CN112329534B - Radar target identification method based on two-dimensional weighted residual convolution neural network - Google Patents
Radar target identification method based on two-dimensional weighted residual convolution neural network Download PDFInfo
- Publication number
- CN112329534B CN112329534B CN202011067886.3A CN202011067886A CN112329534B CN 112329534 B CN112329534 B CN 112329534B CN 202011067886 A CN202011067886 A CN 202011067886A CN 112329534 B CN112329534 B CN 112329534B
- Authority
- CN
- China
- Prior art keywords
- training
- dimensional
- neural network
- weighted residual
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
- G06F17/141—Discrete Fourier transforms
- G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Discrete Mathematics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a radar target identification method based on a two-dimensional weighted residual convolution neural network, which comprises the following steps: constructing a weighted residual convolution neural network; constructing a one-dimensional radar frequency domain target signal into a two-dimensional data plane based on a Toepliz matrix; generating a training set according to the signal category; training the training set data by using the weighted residual convolution neural network, obtaining a trained model, and completing radar target identification by using the model. According to the invention, the original one-dimensional ground reconnaissance radar frequency domain target signal is constructed into a two-dimensional data plane for training, and compared with the training by directly using a one-dimensional signal, the training performance is greatly enhanced, and the testing accuracy is higher; the invention provides a weighted residual error module, wherein the residual error structure can greatly reduce the operation complexity and effectively alleviate the problems of gradient disappearance and gradient explosion in the back propagation process of a neural network.
Description
Technical Field
The invention belongs to radar target recognition technology, and particularly relates to a radar target recognition method based on a two-dimensional weighted residual convolution neural network.
Background
Radar target identification refers to a technology for extracting robust radar features of a target from a target reflected echo signal received by the radar, and automatically identifying the type or model of the target by using the features. In the 21 st century, due to the complexity of modern military war environments and the diversification of hostile targets and tasks, how to find, detect and effectively identify targets in time in harsh environments is a key place for war wins. Radar, which is an important long-distance detection sensor, plays an important role in military, and radar target recognition technology has become a powerful lever for modern military development, and is a core application of modern electronic warfare.
Currently, many methods for automatically extracting target recognition of radar deep features, such as Support Vector Machines (SVM) and Extreme Learning Machines (ELM), have been developed. However, the method only uses the time domain characteristics of the target and the accuracy of target identification is low.
The weighted residual convolution neural network has good robustness because of good mobility and can extract deep features of the target, so that the intrinsic information of the target can be better represented, and the weighted residual convolution neural network is also used for radar target identification in recent years.
The invention patent with the patent application number 201710838721.3 discloses a radar high-resolution range profile target recognition method based on a one-dimensional convolutional neural network, the invention patent with the patent application number 201811405815.2 discloses a radar target recognition method based on a depth residual multi-scale one-dimensional weighted residual convolutional neural network, and the two patents are all realized by training one-dimensional radar frequency domain target signals by using the one-dimensional convolutional neural network. The one-dimensional convolution kernel is limited by the dimension, so that the feature extraction can only be performed by transverse sliding, and long-distance related features cannot be well captured.
Disclosure of Invention
The invention aims to provide a target identification method for ground reconnaissance radar based on a two-dimensional weighted residual convolution neural network, which can extract more relevant features, and has the advantages of higher calculation efficiency, more complete convolution and deeper extracted features.
The technical solution for realizing the purpose of the invention is as follows: a ground reconnaissance radar target identification method based on a two-dimensional weighted residual convolution neural network comprises the following steps:
step 1, constructing a weighted residual convolutional neural network, wherein the weighted residual convolutional neural network comprises a ten-layer weighted residual module, five-layer convolutional downsampling layers, a maximum pooling layer and two full-connection layers;
step 2, constructing a one-dimensional radar signal into a two-dimensional data plane based on a Toepliz matrix;
step 3, generating a training set and a testing set according to a certain proportion from the data processed in the step 2;
step 4, putting the generated training set into the network constructed in the step 1 for training to obtain a trained model;
and 5, testing the trained model by using the test set.
