CN112986950A - Single-pulse laser radar echo feature extraction method based on deep learning - Google Patents
Single-pulse laser radar echo feature extraction method based on deep learning Download PDFInfo
- Publication number
- CN112986950A CN112986950A CN202011568745.XA CN202011568745A CN112986950A CN 112986950 A CN112986950 A CN 112986950A CN 202011568745 A CN202011568745 A CN 202011568745A CN 112986950 A CN112986950 A CN 112986950A
- Authority
- CN
- China
- Prior art keywords
- sequence
- echo
- lstm
- neural network
- laser radar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 23
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 49
- 238000000034 method Methods 0.000 claims abstract description 24
- 229920006395 saturated elastomer Polymers 0.000 claims abstract description 24
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 19
- 238000011084 recovery Methods 0.000 claims abstract description 13
- 238000004364 calculation method Methods 0.000 claims abstract description 7
- 238000013528 artificial neural network Methods 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 14
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000001208 nuclear magnetic resonance pulse sequence Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000000306 recurrent effect Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 210000004027 cell Anatomy 0.000 claims description 2
- 210000002569 neuron Anatomy 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims 1
- 238000007781 pre-processing Methods 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 7
- 238000005259 measurement Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
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
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention discloses a waveform feature extraction method of a single pulse laser radar echo based on deep learning, which comprises the steps of collecting various laser radar echo signals with complete signal features, preprocessing and splicing effective parts of the collected echo signals, forming training data by an obtained splicing sequence with echo feature information, performing waveform recovery on distorted echo signals based on an LSTM-RNN model, particularly performing calculation and prediction on saturated parts of saturated distortion, thereby obtaining partial echo information missing due to waveform distortion, and then performing feature extraction through a convolutional neural network to obtain the wave feature of the radar echo. The method is applied to analysis of the echo signal of the single-pulse laser radar, does not depend on hardware equipment completely, and therefore characteristic information of the echo signal can be extracted from a distorted waveform signal.
Description
Technical Field
The invention relates to a waveform feature extraction technology, in particular to a monopulse laser radar echo feature extraction method based on deep learning.
Background
The laser radar is a product combining a radar principle and a laser technology, takes laser as a detection beam, has a series of advantages of long measuring range, high sensitivity, high spatial resolution, strong anti-interference capability and the like, and has wide application prospect in the fields of distance measurement, target tracking, three-dimensional imaging, environment perception and the like.
The signal processing is the most important part in a laser radar detection system, and aims to accurately analyze, diagnose, compress and quantize an echo signal, and quickly realize the transmission, storage and accurate reconstruction of the signal. Due to the influence of a complex and variable detection environment, the laser radar echo signal not only carries information components of a target, but also is more a noise signal, and even shows the situation that the target signal is submerged, so the detection precision and the detection distance of the laser radar are seriously influenced by the existence of the noise, how to remove the noise is realized, and the effective extraction of target object information becomes the primary task and difficulty of a detection system.
The single measurement precision of the low-cost pulse laser range finder can generally reach the meter level, and the precision is not enough in the application occasions of the vehicle-mounted laser sensor. The main reason for the poor accuracy is due to the limitation of the dynamic range of the receiver subsystem, and in order to ensure high accuracy of distance measurement, the error of time discrimination needs to be reduced, so that in order to avoid saturation distortion of the detected echo pulse signal, and at the same time, for less missing detection, the echo signal strength in the measurement range should also be greater than the threshold voltage, which means that the dynamic range of the receiving circuit needs to be larger.
At present, most of research achievements of deep learning in the field of radar target identification are focused on processing extracted image data by using a deep learning technology, and the essence of the research achievements is still in the category of processing the image data and is not really applied to the learning identification of radar target sequence signals.
