CN113064117B - Radiation source positioning method and device based on deep learning - Google Patents
Radiation source positioning method and device based on deep learning Download PDFInfo
- Publication number
- CN113064117B CN113064117B CN202110269610.1A CN202110269610A CN113064117B CN 113064117 B CN113064117 B CN 113064117B CN 202110269610 A CN202110269610 A CN 202110269610A CN 113064117 B CN113064117 B CN 113064117B
- Authority
- CN
- China
- Prior art keywords
- model
- receiver
- radiation source
- map
- coordinates
- 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.)
- Active
Links
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
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/04—Position of source determined by a plurality of spaced direction-finders
-
- 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
- G01S1/00—Beacons or beacon systems transmitting signals having a characteristic or characteristics capable of being detected by non-directional receivers and defining directions, positions, or position lines fixed relatively to the beacon transmitters; Receivers co-operating therewith
- G01S1/02—Beacons or beacon systems transmitting signals having a characteristic or characteristics capable of being detected by non-directional receivers and defining directions, positions, or position lines fixed relatively to the beacon transmitters; Receivers co-operating therewith using radio waves
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention provides a radiation source positioning method and device based on deep learning, wherein the method comprises the following steps: obtaining a map model of a sample area, wherein the map model comprises: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building; acquiring the receiving intensity of a receiver with set coordinates in a map model by using a ray tracing method based on signals emitted by a radiation source; constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver; a target neural network model is obtained by utilizing the neural network model pre-constructed by the sample set; and acquiring the receiving intensity of a receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by using the target neural network model. By applying the method provided by the invention, the positioning accuracy of the radiation source can be improved.
Description
Technical Field
The invention relates to the technical field of electromagnetic signal detection, in particular to a radiation source positioning method and device based on deep learning.
Background
The interaction of the various signal radiation sources in a certain geographical space creates a complex electromagnetic environment. Electromagnetic environment refers to the situation of electromagnetic energy distribution due to electromagnetic signals of various radiation sources and their interactions within a certain geographical space and a specific time frame. The electromagnetic environment formed by fewer radiation sources is generally simpler within a certain space, and the distribution of electromagnetic energy in the space is also more regular. However, in the case of multiple radiation sources, various disturbances may occur between the different radiation sources, which disturbances vary correspondingly with the variation of the radiation sources, thus resulting in a very complex electromagnetic environment in space. Furthermore, the presence of one or more radiation sources may affect the proper operation of the other radiation source or sources, which are called interference (radiation) sources. Particularly in urban communication environment, the existence of an interference source seriously weakens the communication effect of a communication base station and reduces the communication quality of users, so that how to eliminate the interference source is a technical problem to be solved. From a technical point of view, the premise of eliminating the interference source is to find the spatial position of the interference source. Conventional radiation source positioning methods include both active positioning and passive positioning. The active positioning refers to a radiation source positioning method based on deep learning, wherein the positioning detector radiates electromagnetic waves outwards, and then judges the position of an object according to reflected electromagnetic wave information. Radar is the most common active positioning equipment, and by sending radar waves and receiving radar echoes, the running speed and running track of an object can be judged according to information such as wavelength, time and the like. Such positioning methods are often used for military use because of their high positioning accuracy and wide application space, but have the disadvantage of easily exposing their own space and striking them. In order to solve the above-mentioned problems, passive positioning technology has been developed, and passive positioning technology refers to a positioning method that only relies on receiving electromagnetic radiation signals of the other party to know the position of the radiation source. And obtaining the spatial position of the target radiation source through a series of positioning solutions according to the information such as the incoming wave direction, the frequency, the arrival time and the like of the electromagnetic signals of the received interference source or the enemy radiation source. The positioning mode has good concealment, does not produce secondary electromagnetic pollution to the surrounding environment, and has great application value. For passive positioning, the positioning calculation algorithm of the radiation source is most important. Commonly used are position solutions using time difference of arrival (Time Difference of Arrival, TDOA), frequency difference of arrival (Frequency Difference of Arrival, FDOA). The inventors have found that existing passive positioning methods are suitable for large-range, long-range, low-precision probing. In local urban areas, particularly in indoor environments, how to perform high-precision positioning of civil interference sources is a technical problem to be solved urgently.
In 2016, a wireless positioning method based on a deep neural network (Deep Neural Networks, DNN) has been proposed, and for the variation and unpredictability of wireless signals, a four-layer DNN structure is used for positioning, and the structure is pre-trained by stacked denoising self-encoding (Stacked Denoising Auto encoder, SDA), so that reliable features can be learned from a large number of noise samples, and manual engineering is avoided. Meanwhile, in order to maintain the time coherence, a fine positioning algorithm based on a hidden Markov model (Hidden Markov Model, HMM) is provided, and initial positioning estimation obtained by a coarse positioning algorithm based on DNN is smoothed. The data required for the experiment are collected from the real world at different times to meet the requirements of the actual environment. Experimental results show that the system has remarkable improvement on the positioning accuracy of wireless signals. In addition, scholars have also proposed a multi-story building WiFi fingerprint indoor positioning system based on convolutional neural networks: a new classification model and a new positioning model are designed with a one-dimensional CNN stacked automatic encoder (Stacked Auto Encoder, SAE) that uses SAE to extract features from the received signal strength RSSI data while training the CNN effectively achieves high accuracy in the positioning phase. In 2019, RNN networks have also been applied to the field of wireless positioning, not to locate the position of one mobile user at a time as in conventional positioning algorithms, but to track positioning, taking into account the correlation between the received signal strength indicator, RSSI, measurements in the track. In order to improve the accuracy of the RSSI time fluctuations, a weighted average data filter is proposed that inputs RSSI data and sequentially outputs positions. They tested calculations using different types of recurrent neural networks, such as LSTM, gated loop units (Gate Recurrent Unit, GRU) and BRNN, etc., and the experiments showed an improvement of about 30% over CNN and probabilistic algorithms under the same test environment.
However, the algorithm is used for training the model based on electromagnetic data measured in a real environment, the sample collection cost is high, the sample number is small, the training precision of the model is low, and finally the technical problem of poor positioning precision of the radiation source is solved
Disclosure of Invention
The technical problem to be solved by the invention is how to improve the positioning accuracy of the radiation source.