Compared with the prior art, the invention has the remarkable advantages that: (1) Compared with other target recognition methods, the weighted residual convolution neural network has the advantages of deeper extraction characteristics and stronger recognition capability, and has good adaptability with the ground reconnaissance radar; (2) The method has the advantages that the structure of the original one-dimensional ground reconnaissance radar frequency domain target signal into a two-dimensional data plane is realized based on the Toepliz matrix, and the recognition performance of the ground reconnaissance radar frequency domain target signal is greatly improved; (3) The weighted residual error module firstly uses a convolution kernel of 1 multiplied by 1 to reduce the channel dimension of an input signal so as to reduce the computational complexity, and then respectively performs feature extraction through three convolution kernels with different scales, so that the module can extract feature structures under different sensing fields, and performs feature fusion in a weighted summation mode so as to extract wider features and reduce the computational complexity; meanwhile, the weighted residual error module can reduce the calculated amount of the network and alleviate the problems of gradient disappearance and gradient explosion in the back propagation process; (4) Compared with a one-dimensional weighted residual convolution neural network, the two-dimensional weighted residual convolution neural network has the advantages of stronger convolution performance, deeper extracted features and stronger learning ability.
The invention is described in further detail below with reference to the accompanying drawings.
Drawings
FIG. 1 is a flow chart of a target recognition method for a ground reconnaissance radar frequency domain target signal based on a weighted residual convolution neural network.
Fig. 2 is a weighted residual block diagram.
Fig. 3 is a network configuration diagram constructed in accordance with the present invention.
Fig. 4 is a diagram illustrating a target frequency domain of a ground reconnaissance radar.
Fig. 5 is a signal diagram of the radar signal after the conversion in steps 2 to 4.
FIG. 6 is a graph showing the test results of examples 1, 2 and 3 of the present invention.
Detailed Description
Referring to fig. 1, the ground reconnaissance radar target identification method based on the two-dimensional weighted residual convolution neural network provided by the invention comprises the following steps:
step 1, constructing a weighted residual convolutional neural network, and combining fig. 2 and fig. 3, the weighted residual convolutional neural network structure provided by the invention totally uses ten layers of weighted residual modules, five layers of convolutional downsampling layers, one maximum pooling layer and two full-connection layers. Each convolution downsampling layer is connected with two layers of weighted residual modules; each of blocks 1 through 5 in fig. 3 uses a two-layer weighted residual block and a one-layer convolved downsampling layer. The module 6 uses two fully connected layers and one Softmax layer. The weighted residual module firstly uses a convolution kernel of 1 multiplied by 1 to reduce the channel dimension of an input signal so as to reduce the computational complexity, and then respectively performs feature extraction through convolution kernels (3 multiplied by 3,5 multiplied by 5 and 7 multiplied by 7) of three different scales, so that the module can extract feature structures under different sensing fields, and performs feature fusion in a weighted summation mode so as to extract wider features and reduce the computational complexity; meanwhile, the weighted residual error module can reduce the calculated amount of the network and alleviate the problems of gradient disappearance and gradient explosion in the back propagation process.
Step 2, for radar target frequency domain signal f (n) = [ f 0 ,f 1 ,…,f m ]Pretreatment is carried out, n=0, … m, f 0 ,f 1 ,...,f m The radar frequency domain signal is obtained after the ground reconnaissance radar signal is subjected to FFT, and m=1024; based on Toepliz matrixThe method comprises the steps of constructing an original one-dimensional target signal into a two-dimensional data plane as training and testing data for model training and testing, wherein the specific method comprises the following steps:
m∈even；
m∈odd；
becomes H Toepliz The completed two-dimensional data plane is constructed as in fig. 5.
Fig. 4 is a diagram illustrating a target frequency domain of a ground reconnaissance radar.
Step 3, processing the signals measured by all ground reconnaissance radars in the step 2 to obtain a total data set, classifying the total data set according to data types, and dividing the data of each type into a Training set (Training set) and a Test set (Test set) according to the proportion;
step 4, taking the training set data obtained in the step 3 as input, putting the training set data into the network in the step 1 for training, taking an Optimizer (Optimizer) as an Adam Optimizer for training, taking a Loss Function (Loss Function) as Sparse categorical cross entropy, and storing a training model after training;
and 5, testing the test set obtained in the step 3 by using the trained model.
The invention is described in further detail by the embodiments below, wherein the unconverted one-dimensional ground reconnaissance radar frequency domain target signal and the two-dimensional signal constructed by the invention are respectively used as the input of the weighted residual convolution neural network.
