CN113341395B - Simulated radar filtering delay compensation method based on neural network - Google Patents
Simulated radar filtering delay compensation method based on neural network Download PDFInfo
- Publication number
- CN113341395B CN113341395B CN202110906174.4A CN202110906174A CN113341395B CN 113341395 B CN113341395 B CN 113341395B CN 202110906174 A CN202110906174 A CN 202110906174A CN 113341395 B CN113341395 B CN 113341395B
- Authority
- CN
- China
- Prior art keywords
- output
- neural network
- value
- filtering
- layer
- 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
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
- 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)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention discloses a method for filtering delay compensation of a simulated radar based on a neural network, which comprises the following steps: acquiring a radar height measurement value; outputting the height value of the current moment by each frame to obtain sampling data arranged at a certain time interval, and adding N data in a time sequence to obtain a height value after moving average filtering; the length N of a sliding window at the current moment is adjusted in a self-adaptive manner according to the change characteristic machine learning of the data by the neural network; when the obstacle and the unmanned aerial vehicle are identified to be in the up-and-down motion state, obtaining a height value by using the sliding average filtering with the window length of N; and when the unmanned aerial vehicle is in hovering and level flying states, obtaining a height value by using a filtering window with the window length being a preset value. The invention has fast operation speed, good real-time performance and stability, and can effectively eliminate the delay generated by the moving average filter.
Description
Technical Field
The invention belongs to the technical field of radars, and particularly relates to a method for filtering delay compensation of a simulated radar based on a neural network.
Background
In the signal processing of the stability optimization of the simulated radar height value, the signal is filtered by adopting a sliding average filter, however, the data can generate a delayed phenomenon after being filtered, the delay problem can cause the height lag acquired by the unmanned aerial vehicle flight control, when the unmanned aerial vehicle climbs to the preset height of the simulated radar at a certain speed, the current measurement height value is smaller than the actual height value due to the delay, the flight control can obtain a continuously rising instruction, so that the unmanned aerial vehicle shakes up and down near the preset height value, and the flight stability of the unmanned aerial vehicle is seriously influenced. Meanwhile, when the timeliness of the radar is not enough, the ascending instruction of the flight control in the climbing process of the unmanned aerial vehicle is executed too slowly, and therefore the hidden danger of the explosion machine colliding to the steep slope exists.
Disclosure of Invention
Aiming at the delay phenomenon after filtering, the invention provides a method for analyzing the characteristics of the moving average filter from the time domain and the frequency domain, and finds that the cut-off frequency of the moving average filter is reduced, the passband of the moving average filter is narrowed, the response of the filter is slowed down as the length of a window is increased, and finally the data delay is increased. The invention provides a neural network-based adaptive data matching method for carrying out condition judgment on an original signal, identifying whether a radar detects an obstacle or not and the motion state of an unmanned aerial vehicle, and adaptively adjusting the length of a filtering window, so that the height value detected by the radar has high stability and high real-time property. When the radar detects an obstacle and the unmanned aerial vehicle moves up and down, a filtering window with a shorter length is selected, and the flight control response is quicker; when the unmanned aerial vehicle hovers or flies on the flat ground, the radar detects that the height value floats up and down less, the flight control does not need the timeliness of the height information of the radar at the moment, the length of a filter window can be properly lengthened, and the height value output by the radar is stable.
The simulated radar filtering delay compensation method based on the neural network comprises the following steps:
s1: acquiring a radar height measurement value;
s2: outputting the height value of the current moment by each frame to obtain sampling data arranged at a certain time interval, and adding N data in a time sequence to obtain a height value after moving average filtering;
s3: the length N of a sliding window at the current moment is adjusted in a self-adaptive manner according to the change characteristic machine learning of the data by the neural network;
s4: when the obstacle and the unmanned aerial vehicle are identified to be in the up-and-down motion state, obtaining a height value by using the sliding average filtering with the window length of N; and when the unmanned aerial vehicle is in hovering and level flying states, obtaining a height value by using a filtering window with the window length being a preset value.
Further, selecting the strongest point in the point cloud detected by the radar, clustering a plurality of points near the strongest point into a target cluster, solving the centroid, and calculating the distance from the target centroid to the radar, namely the current height measurement value H.
