CN110764084A - Radar detection method for shielding target under complex terrain - Google Patents
Radar detection method for shielding target under complex terrain Download PDFInfo
- Publication number
- CN110764084A CN110764084A CN201911079151.XA CN201911079151A CN110764084A CN 110764084 A CN110764084 A CN 110764084A CN 201911079151 A CN201911079151 A CN 201911079151A CN 110764084 A CN110764084 A CN 110764084A
- Authority
- CN
- China
- Prior art keywords
- target
- channel
- training
- multipath
- radar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a radar detection method for shielding targets under complex terrain, which comprises the steps of firstly, obtaining corresponding multipath propagation characteristics of detected targets at different positions by utilizing a channel modeling and channel measuring method, and establishing a deep learning training data set; then, training and learning the target position and the multipath propagation characteristics by utilizing deep learning, and further establishing a correlation model of the target position and the multipath propagation characteristics; and finally, the echo multipath signal characteristics of the unknown target position are used as input to obtain the position information of the shielding target. The invention can quickly and accurately output the position information of the target, and has higher target recognition rate, better instantaneity, universality and simplicity.
Description
Technical Field
The invention relates to the technical field of radar detection.
Background
The detection and localization of non-direct-view targets has been one of the important challenges of conventional radars. Because there is no direct-view path, the reflected echoes of the target are superposed with each other, so that the electromagnetic echo energy fluctuates sharply and sometimes even is close to cancellation, which is a fast fading characteristic caused by a multipath propagation environment. Fast fading increases the measurement error (distance error, angle error, doppler frequency error, etc.) of the radar, and when the echo energy is seriously attenuated, the radar completely loses the detection and tracking ability for the target. The detection of non-direct-view objects is difficult, and fast fading, which is mainly due to a multipath propagation environment, has a complex characteristic of fast time variation. For target detection in multipath propagation environment, the main measures currently taken in the industry are to clarify the influence caused by multipath and then take measures to eliminate or avoid the influence.
The study of the detection performance of the radar targets of different systems in the low-altitude multipath environment is carried out by Zhouhao, Hu nationality Hei et al in the research on the detection performance of the radar targets in the low-altitude multipath environment, the detection probability of the radars of different systems in the low-altitude multipath environment is deduced by establishing a radar receiving signal model in the low-altitude multipath environment and adopting a characteristic function method, the purpose of achieving the optimal detection performance in the multipath environment by optimally configuring radar array elements is provided, and the result shows that the multi-input single-output radar has the advantages of both a phased array radar and a multi-input multi-output radar and has stable detection performance in the multipath environment. The low-altitude target radar detection method achieves the purpose of relieving the multipath influence by changing the number of radar array elements, does not solve the problem of low-altitude target detection from the perspective of eliminating the multipath influence, still has the influence caused by multipath, and has poor performance.
In the sky wave over-the-horizon radar multi-path channel equalization method, Wang summons, Zhan and Zhang Yu, aiming at the problem of the pollution of the multi-path effect to the radar echo signal in the operation of the sky wave over-the-horizon radar, a channel equalization method for inhibiting the multi-path pollution is provided. The method is a method for determining the influence caused by multipath and then eliminating the influence by equalization, needs to estimate the multipath channel in real time and has the characteristics of high complexity and poor real-time property.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a radar detection method for shielding targets under complex terrains based on deep learning, which has higher target identification rate and better instantaneity, universality and simplicity.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps: firstly, acquiring corresponding multipath propagation characteristics of a detection target at different positions by using a channel modeling and channel measuring method, and establishing a deep learning training data set; then, training and learning the target position and the multipath propagation characteristics by utilizing deep learning, and further establishing a correlation model of the target position and the multipath propagation characteristics; and finally, the echo multipath signal characteristics of the unknown target position are used as input to obtain the position information of the shielding target.