The inherent nature of the detectors and hardware means that high power radar returns can often produce saturated signals. Typically, these saturated signals are discarded during data processing, and therefore some useful information is lost. Therefore, it is worthwhile to restore the saturated signal to the normal state. The mapping between the saturated signal waveform and the normal signal waveform constitutes a regression problem. Since scintillators and sets do not usually constitute a linear system, typical regression methods such as multi-parameter fitting cannot be applied immediately. An important advantage of neural networks is the ability to handle non-linear regression problems.
For the saturation problem of the laser radar echo, two active solutions are available at present, one is based on hardware and avoids the saturation of the echo signal on the basis of a hardware environment, and the other adopts fitting iteration to recover the saturated signal. The former method currently adopts a multi-layer hierarchical echo processing method, that is, when measuring a short-distance target, a hardware device conforming to a short-distance dynamic range is adopted, but when measuring a long-distance target, a hardware device conforming to a long-distance dynamic range is adopted. This approach can effectively extend the dynamic range of the lidar, but at twice the cost of previous hardware devices with only a single dynamic range. Another method using iterative fitting takes a lot of time to perform iterative calculations. Both of these approaches have certain limitations. The problem of saturation is solved, and it has very big help to characteristic extraction afterwards, and traditional laser radar ranging system can give up the saturation signal generally speaking, but this can have certain influence to the radar performance, consequently need avoid the appearance of saturation phenomenon as far as possible, or reduce the influence that the saturation signal was drawed to laser radar echo characteristic.
Disclosure of Invention
The invention aims to provide a method for extracting echo characteristics of a single-pulse laser radar based on deep learning, which can recover saturated radar signals caused by insufficient dynamic range under the condition of limited hardware conditions, so that lost characteristic information can be recovered, and finally, the recovered waveform is subjected to characteristic extraction, thereby providing guarantee for subsequent work.
The technical scheme for realizing the purpose of the invention is as follows: a method for extracting echo characteristics of a single-pulse laser radar based on deep learning comprises the following steps:
step S1: sampling by using a pulse laser radar, and taking a series of laser radar echo signals containing complete waveform information as a training data set of a neural network;
step S2: respectively sending the sorted training data set into a pre-built cyclic neural network and a pre-built convolutional neural network for data training, and continuously updating parameter information through repeated training iteration to finally establish a proper LSTM network and an LSTM-CNN network;
step S3: sending a pulse laser radar signal which generates distortion, particularly a saturated pulse laser radar signal into a well-built LSTM neural network which is trained and iterated to carry out waveform recovery, so that waveform information lost due to waveform distortion is recovered;
step S4: and the pulse laser radar echo recovered in the step S3 is sent to the LSTM-CNN neural network trained in the step S2 for feature extraction.
Compared with the prior art, the invention has the following remarkable advantages:
(1) aiming at the problem of insufficient dynamic range of the laser radar, an LSTM model-based laser radar saturated echo recovery algorithm is provided, and the laser radar saturated echo signal caused by the insufficient dynamic range can be quickly recovered in real time without depending on hardware conditions; the hardware is convenient, and extra hardware facilities are not needed, so that the cost is saved; in the aspect of algorithm, a large amount of time is saved, and the saturated signal can be quickly and accurately recovered;
(2) because the laser radar is mainly used for ranging, the feature extraction of the invention mainly extracts the distance features of laser radar signals; the method can quickly and accurately identify the distance features with the precision of 0.15m, and has certain real-time property.
Drawings
In order to make the object, technical scheme and beneficial effect of the invention more clear, the invention provides the following drawings for explanation:
FIG. 1 is a complete flow diagram of the process.
FIG. 2 is a flow chart of waveform recovery based on the LSTM-RNN model.
FIG. 3 is a block diagram of the LSTM-RNN model used in the method.
Fig. 4 is a graph of experimental results of a portion of a saturated lidar echo signal recovery based on an LSTM model.
Detailed Description
The following description will proceed with reference being made to specific embodiments of the invention, but it should be understood that the scope of the invention is not limited to the specific embodiments.