Further, the preset value is 3.
Further, the moving average filtering time domain mathematical formula is expressed as follows:
whereinx(n) Represents the current height measurement, andx(n-1) represents the last measurement, and so on,x(n−N+1) is the firstN1 height measurement.
Further, the step of S3 includes the following steps:
s31: the radar measurements are first taken over a time window of lengthnFrame, each sampling obtains a new data to put on the tail of the column, and the original head of the column is lost;
s32: and constructing a neural network to identify the state of the unmanned aerial vehicle so as to adaptively adjust the length of the real-time filtering window.
Further, the neural network is composed of an input layer, a hidden layer and an output layer, and is used for measuring the value sequence with height
As an input to the neural network sample set,
obtaining training output of the neural network as the output of the neural network sample set
And the state of the unmanned aerial vehicle is judged according to the output result through the characteristic learning of the sample.
Further, the step S32 includes the following steps:
s321 input layer to hidden layer:
x i is as followsiIs hiddenThe input of the neurons of the hidden layer,v ih adjusting the weight ratio of each input quantity for the hidden layer connection weight;
s322 activation function via hidden layer:
wherein
Is shown ashA threshold for each hidden layer neuron, which is activated only when information received by the neuron reaches the threshold;
is the firsthInput of hidden layer neurons;
is the firsthThe output of individual hidden layer neurons;
s323 hidden layer to output layer:
s324 activation function via output layer:
wherein
Is shown asjA threshold for individual output layer neurons;
is the firstjInputs to individual output layer neurons;
is the firstjThe output of each of the output layer neurons,
is as followsjActual output in samples;
s325 training error: the error of the prediction result is expressed by a least square method, the weight of each neuron is adjusted according to the error, and the error is reduced step by step;
further, a Sigmod function of the following formula is used in step S324:
the Sigmod function converts a signal with an input from negative infinity to positive infinity into a signal with an input between 0 and 1, and the output is output, wherein 1 represents that an obstacle and the unmanned aerial vehicle are identified to be in an up-and-down motion state; 0 represents the recognition that the drone is in hover and level flight.
Compared with the prior art, the invention has the following beneficial effects:
1) the operation speed is high, and the real-time performance and the stability are good;
2) the unmanned aerial vehicle hanging flight experiment and analysis result prove that the method can effectively eliminate the delay generated by the moving average filter.
Drawings
FIG. 1 is a diagram illustrating the steps of the delay compensation method of the present invention;
FIG. 2 is a graph of delay comparison for different filter window lengths;
FIG. 3 is a comparison graph before and after algorithm optimization.
Detailed Description
The invention is further described with reference to the accompanying drawings, but the invention is not limited in any way, and any alterations or substitutions based on the teaching of the invention are within the scope of the invention.
As shown in fig. 1, the method for compensating delay of filtering of a simulated radar based on a neural network disclosed by the invention comprises the following steps:
s1 obtaining a radar height measurement: selecting the strongest point in the point cloud detected by the radar, clustering a plurality of points near the strongest point into a target cluster, calculating the mass center, and calculating the distance from the target mass center to the radar, wherein the distance is the current height measurement value H.
The centroid coordinates are expressed as:
whereinx i Is shown asiOf dotsxThe coordinates of the position of the object to be imaged,y i is shown asiOf dotsyThe coordinates of the position of the object to be imaged,nrepresenting the number of point clouds.
S2 moving average filtering: outputting the height value of the current moment per frame, obtaining sampling data arranged according to a certain time interval, adding N data in a time sequence for averaging, and expressing a time domain mathematical formula as follows:
whereinx(n) Represents the current height measurement, andx(n-1) represents the last measurement, and so on,x(n−N+1) is the firstN1 height measurement. The frequency domain of the moving average filter is expressed as:
the amplitude-frequency response of the filter is a low-pass filter, the amplitude-frequency response is attenuated along with the rise of the frequency, and the phase-frequency response of the filter is linear. As can be seen from the above, as the window length of the moving average filter becomes longer, its cutoff frequency becomes smaller, and the pass band becomes narrower, so that the response of the filter becomes slower, and the delay becomes larger, as shown in fig. 2.