The channel modeling and channel measurement firstly set the geographic environment of the target to obtain a three-dimensional electronic map of the environment, the geographic environment is divided into different vertical grids according to space, the detection target is respectively arranged at the center of each vertical grid, the position of a radar detector is fixed, and the channel modeling is carried out by utilizing a ray tracing model to obtain the channel impulse response of the target in the different vertical grids; then setting a real geographic environment where the target is located, dividing the geographic environment into different vertical grids according to space, respectively placing the target in the center of each vertical grid, fixing the position of a radar detector, and obtaining channel impulse responses of the target in the different vertical grids by using a time domain or frequency domain channel measurement method; and carrying out discretization sampling on the simulated channel impulse response obtained by channel modeling or actual measurement, and obtaining the discretization channel impulse response by utilizing a CLEAN algorithm to serve as a training data set.
In the training and learning, designing hyper-parameters of a deep neural network DNN, adopting a self-adaptive moment estimation method as an optimizer of a loss function, adopting a Relu function as an excitation function of a hidden layer and an output layer, and adopting a mean square error loss function; feeding a training data set into a designed DNN model, wherein each sample training data x corresponds to a label y*The network firstly carries out the calculation process of forward propagation, finally generates output y by activating a function, and defines a loss function L to quantize y and y*Then using back propagation algorithm to calculate gradient, and minimizing loss function by gradient descentThrough the continuous forward-backward propagation process, the weights and bias values between the neurons in the network are continuously adjusted and updated along with the increase of the training times until the model converges.
The invention has the beneficial effects that: and acquiring corresponding multipath propagation characteristics of the detection target at different positions by using a channel modeling and channel measuring method, and establishing a deep learning training data set. And training and learning the target position and the multipath propagation characteristics by utilizing deep learning, and further establishing a correlation model of the target position and the multipath propagation characteristics. Under the condition of an actual multipath characteristic input model, the position information of the target can be rapidly and accurately output. Compared with the traditional radar detection method for shielding the target, the method has the advantages of higher target identification rate, better real-time performance, universality and simplicity.
Drawings
FIG. 1 is a schematic diagram of a ray tracing model;
FIG. 2 is a three-dimensional electronic map of an urban environment;
FIG. 3 is a diagram of channel time domain measurements;
FIG. 4 is a schematic diagram of channel frequency domain measurements;
FIG. 5 is a schematic diagram of a training sample;
FIG. 6 is a diagram of a neural network model.
Detailed Description
The invention provides a radar detection method for a sheltered target under a complex terrain based on deep learning. The corresponding relation exists between the target position and the multipath characteristics, firstly, a specific multipath propagation channel is generated under a given geographic environment by utilizing a channel modeling and channel measuring method in wireless communication, and the specific multipath propagation channel comprises the number of paths, the echo signal strength, the phase and the time delay characteristics of each path and the like. When the detection target is at different positions, corresponding multipath propagation characteristics are different, and a target position database and a multipath propagation characteristic database are established according to the multipath propagation characteristics and serve as a training data set for deep learning. The target position and the multipath propagation characteristics are trained and learned by utilizing deep learning, so that a correlation model of the target position and the multipath propagation characteristics is established, and the position information of the target can be rapidly and accurately output under the condition that the actual multipath characteristics are input into the model. Compared with the traditional radar detection method for shielding the target, the method has the advantages of higher target identification rate, better real-time performance, universality and simplicity.
The invention solves the technical problem by adopting the following specific steps:
(1) data acquisition
In order to obtain a training sample of the deep neural network, the multipath channel response characteristics of the environment where the mask target is located need to be accurately obtained. And data acquisition is carried out by utilizing a channel modeling and channel measuring method.
