quantity into a preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0021] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0022] at least one processor; and

[0023] a memory communicatively connected to the at least one processor; wherein,

[0024] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the UAV positioning method based on the tensorized long short-term memory network described in the first aspect of the present invention.

[0025] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to implement the UAV positioning method based on a tensorized long short-term memory network described in the first aspect of the present invention when executed.

[0026] The embodiment of the present invention provides a method for positioning a UAV based on a tensorized long short-term memory network, wherein the number of UAVs is K, including: constructing an azimuth function and a pitch angle function of the kth UAV relative to a receiving array, wherein the receiving array continuously detects the azimuth and pitch angles of all UAVs; constructing a three-dimensional tensor obtained by the receiving array in T consecutive detections and the tth detection of the UAV according to the azimuth function and the pitch angle function; Based on three-dimensional tensor Construct the network input function for the tth detection The network input function detected T times is stacked in the fourth dimension to obtain the network input function Construct a tensorized long short-term memory network that can merge N previous states; use the tensorized long short-term memory network to input the function from the network The feature intermediate quantity including the angle information of the first N state quantities is extracted, and the feature intermediate quantity is input into the preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0027] The beneficial effect of the scheme of the present invention is that the receiving array uniformly detects the azimuth of K drones, and the received signal is a multidimensional signal. The physical properties and dimensional distribution characteristics corresponding to each dimension of the multidimensional signal may be different. There are complex correlations within the data, and tensors are effective tools for representing high-dimensional multidimensional data. Therefore, this scheme organizes the multidimensional signal into a three-dimensional tensor form, which can obtain the correlation of the azimuth change of the drone corresponding to each dimension in the multidimensional signal, and construct a tensorized long short-term memory network, so as to extract the feature intermediate quantity rich in azimuth information according to the network, and then the azimuth can be estimated based on the feature intermediate quantity. It is suitable for scenes that require continuous detection, such as drone positioning in power grids.

[0028] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is a flow chart of a UAV positioning method based on a tensorized long short-term memory network provided in the first embodiment of the present invention;

[0031] Figure 2 A receiving array structure of a uniform planar array provided in the first embodiment of the present invention;

[0032] Figure 3 It is a flow chart of a UAV positioning method based on a tensorized long short-term memory network provided in the first embodiment of the present invention;

[0033] Figure 4 3 is a schematic diagram comparing the root mean square error of prediction results of different schemes provided in the first embodiment of the present invention when the signal-to-noise ratio is used as a variable;

[0034] Figure 5 is a schematic diagram comparing the root mean square errors of prediction results of different solutions provided by the first embodiment of the present invention when the number of sampling snapshots is used as a variable;

[0035] Figure 6 is a structural schematic diagram of a UAV positioning device based on a tensorized long short-term memory network provided in Embodiment 2 of the present invention;

[0036] Figure 7 It is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0038] Embodiment 1

[0039] Figure 1The flowchart of a method for positioning a UAV based on a tensorized long short-term memory network is provided in the first embodiment of the present invention. This embodiment is applicable to the case of positioning a UAV based on a tensorized long short-term memory network. The method can be performed by a UAV positioning device based on a tensorized long short-term memory network. The UAV positioning device based on a tensorized long short-term memory network can be implemented in the form of hardware and / or software. The UAV positioning device based on a tensorized long short-term memory network can be configured in an electronic device. Figure 1 As shown, the UAV positioning method based on tensorized long short-term memory network includes:

[0040] S101. Construct an azimuth angle function and a pitch angle function of the kth UAV relative to the receiving array.

[0041] Among them, k = 1, ..., K, and the receiving array continuously detects the azimuth and pitch angles of all drones.

[0042] The azimuth function is:

[0043] θ k (t) = θ k (t-1)+Δ1(t);

[0044] The pitch angle function is:

[0045]

[0046] Among them, θ k (t),θ k (t-1) are the azimuth angles of the UAV at the t-th and t-1-th detections, respectively, and Δ1(t) is the azimuth offset between the t-th and t-1-th detections; are the pitch angles of the UAV at the t-th and t-1-th detections, respectively. Δ2(t) is the pitch angle offset between the t-th and t-1-th detections.