Example 1
10000 samples are taken as training data and 30000 samples are taken as test data from the ground reconnaissance radar frequency domain target data, a data set is constructed, after the data set is respectively constructed as a training set and a test set, a one-dimensional structure similar to the two-dimensional weighted residual convolution neural network structure in fig. 3 is directly adopted, and a trained model is tested by the test set to obtain a result.
Example 2
Using the data samples consistent with example 1, passing these sample data through the Toepliz matrix generation two-dimensional data plane in step 2 generates 10000 training samples, 30000 test samples. And then training the training set by using a 34-layer residual convolutional neural network (Res-34) network structure, obtaining a trained model, and testing the testing set by using the model to obtain result data.
Example 3
Using the data samples consistent with example 1, passing these sample data through the Toepliz matrix generation two-dimensional data plane in step 2 generates 10000 training samples, 30000 test samples. Then training the training set by using the network structure shown in FIG. 3 to obtain a trained model, and testing the test set by using the model to obtain result data
Table 1 is the test result accuracy data in examples 1, 2, 3:
TABLE 1
| Number of training wheels | Example 1 | Example 2 | Example 3 |
| 1 | 0.8980 | 0.9090 | 0.9340 |
| 2 | 0.9080 | 0.9210 | 0.9670 |
| 3 | 0.9180 | 0.9320 | 0.9740 |
| 4 | 0.9250 | 0.9300 | 0.9800 |
| 5 | 0.9400 | 0.9300 | 0.9760 |
| 6 | 0.9330 | 0.9480 | 0.9750 |
| 7 | 0.9300 | 0.9420 | 0.9660 |
| 8 | 0.9280 | 0.9550 | 0.9800 |
| 9 | 0.9290 | 0.9790 | 0.9690 |
| 10 | 0.9370 | 0.9670 | 0.9830 |
FIG. 6 is a graph showing the test results of examples 1, 2 and 3. The experimental result shows that the three-comparison between the embodiment 1 and the embodiment 2 shows that the classification effect of the two-dimensional convolution neural network on the ground reconnaissance radar signal is better than that of the one-dimensional convolution neural network; in comparison between examples 2 and 3, the model parameter of example 2 is 3500 ten thousand, the parameter of example 3 is 1900 ten thousand, and the model size of example 3 is nearly half as compared with example 2, but the classification accuracy of the model is increased by nearly two percentage points compared with that of example 2, so that the target classification task of the two-dimensional weighted residual neural network which is more suitable for the ground reconnaissance radar compared with the prior structure can be obtained.
Claims (2)
1. The radar target identification method based on the two-dimensional weighted residual convolution neural network is characterized by comprising the following steps of:
step 1, constructing a weighted residual convolutional neural network, which comprises ten weighted residual modules, five convolutional downsampling layers, a maximum pooling layer and two fully-connected layers, wherein each convolutional downsampling layer is connected with two weighted residual modules; the weighted residual error module firstly uses a convolution kernel of 1 multiplied by 1 to reduce the channel dimension of an input signal, and then respectively performs feature extraction through three convolution kernels with different scales, so that the module can extract feature structures under different sensing fields, and performs feature fusion in a weighted summation mode; the three convolution kernels have dimensions of 3×3,5×5, and 7×7, respectively;
step 2, constructing a one-dimensional radar signal into a two-dimensional data plane based on a Toepliz matrix, wherein an implementation algorithm is as follows:
let the original data be: f (n) = [ f 0 ,f 1 ,…,f m ]，n＝0,…m；
Wherein f 0 ,f 1 ,...,f m The radar frequency domain signal is obtained after the ground reconnaissance radar signal is subjected to FFT, and m=1024;
then it goes through the Toepliz matrix to become:
step 3, generating a training set and a testing set according to a certain proportion from the data processed in the step 2;
step 4, putting the generated training set into the network constructed in the step 1 for training to obtain a trained model;
and 5, testing the trained model by using the test set.