The method comprises the steps of firstly, setting a complex shielded geographic environment where a target is located to obtain a three-dimensional electronic map of the environment, dividing the geographic environment into different vertical grids according to space, respectively placing a detection target in the center of each vertical grid, fixing the position of a radar detector, and performing channel modeling by using a ray tracing model to obtain channel impulse responses of the target in the different vertical grids; then setting a complex shielded real geographical environment where the target is located, dividing the geographical environment into different vertical grids according to space, respectively placing the target in the center of each vertical grid, fixing the position of a radar detector, and obtaining channel impulse responses of the target in the different vertical grids by using a time domain or frequency domain channel measurement method;
(2) data processing
Discretizing and sampling the analog channel impulse response obtained by channel modeling or actual measurement, and obtaining the discrete channel impulse response by utilizing a CLEAN algorithm, namely channel characteristic parameters (obvious multipath number, amplitude, phase and time delay);
(3) constructing a deep neural network model
The Deep Neural Network (DNN), also known as a multilayer perceptron, simulates the human brain neural network to process and memorize information, and the more layers, the deeper the features can be expressed more deeply, and the stronger the function simulation capability can be possessed. The leftmost and rightmost sides of the network are referred to as the input and output layers, respectively, and the middle is referred to as the hidden layer. Each layer has a considerable number of neurons, each neuron having an output, which isA non-linear function of the weighted sum of neurons in the previous layer. Assuming that the input of a four-layer network is x and the network output is y, the input and output satisfy a function y ═ f (x) ═ f4(f3(f2(f1(x) ))) where the function f) is implemented1(x)，f2(x)，f3(x)，f4(x) Respectively, the nonlinear activation functions of the network layers.
The method is based on a DNN method to train a target detection model. The hyper-parameters of the deep neural network need to be designed in advance before training, and the hyper-parameters are parameters set before the DNN starts a learning process, so that the convergence speed and the classification effect of the DNN can be influenced. The hyper-parameters in the neural network include: the number of network layers, the number of nodes in each layer, the selection of an activation function, the learning rate, the size of batch data, the number of network training times, the size of test set and training set data, and the selection of a loss function. In the current deep learning application field, the configuration of the hyper-parameters is also an empirical process, so that constant parameter adjustment is usually required to obtain the best result. The optimal hyper-parameter is determined through multiple parameter adjustment comparisons. The method adopts an adaptive moment estimation (Adam) as an optimizer of a loss function; adopting Relu (Rectifeldinear units) function as the excitation function of the hidden layer and the output layer; a Mean Squared Error (MSE) loss function is used. And each group of channel characteristics is used as an input characteristic sample, the target three-dimensional coordinate position is used as a corresponding label, and the deep neural network is trained.
The training process is as follows:
firstly, the characteristic parameters of the discrete channel impact response obtained in the step (2) are taken as training set data to be fed into a designed DNN model. Each sample training data x will have a corresponding label y*。
The network firstly carries out a forward propagation calculation process, namely, the input x of the deep neural network is weighted and operated through the neuron nodes of each layer, and finally the output y is generated through an activation function. This is a forward propagation process of DNN, and since the output result of DNN has an error with the actual result, in order to represent output y and corresponding tag y*Defining a loss function L to quantify the error between the two
The training target of the network is to minimize a loss function through gradient descent, and the loss function is used for measuring the shielding target position information y and the actual target position information y predicted by the network*The proximity of (a). Deep neural networks typically compute gradients using a back-propagation (BP) algorithm that utilizes a chain rule for faster computation by the network. The network propagates errors backwards from the output layer through a BP algorithm layer by layer, the errors are reversely propagated from the output layer to the hidden layer until the errors are propagated to the input layer, and if the current loss function is large, the network automatically adjusts the weight and the bias according to the set super-parameters to reduce the loss function. Through the continuous forward-backward propagation process, the weights and bias values between neurons in the network are continuously adjusted and updated with the increase of training times until the model converges.
(4) Target location identification
And using the trained DNN model and taking the echo multipath signal characteristics of the unknown target position as input to obtain the position information of the shielding target.
The present invention will be further described with reference to the following drawings and examples, which include, but are not limited to, the following examples.