The invention solves the technical problems by the following technical means:
the invention provides a radiation source positioning method based on deep learning, which comprises the following steps:
step S101, obtaining a map model of a sample area, where the map model includes: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building;
step S102, based on signals emitted by a radiation source, acquiring the receiving intensity of a receiver with set coordinates in a map model by using a ray tracing method;
step S103, a sample set is constructed according to the radiation source coordinates, the transmitting power, the receiving intensity of a corresponding receiver and the receiving coordinates of the receiver;
step S104, training a pre-constructed neural network model by using the sample set to obtain a target neural network model;
Step S105, obtaining the receiving intensity of the receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by using the target neural network model.
Optionally, the step S101 of obtaining a map model of the sample area includes:
acquiring an orthographic image and elevation data of a sample area;
and establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data.
Optionally, in order to further improve the accuracy of the three-dimensional map model of the sample area and improve the accuracy of ray tracing in step S102, an unmanned aerial vehicle carrying a laser radar is adopted to respectively draw a point cloud around each object in the sample area, and then a neural network algorithm trained in advance for identifying electromagnetic medium types of each part in the object is used to identify the electromagnetic medium types of each part in the object, where the electromagnetic medium types include: glass, concrete, grass, trees, ground. And then constructing a three-dimensional model of each object according to the electromagnetic medium type corresponding to each point cloud as a label of the point cloud, and drawing a three-dimensional model of the sample area.
Optionally, the acquiring, by using a ray tracing method, the receiving intensity of the receiver with the set coordinates in the map model based on the signal emitted by the radiation source includes:
setting a receiver on each coordinate of the map model for each transmitting antenna parameter;
and aiming at the receiver on each coordinate, acquiring signals emitted by the radiation sources arranged on each set coordinate in the map model by using a ray tracing method to obtain corresponding receiving intensity, namely distributing a plurality of radiation sources for the receiver and calculating the receiving intensity at the receiver by using the ray tracing method.
Optionally, the assigning a number of radiation sources to the receiver includes:
and respectively setting the radiation sources of each quantity to each set position in the map model according to each quantity of the radiation sources distributed to the same coordinate receiver to obtain a plurality of distribution combinations of the radiation sources, and pairing each distribution combination of the radiation sources with the receiver.
Optionally, the constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver includes:
according to the transmitting power and the coordinates of each radiation source, the receiving coordinates and the receiving intensity of a receiver, mapping the transmitting power and the transmitting coordinates of the radiation source when the receiving intensity is zero into a map model to obtain an intrinsic situation map of the receiver;
Is provided with a radiation source at (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L The definition is as follows:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x，y，z)
wherein the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene;
therefore, the physical meaning of the eigenstate E of the receiver is: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the emission power of all possible radiation sources when the reception intensity of the receiver is 0 dBm;
drawing a transmitting power contour map for the receiver based on the signal intensity corresponding to each coordinate point in the intrinsic situation map;
filling areas among the contour lines according to the transmission power difference between adjacent contour lines in the transmission power contour map and the sequence from high to low, so as to obtain a contour line filling map aiming at the receiver;
and taking the contour filling map as a sample to construct a sample set.
Optionally, in step S105, the neural network model adopts a combination of a convolutional neural network CNN, a pooling layer and a fully-connected network layer, uses the CNN to automatically extract various input features of the combined sample, and the pooling layer performs high-dimensional extraction on the features, and finally outputs a result through the fully-connected layer.
The invention also provides a radiation source positioning device based on deep learning, which comprises:
an acquisition module, configured to acquire a map model of a sample area, where the map model includes: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building;
the receiving intensity module is used for acquiring the receiving intensity of a receiver with set coordinates in the map model by using a ray tracing method based on the signal emitted by the radiation source;
the construction module is used for constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver;
the training module is used for obtaining a target neural network model by utilizing the neural network model pre-constructed by the sample set;
the identification module is used for acquiring the receiving intensity of the receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by utilizing the target neural network model.
Optionally, the acquiring module is configured to:
acquiring an orthographic image and elevation data of a sample area;
and establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data.
Optionally, the acquiring module acquires a map model of the sample area, including:
acquiring an orthographic image and elevation data of a sample area;
establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data;
in the receiving intensity module, receivers are respectively arranged on each coordinate of the map model according to each transmitting antenna parameter;
and aiming at the receiver on each coordinate, acquiring signals emitted by the radiation sources arranged on each set coordinate in the map model by using a ray tracing method, and obtaining corresponding receiving intensity.
Optionally, the specific implementation of the sample set construction module includes;
according to the transmitting power and the coordinates of each radiation source, the receiving coordinates and the receiving intensity of a receiver, the transmitting power and the transmitting coordinates of the radiation source when the receiving intensity is zero are mapped into a map model to obtain an intrinsic situation map of the receiver, wherein the intrinsic situation map is defined as follows:
is provided with a radiation source at (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L The definition is as follows:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x，y，z)
wherein the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene;
therefore, the physical meaning of the eigenstate E of the receiver is: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the emission power of all possible radiation sources when the reception intensity of the receiver is 0 dBm;
drawing a transmitting power contour map for the receiver based on the signal intensity corresponding to each coordinate point in the intrinsic situation map;
filling areas among the contour lines according to the transmission power difference between adjacent contour lines in the transmission power contour map and the sequence from high to low, so as to obtain a contour line filling map aiming at the receiver;
and taking the contour filling map as a sample to construct a sample set.
The invention has the advantages that:
By applying the embodiment of the invention, the training sample is established by using the ray tracing method based on the electronic map model, and the radiation source positioning is realized by completely relying on a software algorithm, so that compared with the radiation source positioning realized by relying on actual measurement data in the prior art, the radiation source positioning method is easier to obtain samples with larger quantity, the training precision of the target neural network model is improved, and further, the high-precision positioning of the radiation source is realized.