Referring to fig. 1, 2 and 3, a method for extracting echo characteristics of a single-pulse laser radar based on deep learning includes the following steps:
step S1: and (3) carrying out sample collection work by using a pulse laser radar, and taking a series of laser radar echo signals containing complete waveform information as a training data set of the neural network.
Step S2: and respectively sending the sorted training data sets into a pre-built cyclic neural network and a pre-built convolutional neural network for data training, and continuously updating parameter information through repeated training iteration to finally establish a proper LSTM network and an LSTM-CNN network.
Step S3: sending the distorted pulse laser radar signal, especially the saturated pulse laser radar signal into the well-built LSTM neural network which is trained and iterated to recover the waveform, so that the waveform information lost due to waveform distortion can be recovered.
Step S31: a complete series of unsaturated echo pulse sequences generated from a single pulse lidar is acquired as a training dataset for a neural network.
Step S311: the radar echo containing complete information acquired by the laboratory laser radar is composed of n sampling points, namely a single sample (a complete waveform) is composed of n time sequences, and a series of acquired unsaturated echo pulse sequences containing complete information have m in total, namely the total amount of the samples is m. Splicing the collected m samples in the time sequence direction to obtain m × n time sequences in total as input of training samples, as follows:
Dataset＝(x11,x12,…,x1n,x21,…,x2n,…,xm1,…,xmn)
step S312: in order to make the subsequent gradient calculation speed faster and make the direction of the updated parameters more fit with the whole training, the spliced sequence sample data is scaled to the range of 0-1 through a MinMaxScale function, namely, the training data set is normalized, and the formula is as follows:
where Max is the maximum value in the sequence samples and Min is the minimum value in the sequence samples.
Step S313: and carrying out input and output on the normalized training sample, and dividing a training set and a test set. Designing a sliding window with the width of s (s is larger than 10), and each time moving in the time direction, wherein the sliding window comprises s sequence samples, the first s/2 data are set as input, all the remaining data are set as output, namely, in the whole sliding window with the width of s, the remaining sequences are predicted through the first s/2 sequence samples. The sliding window is moved forward one sequence length at a time.
Step S32: inputting the sorted training data set into an LSTM-RNN neural network which is set up in advance and initialized by parameters for training, and finally obtaining the model of the radar saturated echo signal waveform recovery method based on the LSTM-RNN model by performing iterative training on a large number of echo sequences containing complete information and continuously updating the initialized parameters.
Step S321: and constructing a double-layer LSTM recurrent neural network model with N (N is more than 1) dimensions of intermediate hidden layer vectors. The LSTM-RNN neural network model of the first layer finally returns the output of all time steps, and the LSTM-RNN neural network of the second layer finally returns the output of the last time step. The loss function adopts a Mean Square Error (MSE), and the stochastic gradient descent algorithm adopts an Adam optimization algorithm to iteratively update each weight parameter of the recurrent neural network on the training sample sequence, so as to finally obtain the optimal parameter.
Step S322: setting initial values of various parameters of the LSTM neural network, values of Dropout functions, the number of neurons in each layer, a decapay value of an Adam algorithm and iteration times.
Step S323: inputting the processed sample sequence of step S31 into the neural network with various parameters setAnd (5) performing training, and finally obtaining a trained model, namely the model of the single-pulse laser radar saturated echo signal recovery method. The sample sequence in the sliding window with the width S set in step S313 is input into the training model, for example, the sample in the sliding window at time t is P ═ x (x)t，xt+1，…，xt+(s-2)，xt+(s-1)) Taking the sequence value of the first s/2 moments as an input vector of an LSTM input layer, and taking the subsequent sequence value as a network predicted value to compare with a real network, wherein the calculation formula of each LSTM is as follows:
wherein x istRepresenting the value of the sample in the t-th time series, ht-1Representing the implicit vector, W, at t-1 time sequencesfDoor f for indicating forgettingtWeight matrix of bfDoor f for indicating forgettingtOffset vector of WiRepresentation input gate itWeight matrix of biRepresenting the offset vector of the input gate, WCWeight matrix representing the state values of the LSTM cells, bCOffset vector, W, representing the state value of the LSTM cellOWeight matrix representing output gates, bORepresenting the programming vector of the output gate.