S3 filtered delay elimination algorithm: the length of the window of the moving average filter is longer, so that data delay is larger, and the algorithm can adaptively adjust the length of the sliding window at the current moment according to the change characteristics of the neural network on the data through machine learning. The specific implementation steps are as follows:
the method comprises the following steps: firstly, time window data of radar measured values are taken, a new data obtained by sampling each time is put at the tail of a column, and the original head of the column is lost; step two: then, the neural network is constructed to identify the state of the drone so as to adaptively adjust the length of the real-time filtering window. The neural network is composed of input layer, hidden layer and output layer, and uses the sequence of height measurement values
As an input to the neural network sample set,
obtaining training output of the neural network as the output of the neural network sample set
. The algorithm judges the state of the unmanned aerial vehicle according to the output result through the characteristic learning of the sample. The method comprises the following specific steps:
a. input layer to hidden layer:
whereinx i Is as followsiThe input of the individual hidden layer neurons,v ih adjusting the weight ratio of each input quantity for the hidden layer connection weight;
b. activation function through hidden layer:
a threshold value representing the neuron, the threshold value being activated only when information received by the neuron meets the threshold value;
wherein
Is shown ashA threshold for each hidden layer neuron, which is activated only when information received by the neuron reaches the threshold;
is the firsthInput of hidden layer neurons;
is the firsthThe output of each hidden layer neuron.
c. Hiding the layer to the output layer, the same way as the step a;
wherein the content of the first and second substances,w hj adjusting the weight ratio of each input quantity for the connection weight value;
d. hiding the layer to the output layer, the same as step b;
wherein
Is shown asjA threshold for individual output layer neurons;
is the firstjInputs to individual output layer neurons;
is the firstjThe output of each of the output layer neurons,
is as followsjActual output in one sample.
e. Training errors: the error of the prediction result is expressed by a least square method, the weight of each neuron is adjusted according to the error, and the error is reduced step by step;
in step d, a Sigmod function is used, which can transform the input from a signal of negative infinity to positive infinity into an output between 0 and 1, as shown in equation 8. 1 represents that the obstacle and the unmanned aerial vehicle are identified to be in an up-and-down motion state; 0 represents the recognition that the drone is in hover and level flight.
Step three: when the obstacle and the unmanned aerial vehicle are identified to be in the up-and-down motion state, obtaining a height value by using the sliding average filtering with the window length of N; and when the unmanned aerial vehicle is identified to be in a hovering state and a flat flying state, obtaining a height value by using a filtering window with the length of the window being 3.
FIG. 3 is a comparison of an experiment using the present invention with an experiment not using the present invention, after which the height curve becomes as smooth as after sliding filtering, while its timeliness is about the same as before sliding filtering, illustrating that this method effectively eliminates the delay produced by the sliding average filter, resulting in both smoothness and timeliness of the height curve.
The invention has the following beneficial effects:
1) the operation speed is high, and the real-time performance and the stability are good;
2) the unmanned aerial vehicle hanging flight experiment and analysis result prove that the method can effectively eliminate the delay generated by the moving average filter.
The above embodiment is an embodiment of the present invention, but the embodiment of the present invention is not limited by the above embodiment, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements within the protection scope of the present invention.
Claims (8)
1. The method for compensating the filtering delay of the simulated radar based on the neural network is characterized by comprising the following steps of:
s1: acquiring a radar height measurement value;
s2: outputting the height value of the current moment by each frame to obtain sampling data arranged at a certain time interval, and adding N data in a time sequence to obtain a height value after moving average filtering;
s3: the length N of a sliding window at the current moment is adjusted in a self-adaptive manner according to the change characteristic machine learning of the data by the neural network;
s4: when the obstacle and the unmanned aerial vehicle are identified to be in the up-and-down motion state, obtaining a height value by using the sliding average filtering with the window length of N; and when the unmanned aerial vehicle is in hovering and level flying states, obtaining a height value by using a filtering window with the window length being a preset value.
2. The method of claim 1, wherein the strongest point in the point cloud detected by the radar is selected, a plurality of points near the strongest point are clustered into a target cluster, the centroid is calculated, and the distance from the target centroid to the radar, i.e., the current height measurement value H, is obtained by calculation.