An embodiment of the invention comprises the following steps:
(1) data acquisition
In order to obtain a training set sample of the deep neural network, the multipath channel response characteristics of the environment where the mask target is located need to be accurately obtained. And data acquisition is carried out by utilizing a channel modeling and channel measuring method. Setting a complex shielding geographic environment where a target is located to obtain a three-dimensional electronic map of the environment, dividing the geographic environment into different vertical grids according to space, respectively placing the detection target in the center of each vertical grid, fixing the position of a radar detector, and performing channel modeling by using a ray tracing model to obtain channel impulse responses of the target in the different vertical grids.
In this example, a channel modeling method is selected for data acquisition. As shown in fig. 1, taking a teaching area scene of 1kmX1km X1km in a university campus as an example, the teaching area scene is spatially divided into different cubic squares, and the size of each grid is set to be 10m X10 m X10 m, that is, each dimension is divided into 100 equally, and a total of 1000000 cubic squares. Respectively placing the target in the center of each cube, fixing the position of a radar detector, performing channel modeling by using a ray tracing model, and obtaining channel impulse responses of the target in different cube grids by adopting a time domain or frequency domain channel measuring method; finally, we collected a data set of 1000000 sample data, each consisting of a channel impulse response at the mask target and its corresponding actual position information.
(2) Data pre-processing
1) And (3) training set processing:
the training data set used for deep learning of the invention consists of a target position database and a multipath propagation characteristic database.
Firstly, making discretization sampling on the analog channel impulse response obtained by channel measurement, utilizing CLEAN algorithm to obtain discrete channel impulse response, and making characteristic extraction to obtain multipath propagation characteristics, in the invention, all multipath channel impulse responses measured at every position are sorted according to signal attenuation size, and the multipath propagation characteristics of first five multipath channels are only taken as input of neural network, i.e. channel characteristics h ═ α, p, t }, in which α ═ { a ═ a { (α, p, t }, in which1,a2,…,a5Represents the normalized signal amplitude of five multipath channels, wherein, 0 is less than or equal to ai≤1，i＝1,2,3,4,5；p＝{p1,p2,…,p5Denotes the normalized signal phase, 0 ≦ pi≤1；t＝{t1,t2,…,t5Represents the time delay of the normalized signal, and t is more than or equal to 0iLess than or equal to 1; the size of the database of multipath propagation characteristics is set to 1000000.
Then aiming at the channel characteristics at different positions, three-dimensional coordinate information of the positions of the shielding targets in different vertical squaresEstablishing a one-to-one correspondence, e.g. multipath propagation characteristicsMth channel characteristic h in the librarymThe corresponding three-dimensional coordinate information is the mth coordinate information in the target position database
Wherein
Representing the coordinates of the occluding object in the X dimension,
representing the coordinates in the Y-dimension,
represents coordinates in the Z dimension, wherein
The size of the target location database is set to 1000000.
2) Processing the test set:
randomly selecting 20% of samples from the training set to form a test set. And testing the generalization ability of the DNN model after the model is trained.
(3) Constructing a deep neural network model
Deep learning is used as a method for extracting and classifying data nonlinear features, and low-level features are combined through multiple layers of hidden layers to form more abstract high-level category attributes or features, so that the method has strong feature expression capability and complex task modeling capability. When the monitored target moves under a complex environment, a multipath phenomenon related to the target exists, and due to the change of the environment, especially the change of the relative position relationship, the multipath following the target is changed. Therefore, there is a corresponding relationship between the target position and the multipath characteristics, but the corresponding relationship does not have a specific functional relation at present, and fast fading mainly originating from the multipath propagation environment has a complex characteristic of fast time variation. Therefore, the correlation model of the target position and the multipath propagation characteristics can be automatically deduced from a big data training set through continuous iterative training by utilizing the modeling capability of deep learning on the complex function relationship, and the position information of the target can be rapidly and accurately output under the condition that the actual multipath characteristics are input into the model, so that the radar detection of the shielded target under the complex terrain is realized.
Through multiple parameter adjustment attempts, the optimal hyper-parameter obtained by the method is set as follows, the network layer number of the DNN model is set to be 8 layers, and the DNN model consists of an input layer, 6 hidden layers and an output layer. The neuron node of the input layer is 15, which respectively corresponds to the multipath number, amplitude, phase and time delay; hidden layer nodes are 512, 256, 128, 64, 32 and 16; the output layer node is 3, and corresponds to the three-dimensional coordinate of the shielding target in the actual environment. The activation functions of the hidden and output layers are Relu functions, i.e. frMax (x,0), which can increase the non-linearity and speed up the convergence of the system; the loss function is MSE function defined as
N is the number of neuron nodes corresponding to the output layer, wherein N is 3, and L is in an ideal size range of 0-0.01, and when L is in the range, the prediction error can be considered to meet the allowable error range; the optimizer selects an Adam optimizer, wherein Adam is a method capable of adaptively learning the learning rate for different parameters, is suitable for large data and high-dimensional space, and has the advantages of low memory requirement and the like; the batch size (batch size) of each feed to the network is 256; the learning rate is set to 0.001; the training times of the network are 100000 times; the training test set and test data set sizes were 800000 and 200000, respectively.
Software configuration of the present invention: the programming language python 3.0, the deep learning platform used is the Tensorflow supported by Google. Hardware configuration, the CPU is an Inter Xeon Gold 5118 and uses GTX NVIDIA 1080Ti for acceleration training.
And training and learning the target position and the multipath propagation characteristic data set by deep learning, and further establishing a correlation model of the target position and the multipath propagation characteristic.
Training process:
1) all hyper-parameters in the network are first initialized. Then the training set data after data preprocessing is sent to the DN designed by usAnd starting training in the N model. Each channel characteristic sample h has corresponding three-dimensional coordinate position information y*。
2) The network firstly carries out a calculation process of forward propagation, namely the input channel characteristic h of the deep neural network is weighted and calculated through the neuron nodes of each layer, and finally the output predicted three-dimensional coordinate position information y is generated through an activation function. The training is started with a large error between them, so that a loss function is defined
For quantifying the error between the two, the training objective of the network is to reduce the loss function by a gradient descent algorithm so as to lead the network to predict the position information y of the shielding object and the position information y of the actual object*As close as possible. The network propagates the error backward from the output layer through the BP algorithm, the error propagates backward from the output layer to the hidden layer until it propagates to the input layer, at which point the network computes and updates the weights and offsets to reduce the loss function. Through different forward-backward propagation processes, parameters (weight and bias value) among neurons in the network are continuously adjusted and updated along with the increase of the training times until the model converges or the training times is finished, and when the final loss function meets an error range, the complex function relationship between the multipath channel characteristics of the shielding target under the complex environment and the corresponding position of the shielding target is automatically deduced from the data set by the network.
(4) Target location identification
And storing the trained DNN model, and taking the echo multipath signal characteristic h of the unknown target position as input to obtain the position information y of the shielding target predicted by the network. It is compared with the actual position y of the target*And (6) carrying out comparison. Through multiple sets of tests, the output of the DNN meets the allowable error range, and the DNN-based radar detection method for the shielded target under the complex terrain is proved to be feasible. Compared with the traditional radar detection method for shielding the target, the method can quickly and accurately output the position information of the target, and has higher target identification rate, better instantaneity, universality and simplicity.
Claims (3)
1. A radar detection method for a sheltered target under complex terrain is characterized by comprising the following steps: firstly, acquiring corresponding multipath propagation characteristics of a detection target at different positions by using a channel modeling and channel measuring method, and establishing a deep learning training data set; then, training and learning the target position and the multipath propagation characteristics by utilizing deep learning, and further establishing a correlation model of the target position and the multipath propagation characteristics; and finally, the echo multipath signal characteristics of the unknown target position are used as input to obtain the position information of the shielding target.
2. The radar detection method of a shaded object under complex terrain according to claim 1, wherein: the channel modeling and channel measurement firstly set the geographic environment of the target to obtain a three-dimensional electronic map of the environment, the geographic environment is divided into different vertical grids according to space, the detection target is respectively arranged at the center of each vertical grid, the position of a radar detector is fixed, and the channel modeling is carried out by utilizing a ray tracing model to obtain the channel impulse response of the target in the different vertical grids; then setting a real geographic environment where the target is located, dividing the geographic environment into different vertical grids according to space, respectively placing the target in the center of each vertical grid, fixing the position of a radar detector, and obtaining channel impulse responses of the target in the different vertical grids by using a time domain or frequency domain channel measurement method; and carrying out discretization sampling on the simulated channel impulse response obtained by channel modeling or actual measurement, and obtaining the discretization channel impulse response by utilizing a CLEAN algorithm to serve as a training data set.
3. The radar detection method of a shaded object under complex terrain according to claim 1, wherein: in the training and learning, designing hyper-parameters of a deep neural network DNN, adopting a self-adaptive moment estimation method as an optimizer of a loss function, adopting a Relu function as an excitation function of a hidden layer and an output layer, and adopting a mean square error loss function; feeding a training data set into a designed DNN model, and training x pairs of data for each sampleShould be a label y*The network firstly carries out the calculation process of forward propagation, finally generates output y by activating a function, and defines a loss function L to quantize y and y*The gradient is calculated by using a back propagation algorithm, the loss function is minimized by gradient reduction, and the weight and the bias value between the neurons in the network are continuously adjusted and updated along with the increase of the training times through the continuous forward-back propagation process until the model converges.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911079151.XA CN110764084A (en) | 2019-11-07 | 2019-11-07 | Radar detection method for shielding target under complex terrain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911079151.XA CN110764084A (en) | 2019-11-07 | 2019-11-07 | Radar detection method for shielding target under complex terrain |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110764084A true CN110764084A (en) | 2020-02-07 |
Family
ID=69336593
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911079151.XA Pending CN110764084A (en) | 2019-11-07 | 2019-11-07 | Radar detection method for shielding target under complex terrain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110764084A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113009448A (en) * | 2021-03-09 | 2021-06-22 | 森思泰克河北科技有限公司 | Method, device, equipment and storage medium for detecting multipath target |
| CN114779202A (en) * | 2022-06-17 | 2022-07-22 | 之江实验室 | Method, device and system for detecting, identifying and positioning forest shielding target |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011112618A (en) * | 2009-11-30 | 2011-06-09 | Fujitsu Ltd | Radar system and detection method |
| US20160132768A1 (en) * | 2014-11-10 | 2016-05-12 | The Boeing Company | Systems and methods for training multipath filtering systems |
| CN107678022A (en) * | 2017-10-20 | 2018-02-09 | 西安电子科技大学 | The radar Direct Three-dimensional space target positioning method utilized based on multipath |
| CN108696932A (en) * | 2018-04-09 | 2018-10-23 | 西安交通大学 | It is a kind of using CSI multipaths and the outdoor fingerprint positioning method of machine learning |
| CN109922432A (en) * | 2019-03-28 | 2019-06-21 | 南京邮电大学 | Pass through the object localization method of optimization fingerprint elements number under wireless communications environment |
| US20190279366A1 (en) * | 2018-03-12 | 2019-09-12 | Zf Friedrichshafen Ag | Object identification using radar data |
-
2019