Step S33: and predicting the part of the information sequence which is missing from the saturated and distorted signal sequence by using the constructed model so as to obtain a complete predicted waveform sequence.
Step S34: and correcting the predicted waveform sequence to obtain a finally recovered complete waveform sequence. Because partial leading edge and partial trailing edge of a saturated echo signal pulse of the monopulse laser radar exist, the method predicts a partial signal sequence after loss due to saturation by a real leading edge signal sequence in which a saturated signal exists, in order to enable a prediction result to follow the accuracy, the method performs mirror image inversion on the whole sample sequence of a training sample to obtain a new sample sequence after mirror image, then inputs the new sample sequence into an LSTM-RNN neural network for training, predicts a partial signal sequence before loss due to saturation by the trailing edge signal sequence of the saturated signal, and finally performs weighted calculation on a result predicted by a forward sequence and a result predicted by a reverse sequence to obtain a final corrected predicted waveform sequence.
Step S4: and the pulse laser radar echo recovered in the step S3 is sent to the LSTM-CNN neural network trained in the step S2 for feature extraction.
Step S41: acquiring a series of complete unsaturated echo pulse sequences generated from a single-pulse laser radar as a training data set of a convolutional neural network, and carrying out classification calibration on the data set with the resolution of 0.15 m.
Step S42: inputting the sorted training data set into an LSTM-CNN neural network which is set up in advance and initialized by parameters for training, and continuously updating the initialized parameters by performing iterative training on a large number of echo sequences to finally obtain the model of the radar echo feature extraction method based on the LSTM-CNN model.
Step S43: and comparing the characteristics of any radar echo signal by using the constructed model, and putting the radar signals with the same characteristics into corresponding classification labels, thereby obtaining the distance characteristic information of the radar signals.
Fig. 4 is an experimental result of the algorithm on the recovery part of the saturation laser radar echo signal based on the LSTM model, so that it can be seen that the algorithm can recover the pulse laser radar saturation echo signal, the recovery effect is relatively ideal, and the evaluation result of the final evaluation using the mean square error is as follows: the mean square error between the LSTM recovered waveform and the original waveform is 0.045, and the range accuracy of the recovered echo is 0.15 m.
The final experiment result shows that the pulse laser radar echo feature extraction method based on deep learning can extract the distance features, the identification accuracy of the obtained laser radar echo feature extraction model based on the LSTM-CNN model reaches 95.72% in 1ns, the identification accuracy of the obtained laser radar echo feature extraction model based on the LSTM-CNN model reaches 2ns, and the requirement that the laser radar ranging accuracy is 1ns is basically met.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (9)
1. A method for extracting echo features of a single-pulse laser radar based on deep learning is characterized by comprising the following steps:
step S1: sampling by using a pulse laser radar, and taking a series of laser radar echo signals containing complete waveform information as a training data set of a neural network;
step S2: respectively sending the sorted training data set into a pre-built cyclic neural network and a pre-built convolutional neural network for data training, and continuously updating parameter information through repeated training iteration to finally establish a proper LSTM network and an LSTM-CNN network;
step S3: sending a pulse laser radar signal which generates distortion, particularly a saturated pulse laser radar signal into a well-built LSTM neural network which is trained and iterated to carry out waveform recovery, so that waveform information lost due to waveform distortion is recovered;
step S4: and the pulse laser radar echo recovered in the step S3 is sent to the LSTM-CNN neural network trained in the step S2 for feature extraction.