[0047] Considering that the drone has four movement modes, namely vertical movement, horizontal movement, circular movement, and yaw, the corresponding angle change modes can be classified into three types, namely:

[0048] Only the azimuth varies uniformly:

[0049] θ k (t) = θ k (t-1)+Δ1(t);

[0050]

[0051] Only the elevation angle varies uniformly:

[0052] θ k (t) = θ k(t-1);

[0053]

[0054] Both elevation and azimuth change nonlinearly:

[0055] θ k (t) = θ k (t-1)+Δ1(t);

[0056]

[0057] Thus, the wave arrival angle sequence corresponding to the UAV is constructed as θ k (t) and

[0058] Specifically, the receiving array uses N x ×N y physical antenna elements, where N x and N y Represents the number of array elements along the x-axis and y-axis respectively. The array element spacing in the x-axis direction and the y-axis direction is d. The unit spacing d is taken as half of the wavelength λ of the incident narrowband signal, that is, d = λ / 2. The structure is constructed according to the uniform planar array structure. Setting d = λ / 2 is to comply with the Nyquist sampling theorem. According to the Nyquist sampling theorem, the sampling frequency of the signal must be greater than twice the highest frequency component in the signal to ensure that the sampled digital signal completely retains the information in the original signal. The purpose is to prevent aliasing during the sampling and recovery process. The receiving array is based on Figure 2 The receiving array structure of the uniform array is shown.

[0059] The receiving array sends a transmission signal in the direction of the drone. After encountering the drone, the signal returns to the receiving array, and then the position information of the drone at each collection moment can be obtained.

[0060] S102, based on the azimuth function and the pitch angle function, construct a three-dimensional tensor obtained by the receiving array when detecting the drone for the tth time in T consecutive detections

[0061] Wherein, t = 1, ..., T. It should be noted here that since the receiving array detects K drones at the same time during each detection, the position information obtained is a multi-dimensional mixed signal, so this solution processes the position information of K drones in a unified manner. Specifically, a drone can detect another drone once every 0.5 seconds.

[0062] Assume that there are K drones, and their azimuth angles (including pitch angle and azimuth angle) are The receiving array continuously detects the far-field narrowband non-correlated signals emitted by the drone T times. Non-correlated signals refer to those signals that have no statistical similarity. Their changing trends in the time domain or frequency domain are unrelated and are usually used to describe that there is no obvious correlation between two signals. k (t) and are the azimuth and pitch angle of the kth UAV at time t, k = 1, 2, ..., K, t = 1, 2, ..., T, then the Q sampling snapshot signals under T detections are expressed as a three-dimensional tensor It is expressed as:

[0063]

[0064] Among them, θ k (t) and are the azimuth and pitch angle of the kth UAV at the tth detection, is the noise tensor, denote the corresponding azimuth and elevation steering vectors along the x-axis and y-axis, respectively, s k is the source signal of the receiving array, Represents a vector outer product operation;

[0065] s k =[s k,1 ,s k,2 ,···,s k,T ] T ;

[0066]

[0067] in,

[0068] ω k (t) = cos(θ k (t)), [·] T Represents a transpose operation.

[0069] Indicates that the dimension is 3, and the length of each dimension is N x 、N y ,Q.

[0070] The noise tensor refers to the data tensor containing noise in high-dimensional data due to measurement errors, equipment noise or other external interference. It usually manifests as inaccurate data or outliers, affecting the overall quality of the data and the analysis results. The noise tensor can be regarded as a stack of the original data tensor and random noise.

[0071] S103, based on three-dimensional tensor Construct the network input function for the tth detection

[0072] Specifically include: Calculate three-dimensional tensor The covariance tensor R(t) of ; the different dimensions of the covariance tensor R(t) are merged to obtain the intermediate tensor The intermediate tensor The imaginary and real parts of are connected in the third dimension to obtain the network input function

[0073] Based on the above signal model, the corresponding The covariance tensor of [·] * represents the conjugation operation, represents the vector outer product operation,

[0074] Among them, J is the unit tensor, σ n is the tensor strength.

[0075] Optionally, R(t) is replaced by a sampling covariance matrix. Covariance refers to the relationship between two tensors and is used to measure the linear correlation between them. The covariance matrix is ​​symmetric, that is, Cov(X,Y) = Cov(Y,X). The covariance matrix is ​​used to reduce the dimension of the data and improve the interpretability and prediction accuracy of the model.

[0076] Covariance matrix The expression is:

[0077] in, is the sampling covariance matrix, Q is the number of sampling snapshots, <·>3 represents the tensor shrinking operation in the third dimension, [·] * Represents the conjugation operation.

[0078] Q is the number of sampling snapshots, <·>3 means that the tensor is shrunk and merged in the third dimension, [·] * Represents the conjugation operation.

[0079] Each contraction will reduce the order of the tensor by two, that is, the dimension of the tensor can be reduced, which can greatly improve the calculation speed, especially when there are multiple drones in this solution.