3. The method of claim 1, wherein the preset value is 3.
4. The method for compensating the delay of the filter of the simulated radar based on the neural network as claimed in claim 1, wherein the mathematical formula of the moving average filter time domain is expressed as follows:
whereinx(n) Represents the current height measurement, andx(n-1) represents the last measurement, and so on,x(n−N+1) is the firstN1 height measurement.
5. The method for compensating delay of filtering of the simulated radar based on the neural network as claimed in claim 1, wherein the step of S3 comprises the steps of:
s31: the radar measurements are first taken over a time window of lengthnFrame, each sampling obtains a new data to put on the tail of the column, and the original head of the column is lost;
s32: and constructing a neural network to identify the state of the unmanned aerial vehicle so as to adaptively adjust the length of the real-time filtering window.
6. The method of claim 5, wherein the neural network comprises an input layer, a hidden layer, and an output layer, and the height measurement values are obtained by a sequence of height measurement values
As an input to the neural network sample set,
obtaining training output of the neural network as the output of the neural network sample set
And the state of the unmanned aerial vehicle is judged according to the output result through the characteristic learning of the sample.
7. The method for compensating delay of filtering of a simulated radar based on a neural network as claimed in claim 6, wherein the step of S32 comprises the steps of:
s321 input layer to hidden layer:
x i is as followsiThe input of the individual hidden layer neurons,v ih the weight ratio of each input quantity is adjusted for the hidden layer connection weight value, dis an input layerα h The number of neurons;
s322 activation function via hidden layer:
wherein gamma ishIs shown ashIs hiddenA threshold for a layer neuron that is activated only when information received by the neuron reaches the threshold; alpha is alphahIs the firsthInput of hidden layer neurons; bhIs the firsthThe output of individual hidden layer neurons;
s323 hidden layer to output layer:
bhis as followshInput to individual output layer neurons, whjThe weight ratio of each input quantity is adjusted for the output layer connection weight value,qis an output layerβ j The number of neurons;
s324 activation function via output layer:
wherein theta isjIs shown asjA threshold for individual output layer neurons; beta is ajIs the firstjInputs to individual output layer neurons;
is the firstjThe output of each of the output layer neurons,
is as followsjActual output in samples;
s325 training error: the error of the prediction result is expressed by a least square method, the weight of each neuron is adjusted according to the error, and the error is reduced step by step;
whereinlAs the number of neurons in the output layer。
8. The method for delay compensation of filter of artificial radar based on neural network as claimed in claim 7, wherein the following Sigmod function is used in step S324:
the Sigmod function transforms a signal with an input from negative infinity to positive infinity into an output between 0 and 1, wherexIs a time series variable of the height measurement,f’(x) Is the final output value after the training by the output layer neurons,f(x) Is composed off’(x) The output value of the nonlinear form after passing through the Sigmod function expression is in an output range of (0, 1), and 1 represents that the obstacle and the unmanned aerial vehicle are identified to be in an up-and-down motion state; 0 represents the recognition that the drone is in hover and level flight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110906174.4A CN113341395B (en) | 2021-08-09 | 2021-08-09 | Simulated radar filtering delay compensation method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110906174.4A CN113341395B (en) | 2021-08-09 | 2021-08-09 | Simulated radar filtering delay compensation method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113341395A CN113341395A (en) | 2021-09-03 |
| CN113341395B true CN113341395B (en) | 2022-01-18 |
Family
ID=77481056
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110906174.4A Active CN113341395B (en) | 2021-08-09 | 2021-08-09 | Simulated radar filtering delay compensation method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113341395B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101211177A (en) * | 2006-12-29 | 2008-07-02 | 中国科学院沈阳计算技术研究所有限公司 | Filter technique based numerical control system acceleration and deceleration control method |
| CN101762808A (en) * | 2010-01-15 | 2010-06-30 | 山东大学 | Method for extracting radar pulse based on self-adaption threshold value |