2. The method for extracting echo characteristics of monopulse lidar based on deep learning according to claim 1, wherein the method for recovering the waveform of the saturation echo of monopulse lidar based on LSTM-RNN neural network in step S3 comprises the following steps:
step S31: acquiring a series of complete unsaturated echo pulse sequences generated from a single-pulse laser radar as a training data set of a neural network;
step S32: inputting the sorted training data set into an LSTM-RNN neural network which is set up in advance and initialized by parameters for training, and finally obtaining a model of a radar saturated echo signal waveform recovery method based on an LSTM-RNN model by performing iterative training on an echo sequence containing complete information and continuously updating the initialized parameters;
step S33: predicting the information sequence missing from the saturated signal sequence and the distorted signal sequence by using the constructed model so as to obtain a complete predicted waveform sequence;
step S34: and correcting the predicted waveform sequence to obtain a finally recovered complete waveform sequence.
3. The deep learning-based single pulse lidar echo feature extraction method according to claim 2, wherein the step S31 comprises the steps of:
step S311: the radar echo containing complete information and acquired by a pulse laser radar is composed of n sampling points, namely a single sample is composed of n time sequences, and a series of acquired unsaturated echo pulse sequences containing complete information are m in number, namely the total number of the samples is m; splicing the collected m samples in the time sequence direction to obtain m × n time sequences in total as input of training samples, as follows:
Dataset＝(x11,x12,…,x1n,x21,…，x2n，…,xm1,…,xmn)
step S312: the spliced sequence sample data is scaled to be within the range of 0-1 through a MinMaxScaler function, namely, the training data set is subjected to normalization processing, and the formula is as follows:
max is the maximum value in the sequence samples, and Min is the minimum value in the sequence samples;
step S313: and carrying out input and output on the normalized training sample, and dividing a training set and a test set.
4. The deep learning-based single pulse laser radar echo feature extraction method according to claim 3, characterized in that: the sample division method in step S313, designing a sliding window with width S, where S > 10, moving in the time direction each time, where the sliding window includes S sequence samples, where the first S/2 data are set as input, and all the remaining data are set as output, that is, in the sliding window with width S, the remaining sequences are predicted by the first S/2 sequence samples; the sliding window is moved forward one sequence length at a time.
5. The method for extracting echo characteristics of single-pulse lidar based on deep learning according to claim 2, wherein the step S32 of establishing the waveform recovery model comprises the following steps:
step S321: constructing a double-layer LSTM recurrent neural network model with vector of middle hidden layer being N-dimension, wherein N is more than 1;
step S322: setting initial values of various parameters of the LSTM neural network, values of Dropout functions, the number of neurons in each layer, and escape values and iteration times of Adam algorithm
Step S323: and (4) inputting the processed sample sequence of the step (S31) into the neural network with various set parameters for training, and finally obtaining a trained model, namely the model of the single-pulse laser radar saturated echo signal recovery method.
6. The deep learning-based single pulse laser radar echo feature extraction method according to claim 5, characterized in that: in the double-layer LSTM-RNN neural network model in step S321, the first layer LSTM-RNN neural network model finally returns the output of all time steps, and the second layer LSTM-RNN neural network finally returns the output of the last time step; the loss function adopts a mean square error function, and the random gradient descent algorithm adopts an Adam optimization algorithm to iteratively update each weight parameter of the recurrent neural network on the training sample sequence, so as to finally obtain the optimal parameter.
7. The deep learning-based single pulse laser radar echo feature extraction method according to claim 5, characterized in that: in the training model method described in step S323, the set sample sequence in the sliding window with the width S is input into the training model, and the sample in the sliding window at the time t is P ═ x (x)t,xt+1,…,xt+(s-2),xt+(s-1)) Taking the sequence value of the first s/2 moments as an input vector of an LSTM input layer, and taking the subsequent sequence value as a network predicted value to compare with a real network, wherein the calculation formula of each LSTM is as follows:
ft＝σ(Wf·[ht-1，xt]+bf)
it＝σ(Wi·[ht-1，xt]+bi)
Ot＝σ(WO·[ht-1，xt]+bO)
ht＝Ot·tanh(Ct)
wherein x istRepresenting the value of the sample in the t-th time series, ht-1Representing the implicit vector, W, at t-1 time sequencesfDoor f for indicating forgettingtWeight matrix of bfDoor f for indicating forgettingtOffset vector of WiRepresentation input gate itWeight matrix of biRepresenting the offset vector of the input gate, WCWeight matrix representing the state values of the LSTM cells, bCOffset vector, W, representing the state value of the LSTM cellOWeight matrix representing output gates, bORepresenting the programming vector of the output gate.