Drawings
FIG. 1 is a schematic flow chart of a radiation source positioning method based on deep learning according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a radiation source positioning method based on deep learning according to an embodiment of the present invention;
FIG. 3 is an orthographic view of a method for positioning a radiation source based on deep learning according to an embodiment of the present invention;
FIG. 4 is a schematic view of a three-dimensional model of a sample region constructed in an embodiment of the present invention;
FIG. 5 is a schematic diagram of another method for positioning a radiation source based on deep learning according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of the spatial location distribution of a receiver in a three-dimensional model of a sample region in an embodiment of the present invention;
FIG. 7 is a graph showing the reception intensity of each receiver corresponding to the radiation source Tx according to the embodiment of the present invention;
FIG. 8 is a radiation pattern of a radiation source in accordance with an embodiment of the present invention;
FIG. 9 is a graph of the eigen situation when the receiver and the radiation source are shifted and the receiver received intensity is zero in an embodiment of the present invention;
FIG. 10 is a schematic diagram of the position of a radiation source when the receiver receives zero intensity according to an embodiment of the present invention;
FIG. 11 is an intrinsic situation diagram when the receiver receives zero intensity in an embodiment of the present invention;
FIG. 12 is a graph of transmit power contour for a receiver in an embodiment of the invention;
FIG. 13 is a schematic illustration of a filled contour line in an embodiment of the present invention;
FIG. 14 is a schematic view of a portion of a sample obtained in an embodiment of the present invention;
FIG. 15 is a schematic diagram of a neural network constructed in an embodiment of the present invention;
FIG. 16 is a schematic diagram of error variation of a neural network model during training according to an embodiment of the present invention;
FIG. 17 is a distribution histogram of errors in training of a neural network model in an embodiment of the present invention;
FIG. 18 is a schematic diagram of a target neural network used in an embodiment of the present invention;
FIG. 19 is a schematic diagram showing a positioning effect of a target neural network model according to an embodiment of the present invention;
Fig. 20 is a schematic diagram showing a local enlarged positioning effect of a target neural network model according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Fig. 1 is a schematic flow chart of a radiation source positioning method based on deep learning according to an embodiment of the present invention, and fig. 2 is a schematic diagram of a radiation source positioning method based on deep learning according to an embodiment of the present invention, where as shown in fig. 1 and fig. 2, the method includes:
s101: obtaining a map model of a sample area, wherein the map model comprises: building model, road model, tree model and object model, and building model includes: electromagnetic dielectric properties of various parts of the building.
For example, taking an urban area with a sample area of 525m×525m as an example, fig. 3 is an orthographic image of the urban area, which is taken as the sample area by using a satellite or an unmanned aerial vehicle with a synthetic aperture radar, as shown in fig. 3, and the corresponding contour interval is 3m, in the radiation source positioning method based on deep learning according to the embodiment of the present invention; and simultaneously acquiring corresponding elevation data, and establishing a three-dimensional map model of the sample area by utilizing the combination of the orthophoto map and the elevation data. Fig. 4 is a schematic view of a three-dimensional model of a sample area constructed in an embodiment of the present invention, as shown in fig. 4, and the outline of each building is included in fig. 4.
In the foregoing embodiment of the step, fig. 5 is a schematic diagram of another radiation source positioning method based on deep learning according to the embodiment of the present invention, as shown in fig. 5, in order to further improve the accuracy of the urban three-dimensional map model, to improve the accuracy of ray tracing in step S102, an unmanned aerial vehicle carrying a laser radar is adopted to respectively perform point cloud drawing around each object in the sample area. And then identifying the electromagnetic medium categories of each part in the object by using a pre-trained neural network algorithm for identifying the electromagnetic medium categories of each part in the object, wherein the electromagnetic medium categories comprise: glass, concrete, grass, trees, ground. And then constructing a three-dimensional model of each object according to the electromagnetic medium type corresponding to each point cloud as a label of the point cloud, and drawing a three-dimensional model of the sample area.
It should be emphasized that the sample area described in the embodiments of the present invention may be one or a combination of urban area, indoor area, forest area, mountain area.
S102: and acquiring the receiving intensity of a receiver with set coordinates in the map model by using a ray tracing method based on the signal emitted by the radiation source.
For example, assuming that the radiation source has 100 transmitting antenna parameters, for each transmitting antenna parameter, a process is performed in which, taking transmitting antenna parameters-1, 1.6ghz as an example, a receiver is respectively set on each preset coordinate in the map model obtained in step S101 to receive the radiation source signal.
Taking the receiver 501 as an example, a radiation source of the transmitting antenna parameter-1 is set at a set position in the sample area for the receiver 501, and then the radiation source is continuously shifted, and the receiver 501 receives the signal of the radiation source. In order to improve the progress, a radiation source can be arranged at a plurality of positions at one time, so that the receiver 501 and the radiation source realize pairing combination, and the receiver 501 detects the radiation source; or the plurality of radiation sources radiate the signals respectively, the receiver 501 achieves pairing combinations with the plurality of radiation sources respectively, and then calculates each pairing combination in the above manner. Further, the pairing with the receiver 501 may be achieved by two or more radiation sources, which simultaneously emit signals, and the receiving field strength of the receiver 501 may be calculated. After the calculation of the received field strength at the receiver 501 is completed, the position of the receiver 501 is transformed and the above operation is repeated until all the receiver positions and the received field strengths corresponding to the positions of the radiation sources are calculated. The radiation source of the transmit antenna parameter-2 is then switched and the above operation is repeated until all transmit antenna parameters are measured.
Further, in order to improve the calculation efficiency, a plurality of receivers may be set at a time during the calculation, fig. 6 is a schematic diagram of spatial position distribution of the receivers in the three-dimensional model of the sample area in the embodiment of the present invention, as shown in fig. 6, in the embodiment of the present invention, 32 receivers are set, and 501 is the spatial positions of the receivers. If only one radiation source is provided at this time, 32 paired combinations of one radiation source and one receiver are simultaneously produced.
In order to make the accuracy and generalization of deep learning higher, the size of the data set needs to be very suitable, and the quality needs to be high and complete. Considering specific problems and equipment performance, the number of samples of the selected data set is 8000, wherein the number of samples of the training set is 6000, and the number of samples of the testing set is 2000. FIG. 7 is a graph showing the reception intensity of each receiver corresponding to the radiation source Tx according to the embodiment of the present invention; as shown in fig. 7, the emission power of the radiation source was set to 60dBm, the frequency of the emission waveform was set to 1.6ghz, and the field intensity values at 32 receivers were positively correlated with the corresponding pixel gray values.