[0080] Mathematically, the sampled snapshot signal can be represented as a matrix, where the rows represent the number of samples and the columns represent different channels. For example, if the sample is 256 times and there are 8 channels, the sample data is a 256*8 matrix. The sampled snapshot signal can be represented as a matrix, where the rows represent the number of samples and the columns represent different channels. For example, if the sample is 256 times and there are 8 channels, the sample data is a 256*8 matrix.

[0081] At the same time, in order to facilitate network computing, The {1,2} and {3,4} dimensions are combined to obtain the intermediate tensor At the same time The imaginary and real parts of the network are stacked into a third dimension to obtain the network input function The expression is:

[0082]

[0083] Among them, Re(·) and Im(·) represent the operations of taking the real part and the imaginary part respectively. Represents tensor stacking in the i-th dimension. This is because for most neural networks, only real numbers can be operated, not imaginary numbers. Therefore, by stacking the imaginary part with the real part, the imaginary part feature can be converted into a real part feature that can be operated. The tensor concatenation operation will not generate new dimensions, but will only increase the length (dimension size) in a certain dimension direction. The tensor stacking operation will generate new dimensions, merging each tensor as a new dimension with another tensor. The dimensions of the two tensors to be stacked and the length of each dimension must be equal.

[0084] S104, stacking the network input functions detected T times in the fourth dimension to obtain the network input function

[0085] In order to facilitate network training, the T detection results (i.e., the network input function ) are stacked in the fourth dimension to obtain the final network input

[0086]

[0087] Represents stacking of tensors in the 4th dimension.

[0088] S105. Construct a tensorized long short-term memory network that can merge N previous states.

[0089] Specifically, it includes: constructing a convolutional long short-term memory network that can merge previous states, abbreviated as ConvLSTM (Convolutional Neural Networks Long Short-Term Memory), where the convolutional long short-term memory network includes a convolution kernel; performing tensor chain decomposition on the convolution kernel to obtain a tensorized long short-term memory network.

[0090] Convolutional Long Short-Term Memory Network is a deep learning model that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). It is mainly used to process spatiotemporal data, especially for tasks that need to consider spatial features and temporal dependencies.

[0091] The expression of the convolutional long short-term memory network is as follows:

[0092]

[0093] Among them, σ is the sigma activation function, Represents the input The weight of To utilize the information integration function of the previous state. In order to The information extraction method is consistent with The function design is as follows:

[0094]

[0095] Where D(i) is the convolution kernel for the previous state. However, since the large dimension of the convolution kernel will lead to low network efficiency, the convolution kernel {K(0),…,K(N)} is decomposed into tensor chains. Specifically, represents the i-th convolution kernel in the above formula, where K(i)=i[K(1)-1]+1 is the filter size that increases linearly with i; D(1) is the initial filter size; C(i) is C(0) is the number of channels of the function φ, so C(0) = 4×C out , where C out is the number of channels of the high-order convolutional LSTM, and the corresponding convolution tensor chain is decomposed by using a series of low-dimensional factors To express the original high-dimensional convolution kernel:

[0096]

[0097] Then we can get the tensorized long short-term memory network:

[0098]

[0099] S106, using tensorized long short-term memory network to input function from network The feature intermediate quantity including the angle information of the first N state quantities is extracted.

[0100] S107, input the feature intermediate quantity into the preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0101] Figure 3 This is a flowchart of a UAV positioning method based on tensorized long short-term memory network. In general, Figure 3 As shown, the drone positioning process includes the following steps:

[0102] S1, construct a uniform array;

[0103] S2, time series tensor signal modeling;

[0104] S3, tensorized long short-term memory network modeling;

[0105] S4, the fully connected layer outputs the orientation information.

[0106] The embodiment of the present invention provides a method for positioning a UAV based on a tensorized long short-term memory network, wherein the number of UAVs is K, including: constructing an azimuth function and a pitch angle function of the kth UAV relative to a receiving array, wherein the receiving array continuously detects the azimuth and pitch angles of all UAVs; constructing a three-dimensional tensor obtained by the receiving array in T consecutive detections and the tth detection of the UAV according to the azimuth function and the pitch angle function; Based on three-dimensional tensor Construct the network input function for the tth detection The network input function detected T times is stacked in the fourth dimension to obtain the network input function Construct a tensorized long short-term memory network that can merge N previous states; use the tensorized long short-term memory network to input the function from the network The feature intermediate quantity including the angle information of the first N state quantities is extracted, and the feature intermediate quantity is input into the preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0107] The beneficial effect of the scheme of the present invention is that the receiving array uniformly detects the azimuth of K drones, and the received signal is a multidimensional signal. The physical properties and dimensional distribution characteristics corresponding to each dimension of the multidimensional signal may be different. There are complex correlations within the data, and tensors are effective tools for representing high-dimensional multidimensional data. Therefore, this scheme organizes the multidimensional signal into a three-dimensional tensor form, which can obtain the correlation of the azimuth change of the drone corresponding to each dimension in the multidimensional signal, and construct a tensorized long short-term memory network, so as to extract the feature intermediate quantity rich in azimuth information according to the network, and then the azimuth can be estimated based on the feature intermediate quantity. It is suitable for scenes that require continuous detection, such as drone positioning in power grids.