8. The deep learning-based single pulse laser radar echo feature extraction method according to claim 2, characterized in that: modifying the predicted waveform sequence in step S34, mirror-inverting the entire sample sequence of the training sample to obtain a new sample sequence after mirroring, inputting the new sample sequence into the LSTM-RNN neural network for training, predicting the previous partial signal sequence missing due to saturation through the trailing edge signal sequence of the saturation signal, and finally performing weighted calculation on the predicted result of the forward sequence and the predicted result of the reverse sequence to obtain the final modified predicted waveform sequence.
9. The method for extracting echo characteristics of monopulse lidar based on deep learning of claim 1, wherein the method for extracting echo characteristics of monopulse lidar based on LSTM-CNN neural network in step S4 comprises the following steps:
step S41: acquiring a series of complete unsaturated echo pulse sequences generated from a single-pulse laser radar as a training data set of a convolutional neural network, and classifying and calibrating the data set with the resolution of 0.15 m;
step S42: inputting the sorted training data set into an LSTM-CNN neural network which is set up in advance and initialized by parameters for training, and continuously updating the initialized parameters by performing iterative training on an echo sequence to finally obtain a model of the radar echo feature extraction method based on the LSTM-CNN model;
step S43: and comparing the characteristics of any radar echo signal by using the constructed model, and putting the radar signals with the same characteristics into corresponding classification labels, thereby obtaining the distance characteristic information of the radar signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011568745.XA CN112986950B (en) | 2020-12-25 | 2020-12-25 | Single-pulse laser radar echo feature extraction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011568745.XA CN112986950B (en) | 2020-12-25 | 2020-12-25 | Single-pulse laser radar echo feature extraction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112986950A true CN112986950A (en) | 2021-06-18 |
| CN112986950B CN112986950B (en) | 2024-09-06 |
Family
ID=76345144
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011568745.XA Active CN112986950B (en) | 2020-12-25 | 2020-12-25 | Single-pulse laser radar echo feature extraction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112986950B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113655457A (en) * | 2021-08-24 | 2021-11-16 | 中国电子科技集团公司第十四研究所 | Radar target detection capability self-evolution method and device based on sample mining |
| CN113720862A (en) * | 2021-08-17 | 2021-11-30 | 珠海格力电器股份有限公司 | Part abnormality detection method, device, equipment and storage medium |
| CN113759323A (en) * | 2021-07-19 | 2021-12-07 | 西安电子科技大学 | Signal sorting method and device based on improved K-Means combined convolution self-encoder |
| TWI771098B (en) * | 2021-07-08 | 2022-07-11 | 國立陽明交通大學 | Fault diagnosis system and method for state of radar system of roadside units |
| CN115356694A (en) * | 2022-08-26 | 2022-11-18 | 哈尔滨工业大学（威海） | Anti-impact interference method and system for high-frequency ground wave radar, computer equipment and application |
| CN116432703A (en) * | 2023-06-12 | 2023-07-14 | 成都大学 | Pulse height estimation method, system and terminal based on composite neural network model |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107871119A (en) * | 2017-11-01 | 2018-04-03 | 西安电子科技大学 | A kind of object detection method learnt based on object space knowledge and two-stage forecasting |
| CN109146849A (en) * | 2018-07-26 | 2019-01-04 | 昆明理工大学 | A kind of road surface crack detection method based on convolutional neural networks and image recognition |