If 100 radiation sources are provided at the same time, 100 x 32 paired combinations of one radiation source and one receiver are produced; then each radiation source radiates signals respectively, and the receiving intensity of each receiver is calculated; in addition, the reception intensity of each receiver can be calculated when two or more radiation sources radiate signals simultaneously.
Further, considering that the number of 32 receivers is too small relative to the area of the whole sample area, in order to make full use of the information to the receivers, the choice of each receiver position should be made to avoid the positions of corners, inside the building, etc. as much as possible.
In practical applications, the method for calculating the receiving intensity at the receiver may be a UTD ray tracing algorithm; further, the location and number of receivers and radiation sources are not limited by the above examples during the calculation.
Step S102, after the urban electromagnetic model is established, the position of a radiation source and the parameters of a transmitting antenna are given, and the distribution of the field intensity corresponding to each receiver is calculated by using an urban environment electric wave propagation model; the position of the radiation source is continuously changed, calculation is repeated, and the establishment of the urban electromagnetic environment model provides a large amount of data for deep learning. And storing a large amount of data obtained by calculation to obtain an urban electromagnetic environment database.
S103: and constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver.
In the embodiment of the invention, the inventor proposes the concept of an intrinsic situation map, and the concept of the intrinsic situation map is first described below.
The radiation source is shown in (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L Can be defined as:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x, y, z), wherein,
the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene. For the purpose of explaining the loss matrix in detail, fig. 8 is a radiation situation diagram of a radiation source in an embodiment of the present invention, as shown in fig. 8, and fig. 8 is a radiation situation diagram at an altitude of 2 m; the horizontal axis and the vertical axis represent distances respectively, and the depth of a pixel point in the graph represents the receiving intensity. When the radiation source has a coordinate (55, 135,2) with a transmit power of 20.90dBm and a receive intensity at the coordinate (285, 130,2) of-22.79 dBm, the radiation source has a path loss to the receive point of 43.69dB. Meanwhile, according to electromagnetic radiation theory, for the physical model of the urban environment, the field intensity path attenuation between the two points has a value of 43.69dB, and the two values are reciprocal: i.e. if the radiation source is at (285, 130) and the transmit power is 20.90dBm, then the power received by the receiver at point (55, 135) must also be-22.79 dBm; in addition, fig. 9 is a diagram of the eigen situation when the receiver and the radiation source are shifted and the receiver receives the intensity is zero in the embodiment of the present invention, as shown in fig. 9, if the radiation source is at (285, 130), i.e. the receiver in fig. 8, and the transmission power is 43.69dBm, then the power received by the receiver must be 0dBm at the point (55, 135), i.e. the radiation source in fig. 8.
According to the analysis, the physical meaning of the eigenstate E of the receiver is defined as follows: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the transmit power of all possible radiation sources when the receiver receives an intensity of 0 dBm. As analyzed in connection with fig. 9, if the receiver is located at (55, 135) and the field strength received by this receiver is 0dBm, then one possible location of the radiation source can be determined to be at (285, 130) and the transmit power to be 43.69dBm. That is, the loss matrix of the radiation source is numerically equal to the eigenstate of the receiver, and the meaning of expression is opposite. By utilizing the theory of the intrinsic situation map, the sample calculation amount can be reduced by half, and the training efficiency of the model is improved; similarly, because it would be difficult to locate the position of the radiation source by only 32 sensors' field strength data, an eigen situation map can be added to the electromagnetic calculation results to supplement the training samples with information as a priori knowledge of deep learning.
Based on the theory, first, according to the calculation result of each pairing combination, the embodiment of the invention can map the emission power and the emission coordinates of the radiation source when the reception intensity is zero into the map model according to the emission power and the radiation source coordinates of each radiation source, the receiver reception coordinates and the reception intensity in the calculation result, and draw the positions of all possible radiation sources and the emission powers corresponding to the possible radiation sources in the eigen situation map. For example, the radiation intensity, the radiation source coordinates, the receiver coordinates and the receiving intensity of each radiation source corresponding to the receiver 501 are mapped into a map model to obtain a radiation situation map for the receiver 501, then the receiving intensity of the receiver 501 is set to zero, the radiation intensities of the corresponding radiation sources are calculated respectively, and the radiation sources are marked in the map by using pixel points with different pixel values to obtain an intrinsic situation map of the receiver. Fig. 10 is a schematic diagram of the positions of the radiation sources when the receiver receives the intensity zero in the embodiment of the present invention, as shown in fig. 10, when the position of the receiver is at 1001 of coordinates (55, 135), the field strength of the received electromagnetic wave is 0dBm, and two areas with the same transmission power exist, if the transmission power of the radiation sources is between 43dBm and 44dBm, all possible positions of the radiation sources are at 1002 of coordinates (125, 350) and 1003 of coordinates (285, 130). The white area in fig. 10 indicates that if the radiation source is in this area, its emission power must be less than 43dBm; the gray area indicates that if the radiation source is in this area, the emitted power must be greater than 44dBm; the black area in the figure indicates that if the radiation source is in this area, the emitted power is between 43dBm and 44 dBm. Thus, an eigen-situation map of the receiver located at (55, 135) can be obtained as in fig. 11: fig. 11 is an intrinsic situation diagram when the receiver receives zero intensity, as shown in fig. 11, where the black area is a building, and the higher the gray level of the color of the area outside the black area, the lower the corresponding receiving intensity, so that when the receiver receives zero intensity, there are multiple radiation sources with the same radiation intensity, and at the same time, there are more radiation sources with different radiation intensities.
Then, fig. 12 is a transmission power contour diagram for a receiver according to an embodiment of the present invention, and as shown in fig. 12, the transmission power contour diagram for the receiver is drawn based on the signal intensities corresponding to the coordinate points in the eigen situation diagram.
Then, fig. 13 is a schematic diagram of a filled contour line in the embodiment of the present invention, as shown in fig. 13, according to the difference of the transmission power between each adjacent contour line in the transmission power contour line diagram, the region between the contour lines is filled according to the order of the transmission power from high to low. For example, the area between a contour with an emission intensity of 40dBm and a contour with an emission intensity of 30dBm may be filled with a color having a gray value of 50; filling the area between the contour with the emission intensity of 40dBm and the contour with the emission intensity of 50dBm with a color with a gray value of 60; filling the area between the contour with the emission intensity of 50dBm and the contour with the emission intensity of 60dBm with a color with a gray value of 70; and so on, a contour fill map for the receiver can be obtained;
fig. 14 is a schematic view of a portion of a sample obtained in the embodiment of the present invention, and as shown in fig. 14, a sample set is constructed by using a contour filling map as a sample.