[0108] The effects of the present invention are further described below in conjunction with a simulation example.

[0109] Simulation example: Using a uniform array, consider performing T=10 consecutive direction finding of two target sources. At the same time, compared with other similar experimental settings, the accuracy of the incident source angle is 0.01°, which can meet most practical conditions. The Monte Carlo experimental method is used to analyze the impact of different orders on the DOA estimation performance, and the root mean square error (RMSE) is used as the measurement indicator. Among them, K is the number of sources to be estimated, T is the number of consecutive estimates, and M is the number of Monte Carlo tests. Represent the azimuth estimated by the network and the actual azimuth, respectively. They represent the elevation angle estimated by the network and the actual elevation angle respectively.

[0110] Figure 4 3 is a schematic diagram comparing the root mean square error of prediction results of different schemes provided in the first embodiment of the present invention when the signal-to-noise ratio is used as a variable; Figure 5 FIG. 1 is a schematic diagram showing a comparison of the root mean square error of the prediction results of the methods of different solutions provided in the first embodiment of the present invention when the number of sampling snapshots is used as a variable. Figure 4 and Figure 5 The comparison results show that the proposed method has performance advantages in the accuracy of position information estimation in both different expected signal-to-noise ratio scenarios and different sampling snapshot number scenarios. Compared with the traditional methods based on MLP (Multilayer Perceptron), CNN, and ConLSTM, the proposed method adopts the structure of tensorized long short-term memory network, making full use of the correlation information between signals during continuous detection, thus having better position information estimation performance.

[0111] In summary, the present invention designs a time series tensor signal modeling and based on this, proposes an angle feature extraction method based on a tensorized long short-term memory network, thereby achieving high-precision azimuth information estimation.

[0112] Embodiment 2

[0113] Figure 6 A schematic diagram of the structure of a UAV positioning device based on a tensorized long short-term memory network provided in the second embodiment of the present invention. The number of UAVs is K, such as Figure 6 As shown, the UAV positioning device based on the tensorized long short-term memory network includes:

[0114] An azimuth function construction module 601 is used to construct an azimuth function and a pitch angle function of the kth UAV relative to a receiving array, k=1, ..., K, and the receiving array continuously detects the azimuth and pitch angles of all UAVs;

[0115] The three-dimensional tensor construction module 602 is used to construct the three-dimensional tensor obtained by the receiving array in the t-th detection of the drone in T consecutive detections according to the azimuth function and the pitch angle function. t=1,…,T;

[0116] The first input function building module 603 is used to construct a three-dimensional tensor based on Construct the network input function for the tth detection

[0117] The second input function construction module 604 is used to stack the network input functions detected T times in the fourth dimension to obtain the network input function

[0118] A network construction module 605 is used to construct a tensorized long short-term memory network capable of merging N previous states;

[0119] The feature intermediate quantity extraction module 606 is used to extract the feature intermediate quantity from the network input function using a tensorized long short-term memory network. Extracting the characteristic intermediate quantity including the angle information of the first N state quantities;

[0120] The azimuth information estimation module 607 is used to input the feature intermediate quantity into a preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

[0121] The UAV positioning device based on the tensorized long short-term memory network provided in the embodiment of the present invention can execute the UAV positioning method based on the tensorized long short-term memory network provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0122] Embodiment 3

[0123] Figure 7 A schematic diagram of the structure of an electronic device 40 that can be used to implement an embodiment of the present invention is shown, and the electronic device can be a receiving array in the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0124] like Figure 7 As shown, the electronic device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. In the RAM 43, various programs and data required for the operation of the electronic device 40 can also be stored. The processor 41, the ROM 42, and the RAM 43 are connected to each other through a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0125] A number of components in the electronic device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0126] The processor 41 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 41 executes the various methods and processes described above, such as a UAV positioning method based on a tensorized long short-term memory network.

[0127] In some embodiments, the UAV positioning method based on the tensorized long short-term memory network can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the UAV positioning method based on the tensorized long short-term memory network described above can be performed. Alternatively, in other embodiments, the processor 41 can be configured to execute the UAV positioning method based on the tensorized long short-term memory network in any other appropriate manner (for example, by means of firmware).