| CN109270525A (en) * | 2018-12-07 | 2019-01-25 | 电子科技大学 | Through-wall radar imaging method and system based on deep learning |
| WO2019103223A1 (en) * | 2017-11-21 | 2019-05-31 | 연세대학교 산학협력단 | Data input/output saturation detection apparatus and method |
| CN110443127A (en) * | 2019-06-28 | 2019-11-12 | 天津大学 | In conjunction with the musical score image recognition methods of residual error convolutional coding structure and Recognition with Recurrent Neural Network |
| CN110456332A (en) * | 2019-08-12 | 2019-11-15 | 哈尔滨工程大学 | A kind of underwater sound signal Enhancement Method based on autocoder |
| CN111272277A (en) * | 2020-01-21 | 2020-06-12 | 中国工程物理研究院激光聚变研究中心 | Laser pulse waveform measurement distortion correction method and system based on neural network |
| CN111582610A (en) * | 2020-07-13 | 2020-08-25 | 清华四川能源互联网研究院 | Prediction method for family energy decomposition based on convolutional neural network |
| CN111638488A (en) * | 2020-04-10 | 2020-09-08 | 西安电子科技大学 | Radar interference signal identification method based on LSTM network |
| CN111722195A (en) * | 2020-06-29 | 2020-09-29 | 上海蛮酷科技有限公司 | Radar occlusion detection method and computer storage medium |
| CN112055284A (en) * | 2019-06-05 | 2020-12-08 | 北京地平线机器人技术研发有限公司 | Echo cancellation method, neural network training method, apparatus, medium, and device |
-
2020
- 2020-12-25 CN CN202011568745.XA patent/CN112986950B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107871119A (en) * | 2017-11-01 | 2018-04-03 | 西安电子科技大学 | A kind of object detection method learnt based on object space knowledge and two-stage forecasting |
| WO2019103223A1 (en) * | 2017-11-21 | 2019-05-31 | 연세대학교 산학협력단 | Data input/output saturation detection apparatus and method |
| CN109146849A (en) * | 2018-07-26 | 2019-01-04 | 昆明理工大学 | A kind of road surface crack detection method based on convolutional neural networks and image recognition |
| CN109270525A (en) * | 2018-12-07 | 2019-01-25 | 电子科技大学 | Through-wall radar imaging method and system based on deep learning |
| CN112055284A (en) * | 2019-06-05 | 2020-12-08 | 北京地平线机器人技术研发有限公司 | Echo cancellation method, neural network training method, apparatus, medium, and device |
| CN110443127A (en) * | 2019-06-28 | 2019-11-12 | 天津大学 | In conjunction with the musical score image recognition methods of residual error convolutional coding structure and Recognition with Recurrent Neural Network |
| CN110456332A (en) * | 2019-08-12 | 2019-11-15 | 哈尔滨工程大学 | A kind of underwater sound signal Enhancement Method based on autocoder |
| CN111272277A (en) * | 2020-01-21 | 2020-06-12 | 中国工程物理研究院激光聚变研究中心 | Laser pulse waveform measurement distortion correction method and system based on neural network |
| CN111638488A (en) * | 2020-04-10 | 2020-09-08 | 西安电子科技大学 | Radar interference signal identification method based on LSTM network |
| CN111722195A (en) * | 2020-06-29 | 2020-09-29 | 上海蛮酷科技有限公司 | Radar occlusion detection method and computer storage medium |
| CN111582610A (en) * | 2020-07-13 | 2020-08-25 | 清华四川能源互联网研究院 | Prediction method for family energy decomposition based on convolutional neural network |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI771098B (en) * | 2021-07-08 | 2022-07-11 | 國立陽明交通大學 | Fault diagnosis system and method for state of radar system of roadside units |
| CN113759323A (en) * | 2021-07-19 | 2021-12-07 | 西安电子科技大学 | Signal sorting method and device based on improved K-Means combined convolution self-encoder |
| CN113759323B (en) * | 2021-07-19 | 2024-02-06 | 西安电子科技大学 | Signal sorting method and device based on improved K-Means joint convolution self-encoder |
| CN113720862A (en) * | 2021-08-17 | 2021-11-30 | 珠海格力电器股份有限公司 | Part abnormality detection method, device, equipment and storage medium |
| CN113720862B (en) * | 2021-08-17 | 2023-01-13 | 珠海格力电器股份有限公司 | Part abnormality detection method, device, equipment and storage medium |