The method comprises the steps that 32 receivers are placed in a scene (a radiation source is arranged in the scene), each receiver receives electromagnetic waves emitted by the radiation source, the power loss of the electromagnetic waves when the electromagnetic waves reach different receivers is different, the power (scalar P) is recorded, then the eigenvalue matrix of each receiver is added with the received value P, the eigenvalue graphs of the 32 receivers are adjusted one by using the value P (the value of the pixel point of each eigenvalue graph is added with the field intensity value received by each receiver at the moment), and then 32 eigenvalue graphs with actual field intensity offset values can be obtained, and the 32 eigenvalue graphs correspond to one eigenvalue matrix and are used as input of a sample, and then the output of the sample is the spatial position coordinates of the radiation source at the moment. The flow chart is shown in fig. 16:
for example, 32 eigenvectors (each m n), A B … Z …
Data from 32 sensors (each scalar) received a time
β 1 β 2 ...β 32
The dimensions of the input data for sample 1 are (32, m, n)
Input matrix of sample 1
[β 1 +A
β 2 +B
…
]
S104: and obtaining a target neural network model by utilizing the neural network model pre-constructed by the sample set.
Firstly, a machine learning model framework based on a convolutional neural network and a multi-layer perceptron is built, fig. 15 is a schematic diagram of the structure of the neural network built in the embodiment of the invention, as shown in fig. 15, the embodiment of the invention adopts the combination of the convolutional neural network CNN, a pooling layer and a fully-connected network layer, various input features of a combined sample are automatically extracted by using the CNN, the pooling layer performs high-dimensional extraction on the features, and finally, a result is output through the fully-connected layer. The deep learning algorithm builds a neural network by taking TensorFlow as a rear end and Keras as a basis of a high-level abstract structure. The method mainly adopts a convolution layer, a pooling layer, a regularization layer and a full connection layer. Table 1 is a comparison of the structures of the various deep neural networks, as shown in table 1,
TABLE 1
conv3-64 represents a convolution kernel of size 3 x 3 and the convolution layer has 64 channels; similarly conv3-128 means that the size of the convolution kernel is 3 x 3 and that the convolution layer has 128 channels and 256, 512 channels of convolution layers, where all the kernels of the pooling layers have a size of 2 x 2.
In the embodiment of the present invention, the reason for using consecutive 3×3 convolution kernels instead of the larger convolution kernels (11×11,7×7,5×5) is that: for a given receptive field (local size of the input picture in relation to the output), the use of stacked small convolution kernels is preferred over the use of large convolution kernels, because multiple nonlinear layers can increase the network depth to ensure more complex patterns are learned, and at the cost of being smaller (fewer parameters). 3 3×3 convolution kernels are used to replace 7×7 convolution kernels, 2 3×3 convolution kernels are used to replace 5×5 convolution kernels, and the main purpose of this is to increase the depth of the network and to a certain extent the effect of the neural network under the condition of ensuring the same perception field.
The whole sample set is divided into a test set and a training set. Table 2 is a data set partitioning and sample dimensions, as shown in table 2,
TABLE 2
And finally, training a pre-constructed convolutional neural network model by using a training set, testing the accuracy of the convolutional neural network model by using a testing set, and obtaining a target neural network model by adjusting the layer number and parameter tuning of the model and iterating repeatedly until the accuracy of the trained convolutional neural network model reaches a set value.
In the embodiment of the invention, root mean square error (Root Mean Square Error, RMSE) is adopted as the training processIs used as a function of the loss function of the (c),wherein, the liquid crystal display device comprises a liquid crystal display device,
wherein N represents the output dimension of the sample, i.e., N is 3; p (P) 1 ，P 2 ，P 3 ) Representing a predicted radiation source position vector, R (R 1 ，R 2 ，R 3 ) Representing the actual position vector of the radiation source.
The Adamax function was chosen as the optimizer function, which has the following advantages over the direct random gradient descent method: the updating of the parameters is not affected by the expansion transformation of the gradient; the step annealing process can be naturally realized, namely, the learning rate can be automatically adjusted); the method is suitable for being applied to scenes of large-scale data and parameters, unstable objective functions and gradient sparsity or the problem that the gradient has large noise.
Fig. 16 is a schematic diagram of error change of the neural network model in the training process according to the embodiment of the present invention, as shown in fig. 16, a line 1501 is an error of the training set, and a line 1502 is an error of the verification set. When the training round reaches 700 times, the error is basically kept unchanged, the prediction effect on the training set is better than that of the verification set, and the training process of the model can be judged to belong to normal fitting according to fig. 16. Neither an excessive under-fit nor an excessive over-fit occurs. FIG. 16 shows that the model has achieved substantially optimal performance under this data set and network architecture, and that the prediction accuracy of the model has remained substantially unchanged after 700 rounds of iterative training. The predictive performance of the model is evaluated on the dataset on the test set and then the evaluation error is plotted as a distribution histogram of the error. FIG. 17 is a distribution histogram of errors in training of a neural network model according to an embodiment of the present invention, as shown in FIG. 17, the average value of the prediction errors of the model on the data set is 4.05m, the median value of the errors is 2.78m, and the duty ratio of the errors less than 5m is 79.7%.
According to the method, in the training process, network parameters such as the size of a kernel function and the number of filters of a convolutional layer of a neural network, the size of a kernel function and the pooling mode of a pooling layer, the number of full-connection layers, the number of neurons of each layer and the like are continuously searched for so that the model has higher prediction and generalization capabilities, and finally a target neural network model is obtained, and fig. 18 is a schematic diagram of a structure of the target neural network used in the embodiment of the invention, and 13 convolutional layers and multi-layer perceptrons (5 full-connection layers) are selected as structures of a radiation source position prediction network as shown in fig. 18. Table 3 is a set of parameters for each layer of the target neural network, as shown in table 3,
TABLE 3 Table 3
/>
As shown in fig. 18, the input of the target neural network of the present invention is a corrected eigenvector diagram of 32 receivers, and the output is the true position coordinates (3-dimensional vector) of the radiation source at this time. Wherein conv1 is two 64-layer convolutional networks, conv2 is two 128-layer convolutional networks, conv3 is three 256-layer convolutional networks, conv4 is three 512-layer convolutional networks, conv5 is three 512-layer convolutional networks, fc6 is a fully connected layer containing 4096 neurons, fc7 is a fully connected layer containing 1024 neurons, fc8 is a fully connected layer containing 512 neurons, fc9 is a fully connected layer containing 100 neurons, and fc10 is a fully connected layer containing 3 neurons. The total of 5 fully connected layers fc 6-fc 10 together form a network structure of the multi-layer sensor, fc6 is taken as the input of the multi-layer sensor, fc 7-fc 9 is the hidden layer (three layers) of the sensor, and fc10 is the output layer of the sensor.