2. The radar target recognition method based on the two-dimensional weighted residual convolutional neural network according to claim 1, wherein the step 4 is specifically: and (3) taking the training set data obtained in the step (3) as input, putting the training set data into the network in the step (1) for training, wherein an Optimizer adopted for training is an Adam Optimizer, and a loss function adopted for training is Sparse categorical cross entropy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011067886.3A CN112329534B (en) | 2020-10-07 | 2020-10-07 | Radar target identification method based on two-dimensional weighted residual convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011067886.3A CN112329534B (en) | 2020-10-07 | 2020-10-07 | Radar target identification method based on two-dimensional weighted residual convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112329534A CN112329534A (en) | 2021-02-05 |
| CN112329534B true CN112329534B (en) | 2023-07-18 |
Family
ID=74314562
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011067886.3A Active CN112329534B (en) | 2020-10-07 | 2020-10-07 | Radar target identification method based on two-dimensional weighted residual convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112329534B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113989718A (en) * | 2021-10-29 | 2022-01-28 | 南京邮电大学 | Human body target detection method facing radar signal heat map |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764317A (en) * | 2018-05-21 | 2018-11-06 | 浙江工业大学 | A kind of residual error convolutional neural networks image classification method based on multichannel characteristic weighing |
| KR20190056622A (en) * | 2017-11-17 | 2019-05-27 | 주식회사 제이컴스 | Apparatus for portable radar measurement and method thereof |
| CN111624570A (en) * | 2020-04-30 | 2020-09-04 | 南京理工大学 | Radar target identification method based on two-dimensional convolutional neural network |
-
2020
- 2020-10-07 CN CN202011067886.3A patent/CN112329534B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20190056622A (en) * | 2017-11-17 | 2019-05-27 | 주식회사 제이컴스 | Apparatus for portable radar measurement and method thereof |
| CN108764317A (en) * | 2018-05-21 | 2018-11-06 | 浙江工业大学 | A kind of residual error convolutional neural networks image classification method based on multichannel characteristic weighing |
| CN111624570A (en) * | 2020-04-30 | 2020-09-04 | 南京理工大学 | Radar target identification method based on two-dimensional convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112329534A (en) | 2021-02-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110659591B (en) | SAR image change detection method based on twin network | |
| CN111077523B (en) | Inverse synthetic aperture radar imaging method based on generation countermeasure network | |
| Li et al. | Adversarial examples for CNN-based SAR image classification: An experience study | |
| CN112684427B (en) | Radar target recognition method based on serial secondary reinforcement training | |
| CN110619352A (en) | Typical infrared target classification method based on deep convolutional neural network | |
| CN108229551B (en) | Hyperspectral remote sensing image classification method based on compact dictionary sparse representation | |
| CN111624570B (en) | Radar target identification method based on two-dimensional convolutional neural network | |
| CN116258707A (en) | PCB surface defect detection method based on improved YOLOv5 algorithm | |
| CN115291210B (en) | 3D-CNN ground penetrating radar three-dimensional image pipeline identification method combined with attention mechanism | |
| CN114758230A (en) | Underground target body classification and identification method based on attention mechanism | |
| CN111583230A (en) | Local feature-based low-rank sparse decomposition hyperspectral anomaly detection method | |
| CN112329534B (en) | Radar target identification method based on two-dimensional weighted residual convolution neural network | |
| Ye et al. | Recognition algorithm of emitter signals based on PCA+ CNN | |
| CN111368653A (en) | Low-altitude small target detection method based on R-D (R-D) graph and deep neural network | |
| Liu et al. | A multi-scale feature pyramid SAR ship detection network with robust background interference | |
| CN110969203B (en) | HRRP data redundancy removing method based on self-correlation and CAM network | |
| CN117633588A (en) | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network | |
| CN112784777A (en) | Unsupervised hyperspectral image change detection method based on antagonistic learning | |
| CN111275680A (en) | SAR image change detection method based on Gabor convolution network | |
| CN109522830A (en) | A method of the sonar image feature selecting towards seafloor sediment classification | |
| CN115272865A (en) | Target detection method based on adaptive activation function and attention mechanism | |
| CN114818845A (en) | Noise-stable high-resolution range profile feature selection method | |
| CN113807186B (en) | Radar target identification method based on multichannel multiplexing convolutional neural network | |
| CN111860144B (en) | MDR-Net-based radar target recognition system and method | |
| Li et al. | Change detection in sar images based on a multi-scale attention convolution network |
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 |