[0128] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0130] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0131] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0132] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0133] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0134] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0135] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A UAV positioning method based on tensorized long short-term memory network, characterized in that: The number of drones is K, and the method includes: Constructing an azimuth function and a pitch angle function of the kth UAV relative to a receiving array, k=1, ..., K, wherein the receiving array continuously detects the azimuth and pitch angles of all UAVs; According to the azimuth function and the pitch angle function, a three-dimensional tensor is constructed when the receiving array detects the drone for the tth time in T consecutive detections. t=1,…,T; Based on three-dimensional tensor Construct the network input function for the tth detection The network input function detected T times is stacked in the fourth dimension to obtain the network input function Construct a tensorized long short-term memory network that can merge N previous states; Use tensorized long short-term memory network to input function from the network Extracting the characteristic intermediate quantity including the angle information of the first N state quantities; The feature intermediate quantity is input into a preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

2. The UAV positioning method based on tensorized long short-term memory network according to claim 1, characterized in that: The azimuth function is: i k (t)=θ k (t-1)+Δ1(t); The pitch angle function is: Among them, θ k (t),θ k (t-1) are the azimuth angles of the UAV at the t-th and t-1-th detections, respectively, and Δ1(t) is the azimuth offset between the t-th and t-1-th detections; are the pitch angles of the UAV at the t-th and t-1-th detections, respectively. Δ2(t) is the pitch angle offset between the t-th and t-1-th detections.

3. The UAV positioning method based on tensorized long short-term memory network according to claim 1, characterized in that: 3D Tensor The expression is as follows: Among them, θ k (t) and are the azimuth and pitch angle of the kth UAV at the tth detection, is the noise tensor, denote the corresponding azimuth and elevation steering vectors along the x-axis and y-axis, respectively, s k is the source signal of the receiving array, [°] represents a vector outer product operation; in, [·] T Represents a transpose operation.

4. The method for positioning a UAV based on a tensorized long short-term memory network according to claim 3, characterized in that: The three-dimensional tensor Construct the network input function for the tth detection include: Computing 3D tensors The covariance tensor R(t); Concatenate the different dimensions of the covariance tensor R(t) to get the intermediate tensor The intermediate tensor The imaginary and real parts of are tensor-connected in the third dimension to obtain the network input function 5. The method for positioning a UAV based on a tensorized long short-term memory network according to claim 4, characterized in that: Where E{·} is the covariance operator, [°] represents the vector outer product operation, [·] * represents the conjugate operation, J is the unit tensor, σ n is the tensor strength.

6. The method for positioning a UAV based on a tensorized long short-term memory network according to claim 4, characterized in that: in, is the sampling covariance matrix, Q is the number of sampling snapshots, <·>3 represents the tensor shrinking operation in the third dimension, [·] * Represents the conjugation operation.

7. The UAV positioning method based on tensorized long short-term memory network according to any one of claims 1 to 6, characterized in that: The construction of a tensorized long short-term memory network capable of merging N previous states includes: Constructing a convolutional long short-term memory network capable of merging previous states, wherein the convolutional long short-term memory network includes a convolution kernel; The convolution kernel is subjected to tensor chain decomposition to obtain a tensorized long short-term memory network.

8. A UAV positioning device based on a tensorized long short-term memory network, used to execute the UAV positioning method based on a tensorized long short-term memory network according to any one of claims 1 to 7, characterized in that: The number of drones is K, and the device includes: An azimuth function construction module is used to construct an azimuth function and a pitch angle function of the kth UAV relative to the receiving array, k=1, ..., K, and the receiving array continuously detects the azimuth and pitch angles of all UAVs; A three-dimensional tensor construction module is used to construct a three-dimensional tensor obtained by the receiving array in the t-th detection of the drone in T consecutive detections according to the azimuth function and the pitch angle function. t=1,…,T; The first input function builds a module for 3D tensor-based Construct the network input function for the tth detection The second input function building module is used to stack the network input functions of T detections in the fourth dimension to obtain the network input function A network building module for building a tensorized long short-term memory network that can merge N previous states; The feature intermediate quantity extraction module is used to extract the function from the network input using the tensorized long short-term memory network Extracting the characteristic intermediate quantity including the angle information of the first N state quantities; The azimuth information estimation module is used to input the feature intermediate quantity into a preset fully connected layer to obtain the estimated values ​​of the azimuth and pitch angle of the drone during the T+1th detection.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the UAV positioning method based on a tensorized long short-term memory network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the UAV positioning method based on a tensorized long short-term memory network as described in any one of claims 1 to 7 when executed.

Images