S105: and acquiring the receiving intensity of a receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by using the target neural network model.
And receiving signals sent by the radiation source by using a receiver, mapping the receiving coordinates and the receiving intensity of the receiver into a map model of the area to be positioned, and then inputting the map model into a target neural network model by taking the map model as input so as to enable the target neural network to identify the position of the radiation source.
FIG. 19 is a schematic diagram showing a positioning effect of a target neural network model according to an embodiment of the present invention; FIG. 20 is a schematic diagram showing a local enlarged positioning effect of a target neural network model according to an embodiment of the present invention; as shown in fig. 19 and 20, the target neural network identifies a radiation source position 1901 as Tx '(510.9, 573.8) and an actual position 1902 of the radiation source as Tx' (509.6, 570.1), both of which have an error of 3.95m, of about 3.95/500=0.78%.
With the increasing research speed of computing systems and passive positioning in China, particularly, a series of computer hardware can be utilized to improve the detection and display performance at present, and passive positioning settlement of civil radiation sources also makes certain progress. In engineering applications, however, the available electromagnetic environment computing systems and platforms are still monopolized by foreign simulation software and positioning solution software. Therefore, it remains an important topic to delving how to build a simulation platform that simulates an electromagnetic environment and thereby develop a computing application for the positioning of the corresponding radiation source.
In addition, common radiation source positioning algorithms can be classified into conventional passive positioning and deep learning-based radiation source positioning from the standpoint of their algorithm implementation. The passive positioning technology of the radiation source mainly relies on surrounding electromagnetic wave information and other information received by a receiver to obtain other information such as spatial geographic position, speed and the like of the target radiation source. Compared with active positioning, the passive positioning has better concealment and simple structure, and a resolving model can support higher-precision result prediction. For positioning based on deep learning, the method can realize faster positioning speed, and once the model is trained, the time consumed in each radiation source positioning process is far less than that of the traditional positioning method.
Corresponding to the embodiment shown in fig. 1, the present invention further provides a radiation source positioning device based on deep learning, the device comprising:
an acquisition module, configured to acquire a map model of a sample area, where the map model includes: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building;
the receiving intensity module is used for acquiring the receiving intensity of a receiver with set coordinates in the map model by using a ray tracing method based on signals emitted by the radiation source;
The construction module is used for constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver;
the training module is used for obtaining a target neural network model by utilizing the neural network model pre-constructed by the sample set;
the identification module is used for acquiring the receiving intensity of the receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by utilizing the target neural network model.
In a specific implementation manner of the embodiment of the present invention, the obtaining a map model of a sample area in the obtaining module includes:
acquiring an orthographic image and elevation data of a sample area;
establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data;
in the receiving intensity module, receivers are respectively arranged on each coordinate of the map model according to each transmitting antenna parameter;
and aiming at the receiver on each coordinate, acquiring signals emitted by the radiation sources arranged on each set coordinate in the map model by using a ray tracing method, and obtaining corresponding receiving intensity.
In one specific implementation of the embodiment of the present invention, a specific implementation of the sample set building module includes;
According to the transmitting power and the coordinates of each radiation source, the receiving coordinates and the receiving intensity of a receiver, the transmitting power and the transmitting coordinates of the radiation source when the receiving intensity is zero are mapped into a map model to obtain an intrinsic situation map of the receiver, wherein the intrinsic situation map is defined as follows:
is provided with a radiation source at (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L The definition is as follows:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x, y, z), wherein,
the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene;
therefore, the physical meaning of the eigenstate E of the receiver is: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the emission power of all possible radiation sources when the reception intensity of the receiver is 0 dBm;
drawing a transmitting power contour map for the receiver based on the signal intensity corresponding to each coordinate point in the intrinsic situation map;
Filling areas among the contour lines according to the transmission power difference between adjacent contour lines in the transmission power contour map and the sequence from high to low, so as to obtain a contour line filling map aiming at the receiver;
and taking the contour filling map as a sample to construct a sample set.
The specific implementation of each module corresponds to each step, and the invention is not written.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. A method of deep learning-based radiation source positioning, the method comprising:
step S101, obtaining a map model of a sample area, where the map model includes: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building;
Step S102, based on signals emitted by a radiation source, acquiring the receiving intensity of a receiver with set coordinates in a map model by using a ray tracing method;
step S103, a sample set is constructed according to the radiation source coordinates, the transmitting power, the receiving intensity of a corresponding receiver and the receiving coordinates of the receiver;
the construction of the sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver comprises the following steps:
according to the transmitting power and the coordinates of each radiation source, the receiving coordinates and the receiving intensity of a receiver, the transmitting power and the transmitting coordinates of the radiation source when the receiving intensity is zero are mapped into a map model to obtain an intrinsic situation map of the receiver, wherein the intrinsic situation map is defined as follows:
is provided with a radiation source at (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L The definition is as follows:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x，y，z)
wherein the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene;
Therefore, the physical meaning of the eigenstate E of the receiver is: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the emission power of all possible radiation sources when the reception intensity of the receiver is 0 dBm;
drawing a transmitting power contour map for the receiver based on the signal intensity corresponding to each coordinate point in the intrinsic situation map;
filling areas among the contour lines according to the transmission power difference between adjacent contour lines in the transmission power contour map and the sequence from high to low, so as to obtain a contour line filling map aiming at the receiver;
taking the contour line filling map as a sample to construct a sample set;
step S104, training a pre-constructed neural network model by using the sample set to obtain a target neural network model;
step S105, obtaining the receiving intensity of the receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by using the target neural network model.