- 2019-11-07 CN CN201911079151.XA patent/CN110764084A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011112618A (en) * | 2009-11-30 | 2011-06-09 | Fujitsu Ltd | Radar system and detection method |
| US20160132768A1 (en) * | 2014-11-10 | 2016-05-12 | The Boeing Company | Systems and methods for training multipath filtering systems |
| CN107678022A (en) * | 2017-10-20 | 2018-02-09 | 西安电子科技大学 | The radar Direct Three-dimensional space target positioning method utilized based on multipath |
| US20190279366A1 (en) * | 2018-03-12 | 2019-09-12 | Zf Friedrichshafen Ag | Object identification using radar data |
| CN108696932A (en) * | 2018-04-09 | 2018-10-23 | 西安交通大学 | It is a kind of using CSI multipaths and the outdoor fingerprint positioning method of machine learning |
| CN109922432A (en) * | 2019-03-28 | 2019-06-21 | 南京邮电大学 | Pass through the object localization method of optimization fingerprint elements number under wireless communications environment |
Non-Patent Citations (3)
| Title |
|---|
| 孙晶: "基于深度学习和位置指纹的RFID室内定位算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
| 李文杰等: "基于支持向量机的位置相关计算", 《计算机仿真》 * |
| 高欢斌: "基于深度学习的室内定位算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113009448A (en) * | 2021-03-09 | 2021-06-22 | 森思泰克河北科技有限公司 | Method, device, equipment and storage medium for detecting multipath target |
| CN113009448B (en) * | 2021-03-09 | 2022-12-06 | 森思泰克河北科技有限公司 | Method, device, equipment and storage medium for detecting multipath target |
| CN114779202A (en) * | 2022-06-17 | 2022-07-22 | 之江实验室 | Method, device and system for detecting, identifying and positioning forest shielding target |
| CN114779202B (en) * | 2022-06-17 | 2022-10-04 | 之江实验室 | Method, device and system for detecting, identifying and positioning forest shielding target |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109993280B (en) | Underwater sound source positioning method based on deep learning | |
| US11246010B2 (en) | Method and system for positioning and tracking using a plurality of sensors | |
| CN111160176B (en) | Fusion feature-based ground radar target classification method for one-dimensional convolutional neural network | |
| US10275707B2 (en) | Systems and methods for training multipath filtering systems | |
| CN113568055B (en) | Aviation transient electromagnetic data inversion method based on LSTM network | |
| Darzikolaei et al. | Classification of radar clutters with artificial neural network | |
| CN112859014A (en) | Radar interference suppression method, device and medium based on radar signal sorting | |
| CN110764084A (en) | Radar detection method for shielding target under complex terrain | |
| CN112001433A (en) | Flight path association method, system, equipment and readable storage medium | |
| CN115685096B (en) | Secondary radar side lobe suppression method based on logistic regression | |
| CN111983619A (en) | Underwater acoustic target forward scattering acoustic disturbance positioning method based on transfer learning | |
| CN113472415B (en) | Signal arrival angle estimation method and device, electronic equipment and storage medium | |
| Tian et al. | Performance evaluation of deception against synthetic aperture radar based on multifeature fusion | |
| US5488589A (en) | Neural network based three dimensional ocean modeler | |
| Qin et al. | Underwater acoustic source localization using LSTM neural network | |
| Li et al. | Sea/land clutter recognition for over-the-horizon radar via deep CNN | |
| Venkatasubramanian et al. | Toward data-driven stap radar | |
| CN111368653A (en) | Low-altitude small target detection method based on R-D (R-D) graph and deep neural network | |
| Zaarour et al. | Comparative experimental study on modeling the path loss of an UWB channel in a mine environment using MLP and RBF neural networks | |
| Liu et al. | Artificial neural networks for estimation and fusion in long-haul sensor networks | |
| CN113657593B (en) | Plasma parameter diagnosis method based on BP neural network | |
| CN112666528B (en) | Multi-station radar system interference identification method based on convolutional neural network | |
| CN115062526A (en) | Deep learning-based three-dimensional ionosphere electron concentration distribution model training method | |
| de Sá et al. | Lightning distance estimation using LF lightning Radio signals via analytical and machine-learned models | |
| Li et al. | Aerospace target identification-Comparison between the matching score approach and the neural network approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200207 |