| CN101834632A (en) * | 2010-04-16 | 2010-09-15 | 西安电子科技大学 | Method for capturing synchronization in frequency hopping communication |
| CN104132884A (en) * | 2013-10-16 | 2014-11-05 | 深圳市帝迈生物技术有限公司 | Rapid processing method and apparatus for signal baseline in signal processing system |
| US10411744B1 (en) * | 2018-10-11 | 2019-09-10 | Ratheon Company | Waveform transformation and reconstruction |
| CN111637878A (en) * | 2020-06-19 | 2020-09-08 | 四川陆垚控制技术有限公司 | Unmanned aerial vehicle navigation filter |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10148476B2 (en) * | 2014-04-05 | 2018-12-04 | Shanghai National Engineering Research Center Of Digital Television Co., Ltd. | Preamble symbol generation and receiving method, and frequency-domain symbol generation method and device |
-
2021
- 2021-08-09 CN CN202110906174.4A patent/CN113341395B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101211177A (en) * | 2006-12-29 | 2008-07-02 | 中国科学院沈阳计算技术研究所有限公司 | Filter technique based numerical control system acceleration and deceleration control method |
| CN101762808A (en) * | 2010-01-15 | 2010-06-30 | 山东大学 | Method for extracting radar pulse based on self-adaption threshold value |
| CN101834632A (en) * | 2010-04-16 | 2010-09-15 | 西安电子科技大学 | Method for capturing synchronization in frequency hopping communication |
| CN104132884A (en) * | 2013-10-16 | 2014-11-05 | 深圳市帝迈生物技术有限公司 | Rapid processing method and apparatus for signal baseline in signal processing system |
| US10411744B1 (en) * | 2018-10-11 | 2019-09-10 | Ratheon Company | Waveform transformation and reconstruction |
| CN111637878A (en) * | 2020-06-19 | 2020-09-08 | 四川陆垚控制技术有限公司 | Unmanned aerial vehicle navigation filter |
Non-Patent Citations (2)
| Title |
|---|
| Adaptive filter bank multi-carrier waveform design for joint communication-radar system;Wanlu Li 等;《Digital Signal Processing》;20201228;第1-10页 * |
| 基于高度融合的植保无人机仿地飞行方法研究;吴开华 等;《农业机械学报》;20180630;第49卷(第6期);第17-23页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113341395A (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112880688B (en) | Unmanned aerial vehicle three-dimensional track planning method based on chaotic self-adaptive sparrow search algorithm | |
| CN110807806B (en) | Obstacle detection method and device, storage medium and terminal equipment | |
| CN109633599B (en) | Multi-target tracking method for airborne early warning radar | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN113128342B (en) | Flight path data preprocessing method and aerial target identification method | |
| CN113970420A (en) | Deep learning-based shock tunnel force measurement signal frequency domain analysis method | |
| CN113808174A (en) | Radar small target tracking method based on full convolution network and Kalman filtering | |
| CN112699536B (en) | Self-adaptive target ship track prediction and locking method | |
| CN109444840B (en) | Radar clutter suppression method based on machine learning | |
| CN112070103B (en) | Method for inverting atmospheric visibility through microwave link network gridding self-adaptive variable scale | |
| CN113341395B (en) | Simulated radar filtering delay compensation method based on neural network | |
| CN115481680A (en) | Exogenous radar-based flying bird and unmanned aerial vehicle track target classification method and equipment | |
| CN112133323A (en) | Unsupervised classification and supervised modification fusion voice separation method related to spatial structural characteristics | |
| CN114677419A (en) | Radar Doppler signal low-slow small target detection method based on three-dimensional convolution network | |
| CN114545387A (en) | High-altitude parabolic detection and discrimination method based on millimeter wave radar data fitting | |
| CN113627075A (en) | Projectile aerodynamic coefficient identification method based on adaptive particle swarm optimization extreme learning | |
| CN113064134B (en) | Laser radar echo processing method based on improved fuzzy neural network filtering | |
| CN112710470A (en) | Self-adaptive wavelet threshold MEMS gyroscope noise reduction method based on maneuver identification | |
| CN109255321B (en) | Visual tracking classifier construction method combining history and instant information | |
| CN115660039A (en) | Intelligent diagnosis method, system, equipment and medium for fault of shipboard aircraft | |
| CN115063714A (en) | Bird collision accident prevention target detection method based on improved YOLOv5s network | |
| CN113721214A (en) | Helicopter hovering detection method based on air traffic control radar | |
| Peng et al. | Research on plot authenticity identification method based on PSO-MLP | |
| CN104698836A (en) | Method for preprocessing flight data | |
| Jan et al. | Artificial neural networks in the filtration of radiolocation information |
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 |