2. A method of deep learning based radiation source localization as claimed in claim 1, wherein: the step S101 of obtaining a map model of the sample area includes:
Acquiring an orthographic image and elevation data of a sample area;
and establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data.
3. A method of deep learning based radiation source localization as claimed in claim 1, wherein: in order to further improve the accuracy of the three-dimensional map model of the sample area, so as to improve the accuracy of ray tracing in step S102, an unmanned aerial vehicle carrying a laser radar is adopted to respectively perform point cloud drawing around each object in the sample area, and then a neural network algorithm which is trained in advance and is used for identifying electromagnetic medium categories of each part in the object is used for identifying the electromagnetic medium categories of each part in the object, wherein the electromagnetic medium categories comprise: glass, concrete, grass, trees, ground; and then constructing a three-dimensional model of each object according to the electromagnetic medium type corresponding to each point cloud as a label of the point cloud, and drawing a three-dimensional model of the sample area.
4. A method of deep learning based radiation source localization as claimed in claim 1, wherein: in the step S102, based on the signal emitted by the radiation source, the method for acquiring the receiving intensity of the receiver with the set coordinates in the map model by using the ray tracing method includes:
Setting a receiver on each coordinate of the map model for each transmitting antenna parameter;
and aiming at the receiver on each coordinate, acquiring signals transmitted by the radiation sources arranged on each set coordinate in the map model by using a ray tracing method to obtain corresponding receiving intensity, namely distributing a plurality of radiation sources for the receiver and calculating the receiving intensity at the receiver by using the ray tracing method.
5. A method of deep learning based radiation source localization as claimed in claim 4, wherein: said assigning a number of radiation sources to said receiver, comprising:
and respectively setting the radiation sources of each quantity to each set position in the map model according to each quantity of the radiation sources distributed to the same coordinate receiver to obtain a plurality of distribution combinations of the radiation sources, and pairing each distribution combination of the radiation sources with the receiver.
6. A method of deep learning based radiation source localization as claimed in claim 4, wherein: in step S105, the neural network model adopts a combination of a convolutional neural network CNN, a pooling layer and a fully-connected network layer, uses the CNN to automatically extract various input features of the combined sample, and the pooling layer performs high-dimensional extraction on the features, and finally outputs a result through the fully-connected layer.
7. A deep learning-based radiation source positioning device, the device comprising:
an acquisition module, configured to acquire a map model of a sample area, where the map model includes: building model, road model, tree model and object model, and building model includes: electromagnetic medium properties of various parts of the building;
the receiving intensity module is used for acquiring the receiving intensity of a receiver with set coordinates in the map model by using a ray tracing method based on the signal emitted by the radiation source;
the sample set construction module is used for constructing a sample set according to the radiation source coordinates, the transmitting power, the receiving intensity of the corresponding receiver and the receiving coordinates of the receiver;
the concrete implementation of the sample set construction module comprises;
according to the transmitting power and the coordinates of each radiation source, the receiving coordinates and the receiving intensity of a receiver, the transmitting power and the transmitting coordinates of the radiation source when the receiving intensity is zero are mapped into a map model to obtain an intrinsic situation map of the receiver, wherein the intrinsic situation map is defined as follows:
is provided with a radiation source at (x) t ，y t ，z t ) Where the transmitting power is P t The propagation attenuation at (x, y, z) is then P L (x, y, z), then the loss matrix P of the radiation source L The definition is as follows:
P L (x，y，z)＝-[P r (x，y，z)-P t ]＝P t -P r (x，y，z)
wherein the electromagnetic wave emission power level of the radiation source A is P t In dBm, its spatial coordinates are (x) t ，y t ，z t )；P r (x, y, z) is a field intensity value at any point (x, y, z) obtained by electromagnetic calculation of the radiation source A in dBm under the urban physical model structure scene;
therefore, the physical meaning of the eigenstate E of the receiver is: for a given physical model of urban environment, the eigen-situation of the receiver reflects the position and the emission power of all possible radiation sources when the reception intensity of the receiver is 0 dBm;
drawing a transmitting power contour map for the receiver based on the signal intensity corresponding to each coordinate point in the intrinsic situation map;
filling areas among the contour lines according to the transmission power difference between adjacent contour lines in the transmission power contour map and the sequence from high to low, so as to obtain a contour line filling map aiming at the receiver;
taking the contour line filling map as a sample to construct a sample set;
the training module is used for obtaining a target neural network model by utilizing the neural network model pre-constructed by the sample set;
the identification module is used for acquiring the receiving intensity of the receiver arranged in the area to be positioned and a map model of the area to be positioned, and identifying the coordinates of the radiation source by utilizing the target neural network model.