| CN113655457A (en) * | 2021-08-24 | 2021-11-16 | 中国电子科技集团公司第十四研究所 | Radar target detection capability self-evolution method and device based on sample mining |
| CN113655457B (en) * | 2021-08-24 | 2023-11-24 | 中国电子科技集团公司第十四研究所 | Self-evolution method and device for radar target detection capability based on sample mining |
| CN115356694A (en) * | 2022-08-26 | 2022-11-18 | 哈尔滨工业大学（威海） | Anti-impact interference method and system for high-frequency ground wave radar, computer equipment and application |
| CN115356694B (en) * | 2022-08-26 | 2023-08-22 | 哈尔滨工业大学（威海） | High-frequency ground wave radar anti-impact interference method, system, computer equipment and application |
| CN116432703A (en) * | 2023-06-12 | 2023-07-14 | 成都大学 | Pulse height estimation method, system and terminal based on composite neural network model |
| CN116432703B (en) * | 2023-06-12 | 2023-08-29 | 成都大学 | Pulse height estimation method, system and terminal based on composite neural network model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112986950B (en) | 2024-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112986950B (en) | Single-pulse laser radar echo feature extraction method based on deep learning | |
| CN110210463B (en) | Precise ROI-fast R-CNN-based radar target image detection method | |
| CN109407067B (en) | Radar moving target detection and classification integrated method based on time-frequency graph convolution neural network | |
| CN108447057B (en) | SAR image change detection method based on significance and depth convolution network | |
| CN112132042A (en) | SAR image target detection method based on anti-domain adaptation | |
| CN108171119B (en) | SAR image change detection method based on residual error network | |
| CN112036239B (en) | Radar signal working mode identification method and system based on deep learning network | |
| CN110728698A (en) | Multi-target tracking model based on composite cyclic neural network system | |
| CN114781458B (en) | Power distribution network initial fault identification method based on adaptive wavelet and attention mechanism time-frequency memory neural network | |
| Wang et al. | Model-and-data-driven method for radar highly maneuvering target detection | |
| CN116778290A (en) | Radar vision data association method based on deep learning algorithm | |
| Chen et al. | Function recognition of multi-function radar via cnn-gru neural network | |
| CN108507607B (en) | Weak signal detection method based on kernel function | |
| CN117710689A (en) | High-precision SAR image target detection method and system based on particle filtering | |
| Ruan et al. | Automatic recognition of radar signal types based on CNN-LSTM | |
| CN112327286A (en) | Low-complexity daily activity classification method, device, equipment and storage medium | |
| CN111784738A (en) | Extremely dark and weak moving target correlation detection method based on fluctuation analysis | |
| CN114748047A (en) | Blood pressure interval measuring method based on residual error space-time feature fusion and Gaussian process regression | |
| CN113421281A (en) | Pedestrian micromotion part separation method based on segmentation theory | |
| CN112507826A (en) | End-to-end ecological change monitoring method, terminal, computer equipment and medium | |
| Luo et al. | Automatic reconstruction of radar pulse repetition pattern based on model learning | |
| CN113723253B (en) | Clutter pre-classification-based moving target detection method | |
| CN113537299B (en) | Target tracking method of distributed Bayes filter based on expectation maximization Gaussian mixture reduction | |
| CN115015903B (en) | Radar sequential image moving target detection method and system | |
| CN113341391B (en) | Radar target multi-frame joint detection method in unknown environment based on deep learning |
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 |