8. A deep learning based radiation source localization apparatus as claimed in claim 7, wherein: the obtaining module obtains a map model of a sample area, including:
acquiring an orthographic image and elevation data of a sample area;
establishing a three-dimensional map model of the sample area according to the orthographic image and the elevation data;
in the receiving intensity module, receivers are respectively arranged on each coordinate of the map model according to each transmitting antenna parameter;
and aiming at the receiver on each coordinate, acquiring signals emitted by the radiation sources arranged on each set coordinate in the map model by using a ray tracing method, and obtaining corresponding receiving intensity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269610.1A CN113064117B (en) | 2021-03-12 | 2021-03-12 | Radiation source positioning method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269610.1A CN113064117B (en) | 2021-03-12 | 2021-03-12 | Radiation source positioning method and device based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113064117A CN113064117A (en) | 2021-07-02 |
| CN113064117B true CN113064117B (en) | 2023-07-25 |
Family
ID=76560180
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110269610.1A Active CN113064117B (en) | 2021-03-12 | 2021-03-12 | Radiation source positioning method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113064117B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113721613B (en) * | 2021-08-23 | 2023-05-23 | 南京航空航天大学 | Robot autonomous source searching method and device based on deep reinforcement learning |
| CN113676858B (en) * | 2021-08-26 | 2024-01-02 | 杭州北斗时空研究院 | Wireless signal indoor positioning method considering non-line-of-sight identification |
| CN113848549B (en) * | 2021-09-15 | 2023-06-23 | 电子科技大学 | Radiation source positioning method based on synthetic aperture technology |
| CN114337875B (en) * | 2021-12-31 | 2024-04-02 | 中国人民解放军陆军工程大学 | Unmanned aerial vehicle group flight path optimization method for multi-radiation source tracking |
| CN114679231B (en) * | 2022-03-31 | 2022-12-13 | 中国人民解放军战略支援部队航天工程大学 | Method for acquiring space-based radio frequency map |
| WO2024000036A1 (en) * | 2022-06-29 | 2024-01-04 | The University Of Sydney | Signal strength prediction in complex environments |
| CN115754899A (en) * | 2022-09-05 | 2023-03-07 | 北京航空航天大学 | Single-station positioning method for external radiation source of test area in non-line-of-sight environment |
| CN115598592B (en) * | 2022-10-27 | 2023-09-19 | 中国电子科技集团公司信息科学研究院 | Time-frequency difference joint positioning method, system, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105005039A (en) * | 2015-07-17 | 2015-10-28 | 上海交通大学 | Satellite signal positioning method and system based on 3D modeling scene dynamic fingerprints |
| KR20190053470A (en) * | 2017-11-10 | 2019-05-20 | 주식회사 셀리지온 | Positioning system based on deep learnin and construction method thereof |
| CN110596668A (en) * | 2019-09-20 | 2019-12-20 | 中国人民解放军国防科技大学 | Target external radiation source passive positioning method based on reciprocal deep neural network |
| CN111914919A (en) * | 2020-07-24 | 2020-11-10 | 天津大学 | Open set radiation source individual identification method based on deep learning |
| CN112418245A (en) * | 2020-11-04 | 2021-02-26 | 武汉大学 | Electromagnetic emission point positioning method based on urban environment physical model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11259191B2 (en) * | 2018-11-26 | 2022-02-22 | Samsung Electronics Co., Ltd. | Methods and apparatus for coverage prediction and network optimization in 5G new radio networks |
-
2021
- 2021-03-12 CN CN202110269610.1A patent/CN113064117B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105005039A (en) * | 2015-07-17 | 2015-10-28 | 上海交通大学 | Satellite signal positioning method and system based on 3D modeling scene dynamic fingerprints |
| KR20190053470A (en) * | 2017-11-10 | 2019-05-20 | 주식회사 셀리지온 | Positioning system based on deep learnin and construction method thereof |
| CN110596668A (en) * | 2019-09-20 | 2019-12-20 | 中国人民解放军国防科技大学 | Target external radiation source passive positioning method based on reciprocal deep neural network |
| CN111914919A (en) * | 2020-07-24 | 2020-11-10 | 天津大学 | Open set radiation source individual identification method based on deep learning |
| CN112418245A (en) * | 2020-11-04 | 2021-02-26 | 武汉大学 | Electromagnetic emission point positioning method based on urban environment physical model |
Non-Patent Citations (7)
| Title |
|---|
| Ionospheric foF2 Disturbance Forecast Using Neural Network Improved by s Genetic Algorithm;Jun Zhao et al.;Advances in Space Research;全文 * |
| Neural network model for efficient localization of a number of mutually arbitrary positioned stochastic EM sources in far-field;Zoran Stankovic et al.;12th Symposium on Neural Network Applications in Electrical Engineering;全文 * |
| 基于优化神经网络算法的电离层foF2预测;胡小希等;电波科学学报;第33卷(第6期);全文 * |
| 基于机器学习的雷达辐射源识别方法综述;孟磊;曲卫;蔡凯;张婧;;兵器装备工程学报;第37卷(第10期);全文 * |
| 基于深度学习的通信辐射源个体识别技术研究;黄健航;中国优秀硕士学位论文全文数据库信息科技辑(第02期);全文 * |
| 导航自定位辐射源衰落信号智能识别算法;吉丰 等;理论与方法;第39卷(第7期);全文 * |
| 新型雷达辐射源识别;高欣宇;张文博;姬红兵;欧阳成;;中国图象图形学报(第06期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113064117A (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113064117B (en) | Radiation source positioning method and device based on deep learning | |
| CN108596101B (en) | Remote sensing image multi-target detection method based on convolutional neural network | |
| CN110012428B (en) | Indoor positioning method based on WiFi | |
| CN109891781B (en) | Method and apparatus for analyzing communication channel in consideration of material and contour of object | |
| US10820213B2 (en) | Method and apparatus for analyzing communication environment based on property information of an object | |
| CN112418245B (en) | Electromagnetic emission point positioning method based on urban environment physical model | |
| CN110225460A (en) | A kind of indoor orientation method and device based on deep neural network | |
| WO2017121168A1 (en) | Cluster-based magnetic positioning method, device and system | |
| Lee et al. | Path loss exponent prediction for outdoor millimeter wave channels through deep learning | |
| US20230024326A1 (en) | Using maps comprising covariances in multi-resolution voxels | |
| CN109327797A (en) | Mobile robot indoor locating system based on WiFi network signal | |
| Peng et al. | Deep convolutional neural network for passive RFID tag localization via joint RSSI and PDOA fingerprint features | |
| US11288861B2 (en) | Maps comprising covariances in multi-resolution voxels | |
| CN112305559A (en) | Power transmission line distance measuring method, device and system based on ground fixed-point laser radar scanning and electronic equipment | |
| CN111598028A (en) | Method for identifying earth surface vegetation distribution based on remote sensing imaging principle | |
| CN116953653A (en) | Networking echo extrapolation method based on multiband weather radar | |
| Zhang et al. | Regional Double-Layer, High-Precision Indoor Positioning System Based on iBeacon Network. | |
| CN113344954A (en) | Boundary detection method and device, computer equipment, storage medium and sensor | |
| CN114777030B (en) | Dangerous chemical gas monitoring method based on NB-IOT technology | |
| CN113066161B (en) | Modeling method of urban radio wave propagation model | |
| CN109829939A (en) | A method of it reducing multi-view images and matches corresponding image points search range | |
| CN114758364A (en) | Industrial Internet of things scene fusion positioning method and system based on deep learning | |
| CN113141570A (en) | Underground scene positioning method and device, computing equipment and computer storage medium | |
| CN115022889B (en) | RSRP prediction model of 5G network and establishment method and application thereof | |
| CN115310372B (en) | Path loss prediction method based on high-precision electromagnetic propagation model |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |