CN114926745A - Small-sample SAR target identification method based on domain feature mapping - Google Patents

Small-sample SAR target identification method based on domain feature mapping Download PDF

Info

Publication number
CN114926745A
CN114926745A CN202210568939.2A CN202210568939A CN114926745A CN 114926745 A CN114926745 A CN 114926745A CN 202210568939 A CN202210568939 A CN 202210568939A CN 114926745 A CN114926745 A CN 114926745A
Authority
CN
China
Prior art keywords
sar
feature
domain
target
sample
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
Application number
CN202210568939.2A
Other languages
Chinese (zh)
Other versions
CN114926745B (en
Inventor
杨建宇
黄钰林
盛自维
裴季方
霍伟博
张寅�
杨海光
张永超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202210568939.2A priority Critical patent/CN114926745B/en
Publication of CN114926745A publication Critical patent/CN114926745A/en
Application granted granted Critical
Publication of CN114926745B publication Critical patent/CN114926745B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Abstract

The invention discloses a domain feature mapping small-sample SAR target recognition method, which is applied to the field of radar target recognition and aims at solving the problem that enough samples are difficult to obtain for recognition training in the prior art; the method solves the problems of characteristic isomerism and difficulty in utilization of auxiliary information when source information is distributed in a differentiated mode for assistance, homogenizes heterogeneous data characteristic distribution, effectively utilizes the auxiliary information of the optical target sample and the simulated SAR target sample, and improves the SAR target identification accuracy rate of other source-assisted few samples.

Description

Small-sample SAR target identification method based on domain feature mapping
Technical Field
The invention belongs to the field of synthetic aperture radars, and particularly relates to a target identification technology.
Background
Synthetic Aperture Radar (SAR) is an active microwave sensor, has the characteristics of all-time and all-weather operation, can effectively identify camouflage and penetration masks, obtains high-resolution SAR samples similar to optical samples under meteorological conditions with extremely low visibility, and is widely applied to reconnaissance, monitoring and identification of targets. The SAR Automatic Target Recognition (ATR) technology realizes preprocessing, feature extraction and detection Recognition of a Target sample by means of a computer, and provides powerful support for battlefield reconnaissance, information analysis and the like.
The existing SAR ATR technology needs a large amount of target samples for training, also needs to consume a large amount of human resources for collecting samples and labeling labels. However, in practical applications, it is often difficult to obtain enough samples for recognition training, resulting in poor generalization capability of the model. Therefore, in order to avoid over-fitting of the self-adaptive identification model under the condition of few samples, the source auxiliary data is used, the supervision identification information is enriched, and the SAR target identification rate of few samples is improved.
The documents "Malmgren-Hansen D, Kusk A, Dall J, et al, improving SAR automatic target recognition modules with transfer learning from complex data [ J ]. IEEE Geoscience and Remote Sensing Letters,2017,14(9):1484 1488". based on the idea of transfer learning, a large number of simulated SAR target samples are used for pre-training a network, and then network parameters are finely adjusted in an SAR data set, so that the SAR target recognition accuracy is effectively improved. However, the method has limited improvement on the identification accuracy under the condition of small samples, and a complete simulated SAR data set usually depends on strong professional knowledge and special electromagnetic simulation software, so that the acquisition of the simulated SAR samples is difficult to a certain extent. For this reason, the document "Zhong C, Mu X, He X, et al, SAR target image classification based on transfer and model compression [ J ]. IEEE Geoscience and Remote Sensing Letters,2018,16(3): 412-. However, the method does not fully consider the difference of feature distribution between the optical sample and the SAR sample, the auxiliary identification information of the optical sample is not fully utilized, and the small-sample SAR target identification capability is limited.
Disclosure of Invention
In order to solve the technical problem, the invention provides a domain feature mapping small-sample SAR target identification method, which is characterized in that by learning the feature mapping relation between an optical sample and an SAR sample, heterogeneous data is homogenized to assist different domain classification features in a small-sample identification task, and domain feature mapping small-sample SAR target identification of source data supervision identification information is fully utilized.
The technical scheme adopted by the invention is as follows: a domain feature mapping small sample SAR target recognition method comprises the following steps:
s1, collecting an original SAR target sample and an optical target sample;
s2, simulating an SAR target sample;
s3, preprocessing the original SAR target sample, the simulated SAR target sample and the optical target sample;
s4, constructing a source-assisted few-sample SAR target identification data set;
s5, constructing a deep neural network of domain feature mapping;
s6, training the deep neural network of the domain feature mapping constructed in the step S5 by adopting the small-sample SAR target recognition data set constructed in the step S4, so as to obtain a small-sample SAR recognition model;
and S7, adopting the small sample SAR recognition model trained in the step S6 to recognize the target.
The invention has the beneficial effects that: compared with the prior art, the method provided by the invention overcomes the problems of characteristic isomerism and difficulty in utilization of auxiliary information when the source information is differentially distributed for assistance, homogenizes the characteristic distribution of heterogeneous data, effectively utilizes the auxiliary information of an optical target sample and a simulated SAR target sample, and improves the accuracy rate of source-assisted few-sample SAR target identification.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of a domain feature mapping deep neural network according to the present embodiment;
fig. 3 is a characteristic distribution of the SAR image domain and the optical domain according to the embodiment;
wherein, (a) is the distribution condition of SAR image domain and optical domain characteristics before domain characteristic mapping; (b) the distribution condition of the SAR image domain and the optical domain features after domain feature mapping is obtained;
FIG. 4 illustrates training and recognition accuracy for this embodiment;
wherein, (a) is the recognition rate of the training process, and (b) is the recognition rate of the testing process.
Detailed Description
For the convenience of describing the present invention, the following terms are explained herein.
The term 1: k-way N-shot
K-way N-shot represents the identification of N classes, each class having K training samples. The invention relates to 5-way 1-shot and 5-way 5-shot.
The term 2: support and query sets
The support set and the query set respectively represent a training set and a verification set on one task, and the data of the support set and the data of the verification set do not coincide with each other.
The specific implementation steps of the invention are described as follows with reference to the attached drawing 1:
s1, obtaining an original SAR target sample and an optical target sample: and acquiring a target image with the same resolution by using the SAR radar, and acquiring an optical image by using the optical sensor.
S2, simulating SAR target samples: keeping the simulation parameters consistent with the actual SAR radar parameters, such as resolution, carrier frequency and the like, and simulating the SAR target image by using electromagnetic simulation software.
S3, preprocessing the original SAR target sample, the simulated SAR target sample and the optical target sample; the specific implementation method comprises the following steps: cutting the acquired SAR target sample, the simulated SAR target sample and the optical target sample into image slices with the same size and the target positioned at the center, and carrying out normalization operation on the sliced samples: x represents the sample data before normalization, wherein the pixel value at the (i, j) th position can be represented as X (i, j), and X' represents the sample data after normalization, then
Figure BDA0003659438240000031
Wherein min [ X ]]Denotes the minimum value of the pixel values of the sample before normalization, max [ X ]]Representing the maximum value of the sample pixel values before normalization.
S4, constructing a source-assisted few-sample SAR target identification data set; the specific implementation method comprises the following steps: the method comprises the steps of equally dividing an actually measured SAR target sample into two parts according to a target class, wherein one part and a simulated SAR sample form an SAR domain data set
Figure BDA0003659438240000032
The other part is a less-sample SAR target identification test set for target identification performance verification
Figure BDA0003659438240000033
And all the optical target sample data are formed into an optical domain data set
Figure BDA0003659438240000034
Finally, training set in the small sample SAR target recognition dataset is composed of
Figure BDA0003659438240000035
And
Figure BDA0003659438240000036
composition of the test set
Figure BDA0003659438240000037
And the ratio of the actually measured SAR target sample to the simulated SAR target sample to the optical target sample is 1: 2: and 3, ensuring that the proportion of the SAR target sample is the same as that of the optical target sample, and preventing the network from difficultly considering the classification and identification of the optical target and the SAR target at the same time.
S5, constructing a deep neural network of domain feature mapping;
as shown in fig. 2, the deep neural network of domain feature mapping of the present invention comprises a feature extractor, a classifier and two domain feature mappers (optical and SAR); the feature extractor is formed by overlapping a convolution layer, two residual blocks and three maximum pooling layers; the classifier consists of a uniform pooling layer and a full-connection layer; the optical feature mapper consists of two residual blocks and two maximized pooling layer overlaps; the SAR feature mapper consists of two convolutional layers and two max-pooling layers.
S6, training the deep neural network of the domain feature mapping constructed in the step S5 by adopting the small sample SAR target recognition data set constructed in the step S4, thereby obtaining a small sample SAR recognition model; the specific implementation substep is as follows:
s61, pre-training of a feature extractor and a SAR image domain feature mapper. Suppose the SAR domain data set in the small-sample SAR target recognition training set constructed in step S4 is
Figure BDA0003659438240000038
Data set of optical domain is
Figure BDA0003659438240000041
X i And Y i Respectively, SAR domain sample data and a corresponding tag, i 1,2,
Figure BDA0003659438240000042
and Y i O Optical domain sample data and corresponding labels, respectively, and all tasks obey similarDistribution of
Figure BDA0003659438240000043
The number of target classes corresponding to the sample is respectively K SAR And K O Usually with K O ≥K SAR . And assuming that the optical domain and SAR image domain data share the parameter theta of the feature extractor E And a classifier parameter θ C′ The corresponding characteristic mapper parameters are respectively theta SAR And theta O . The parameter θ is then updated according to the following criteria E ,θ SAR ,θ C′ ：
Figure BDA0003659438240000044
Wherein alpha is SAR Representing the SAR image domain pre-training learning step size,
Figure BDA0003659438240000045
represents a loss operator, { (X) 1 ,Y 1 ),(X 2 ,Y 2 ),...,(X n ,Y n ) Expressing that K of n-size batches are randomly drawn in the SAR image domain SAR Class target picture data and its corresponding label data.
S62, pre-training the optical domain feature mapper. Randomizing the optical domain data set in the small-sample SAR target recognition training set constructed in step S4
Figure BDA0003659438240000046
Middle extraction K SAR The optical-like samples form a sub data set used for learning an interactive mapping mechanism between the optical domain characteristics and the SAR image domain characteristics. In the feature mapper T O In the training process, firstly, the parameters of the feature extractor and the classifier need to be fixed, and then the learning step length is reduced to update the feature mapper T O Parameter theta O And reshaping the feature distribution of the optical domain under the feature extractor E. Parameter theta O The update rule is as follows:
Figure BDA0003659438240000047
and S63, adjusting the structure of the classifier. And adjusting the output dimensionality of the classifier according to the requirement of the SAR target identification task with few samples to meet the number of categories to be identified. In particular, assume that the pre-training phase classifier completes f (θ) C′ ):
Figure BDA0003659438240000048
Wherein f (-) is a non-linear transformation function,
Figure BDA0003659438240000049
the method is a real number space, W and P are respectively a feature dimension and the number of classes of a target in a pre-training stage, and if the number of classes of a few-sample SAR target recognition task is Q, a classifier needs to be adjusted to be f (theta) C ):
Figure BDA00036594382400000410
And S64, training the SAR target recognition network with few domain feature mapping samples. The sub-step utilizes a Model-independent Meta-Learning (MAML) algorithm to train the feature extractor and classifier parameters.
Assuming that the iteration times are T, in the T iteration of meta-learning, randomly extracting a few-sample task with the batch size of T in the SAR image domain data set and the optical domain data set respectively
Figure BDA0003659438240000051
And
Figure BDA0003659438240000052
wherein the data of each task is composed of a support set S and a query set Q, i.e.
Figure BDA0003659438240000053
And assume that extractor E, SAR image domain feature mapper T SAR Optical domain feature mapper T O And the parameters of the classifier C are respectively represented by theta E 、θ SAR 、θ O And theta C Wherein theta E 、θ SAR And theta O Inherit the pre-training stage of learning the two domain feature mapper parameters and the feature extractor parameters in steps S61 and S62, and the feature mapper parameter θ SAR And theta O Remains constant throughout this phase, θ C Is randomly generated, a feature extractor parameter theta E Training at S61 yields initial values, which are fixed at S62. Then, in the t-th iteration, the parameter feature extractor parameter and the classifier parameter sequentially execute the following update rules:
Figure BDA0003659438240000054
Figure BDA0003659438240000055
Figure BDA0003659438240000056
Figure BDA0003659438240000057
wherein the content of the first and second substances,
Figure BDA0003659438240000058
representing the parameters of the feature extractor and the classifier at the t-th iteration respectively,
Figure BDA0003659438240000059
respectively representing auxiliary parameters of the feature extractor and the classifier in the t iteration, wherein alpha is the learning step length of the MAML outer loop and alpha is in The step size is learned for the inner loop.
When updating network parameters using SAR image domain data, the inverse propagation path of the gradient is C → T SAR → E; with training aided by optical domain data, the inverse propagation path of the gradient is C → T O → E, network parameters are updated sequentially with (3) and (4), respectively, until the algorithm reaches convergence.
Fig. 3(a) shows the distribution of the SAR image domain and the optical domain features before feature mapping, and fig. 3(b) shows the distribution after feature mapping, wherein cyan and red represent the SAR image domain features and the optical domain features, respectively. It can be seen that before feature mapping, the SAR image domain and the optical domain features are far apart in space, and there is almost no overlap. After feature mapping, the SAR image domain and the optical feature are partially overlapped in the feature space. The result shows that after the SAR image domain characteristics and the optical domain characteristics respectively pass through the domain characteristic mapper designed by the invention, the distribution of characteristic values in a characteristic space tends to be the same. Fig. 4(a) and 4(b) show the training recognition rate iteration curve and the testing recognition rate iteration curve under the 5-way 1-shot and 5-way 5-shot recognition tasks, respectively. The result shows that the method can fully utilize the source auxiliary information, can quickly converge after a plurality of iterations under the condition of few samples, and can obtain higher recognition rate.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims (8)

1. A small sample SAR target recognition method of domain feature mapping is characterized by comprising the following steps:
s1, collecting an original SAR target sample and an optical target sample;
s2, simulating an SAR target sample;
s3, preprocessing the original SAR target sample, the simulated SAR target sample and the optical target sample;
s4, constructing a source-assisted few-sample SAR target identification data set;
s5, constructing a deep neural network of domain feature mapping;
s6, training the deep neural network of the domain feature mapping constructed in the step S5 by adopting the small-sample SAR target recognition data set constructed in the step S4, so as to obtain a small-sample SAR recognition model;
and S7, adopting the small sample SAR recognition model trained in the step S6 to recognize the target.
2. The method for identifying a SAR target with few samples according to claim 1, wherein the preprocessing of step S3 comprises: and cutting the acquired SAR target sample, the simulated SAR target sample and the optical target sample into image slices with the same size, wherein the target is positioned in the center, and performing normalization operation on the sliced samples.
3. The method for identifying the SAR target with less samples of the domain feature mapping according to claim 2, wherein the step S4 is implemented by the following steps: the original SAR target sample is equally divided into a training set and a test set according to the target category, and the simulated SAR target sample and the collected optical target sample are used for expanding the training set to finally form a source-assisted few-sample SAR target identification data set.
4. The method for SAR target recognition with few samples of domain feature mapping as claimed in claim 3, wherein the deep neural network of domain feature mapping of step S5 comprises a feature extractor, a classifier and two domain feature mappers; the two domain feature mappers are specifically an optical feature mapper and an SAR feature mapper; the input of the feature extractor is an SAR target to be identified, the output of the feature extractor is respectively used as the input of the optical feature mapper and the SAR feature mapper, the output of the optical feature mapper and the SAR feature mapper is used as the input of the classifier, and the output of the classifier is a target identification result;
the structure of the feature extractor comprises the following components in sequence: the first rolling layer, the first maximum pooling layer, the first residual block, the second maximum pooling layer, the second residual block and the third maximum pooling layer;
the structure of the classifier comprises in sequence: an average pooling layer and a full-link layer;
the structure of the optical feature mapper sequentially comprises: a third residual block, a fourth maximum pooling layer, a fourth residual block, and a fifth maximum pooling layer;
the structure of the SAR characteristic mapper sequentially comprises: a second convolution layer, a sixth maximum pooling layer, a third convolution layer, and a seventh maximum pooling layer.
5. The method for identifying a domain feature mapped SAR target as claimed in claim 4, wherein the step S6 includes the following sub-steps:
s61, SAR feature mapper training, specifically: parameter θ for shared feature extractor assuming optical feature mapper and SAR feature mapper E And a classifier parameter θ C′ (ii) a Inputting the samples in the SAR image domain data set into a deep neural network of domain feature mapping, and updating the parameter theta of the feature extractor E Classifier parameter θ C′ And SAR characteristic mapper parameter θ SAR ；
S62, training an optical characteristic mapper, specifically: parameter θ assuming shared feature extractor for optical feature mapper and SAR feature mapper E And a classifier parameter θ C′ (ii) a Inputting the samples in the optical domain data set into the deep neural network of the domain feature mapping, and updating the parameter theta of the optical feature mapper O ；
S63, adjusting the output dimension of the classifier according to the requirement of the less-sample SAR target recognition task;
s64, training the parameters of the feature extractor and the classifier by using a model independent element learning algorithm, which comprises the following steps: SAR characteristic mapper parameter θ updated according to step S61 SAR Step S61 updated parameter theta of feature extractor E Step S62, updating the parameters of the optical characteristic mapper; inputting randomly extracted samples in the SAR image domain data set and the optical domain data set into a deep neural network of domain feature mapping, and updating the parameters of the feature extractorNumber and classifier parameters.
6. The method for identifying SAR target with less samples of domain feature mapping as claimed in claim 5, wherein step S61 is to update the parameter θ according to the following criteria E ,θ SAR ,θ C′ ：
Figure FDA0003659438230000021
Wherein alpha is SAR Representing the SAR image domain pre-training learning step size,
Figure FDA0003659438230000022
represents a loss operator, { (X) 1 ,Y 1 ),(X 2 ,Y 2 ),...,(X n ,Y n ) Expressing that K of n-size batches are randomly drawn in the SAR image domain SAR Class target picture data and its corresponding label data.
7. The method for identifying SAR target with few samples of domain feature mapping as claimed in claim 6, wherein step S62 is to update the parameter θ according to the following criteria O ：
Figure FDA0003659438230000023
8. The method for identifying a domain feature mapped SAR target as claimed in claim 7, wherein step S64 is to update the feature extractor parameters and the classifier parameters according to the following criteria:
Figure FDA0003659438230000024
Figure FDA0003659438230000031
Figure FDA0003659438230000032
Figure FDA0003659438230000033
wherein the content of the first and second substances,
Figure FDA0003659438230000034
representing the parameters of the feature extractor and the classifier at the t-th iteration respectively,
Figure FDA0003659438230000035
respectively representing auxiliary parameters of the feature extractor and the classifier in the t iteration, wherein alpha is the learning step length of the MAML outer loop and alpha is in The step size is learned for the inner loop,
Figure FDA0003659438230000036
for the data in the SAR image domain dataset,
Figure FDA0003659438230000037
for the data in the optical domain data set,
Figure FDA0003659438230000038
representing data in a SAR image domain data query set,
Figure FDA0003659438230000039
representing data in the SAR image domain data support set,
Figure FDA00036594382300000310
representing the data in the optical domain data query set,
Figure FDA00036594382300000311
representing data in the SAR optical domain data support set.
CN202210568939.2A 2022-05-24 2022-05-24 Domain feature mapping small sample SAR target recognition method Active CN114926745B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210568939.2A CN114926745B (en) 2022-05-24 2022-05-24 Domain feature mapping small sample SAR target recognition method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210568939.2A CN114926745B (en) 2022-05-24 2022-05-24 Domain feature mapping small sample SAR target recognition method

Publications (2)

Publication Number Publication Date
CN114926745A true CN114926745A (en) 2022-08-19
CN114926745B CN114926745B (en) 2023-04-25

Family

ID=82810434

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210568939.2A Active CN114926745B (en) 2022-05-24 2022-05-24 Domain feature mapping small sample SAR target recognition method

Country Status (1)

Country Link
CN (1) CN114926745B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116094824A (en) * 2023-02-07 2023-05-09 电子科技大学 Detection system and method for few sample malicious traffic

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150324989A1 (en) * 2013-09-03 2015-11-12 Litel Instruments Method & system for high accuracy & reliability registration of multi modal imagery
CN107886123A (en) * 2017-11-08 2018-04-06 电子科技大学 A kind of synthetic aperture radar target identification method based on auxiliary judgement renewal learning
CN108345856A (en) * 2018-02-09 2018-07-31 电子科技大学 The SAR automatic target recognition methods integrated based on isomery convolutional neural networks
CN109712150A (en) * 2018-12-26 2019-05-03 航天恒星科技有限公司 Optical microwave image co-registration method for reconstructing and device based on rarefaction representation
CN109934282A (en) * 2019-03-08 2019-06-25 哈尔滨工程大学 A kind of SAR objective classification method expanded based on SAGAN sample with auxiliary information
CN110751061A (en) * 2019-09-29 2020-02-04 五邑大学 SAR image recognition method, device, equipment and storage medium based on SAR network
CN112990334A (en) * 2021-03-29 2021-06-18 西安电子科技大学 Small sample SAR image target identification method based on improved prototype network

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150324989A1 (en) * 2013-09-03 2015-11-12 Litel Instruments Method & system for high accuracy & reliability registration of multi modal imagery
CN107886123A (en) * 2017-11-08 2018-04-06 电子科技大学 A kind of synthetic aperture radar target identification method based on auxiliary judgement renewal learning
CN108345856A (en) * 2018-02-09 2018-07-31 电子科技大学 The SAR automatic target recognition methods integrated based on isomery convolutional neural networks
CN109712150A (en) * 2018-12-26 2019-05-03 航天恒星科技有限公司 Optical microwave image co-registration method for reconstructing and device based on rarefaction representation
CN109934282A (en) * 2019-03-08 2019-06-25 哈尔滨工程大学 A kind of SAR objective classification method expanded based on SAGAN sample with auxiliary information
CN110751061A (en) * 2019-09-29 2020-02-04 五邑大学 SAR image recognition method, device, equipment and storage medium based on SAR network
CN112990334A (en) * 2021-03-29 2021-06-18 西安电子科技大学 Small sample SAR image target identification method based on improved prototype network

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
CLAIRE BORYAN等: "Early Season Winter Wheat Identification Using Sentinel -1 Synthetic Aperture Radar (Sar) and Optical Data" *
PEI, JIFANG等: "Multiview Deep Feature Learning Network for SAR Automatic Target Recognition" *
周涛等: "基于多时相合成孔径雷达与光学影像的冬小麦种植面积提取" *
王博威等: "少量样本下基于孪生CNN的SAR目标识别" *
王程丽等: "基于迁移学习的小样本SAR船只目标识别" *
盛自维: "少样本SAR目标它源辅助识别方法研究" *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116094824A (en) * 2023-02-07 2023-05-09 电子科技大学 Detection system and method for few sample malicious traffic
CN116094824B (en) * 2023-02-07 2024-02-20 电子科技大学 Detection system and method for few sample malicious traffic

Also Published As

Publication number Publication date
CN114926745B (en) 2023-04-25

Similar Documents

Publication Publication Date Title
Zhao et al. ApLeaf: An efficient android-based plant leaf identification system
CN109558942B (en) Neural network migration method based on shallow learning
Kang et al. SAR image classification based on the multi-layer network and transfer learning of mid-level representations
CN107316013A (en) Hyperspectral image classification method with DCNN is converted based on NSCT
CN104866524A (en) Fine classification method for commodity images
CN111079847B (en) Remote sensing image automatic labeling method based on deep learning
CN110689043A (en) Vehicle fine granularity identification method and device based on multiple attention mechanism
Qiang et al. Identification of plant leaf diseases based on inception V3 transfer learning and fine-tuning
CN111126361B (en) SAR target identification method based on semi-supervised learning and feature constraint
Chen et al. Agricultural remote sensing image cultivated land extraction technology based on deep learning
CN113033321A (en) Training method of target pedestrian attribute identification model and pedestrian attribute identification method
Yang et al. Color texture segmentation based on image pixel classification
Geng et al. An improved helmet detection method for YOLOv3 on an unbalanced dataset
Fan Research and realization of video target detection system based on deep learning
Wei et al. Food image classification and image retrieval based on visual features and machine learning
Tao et al. Highly efficient follicular segmentation in thyroid cytopathological whole slide image
CN104573745A (en) Fruit fly classification method based on magnetic resonance imaging
Chen et al. Intelligent monitoring method of water quality based on image processing and RVFL‐GMDH model
CN114926745B (en) Domain feature mapping small sample SAR target recognition method
Wang et al. Pedestrian detection in infrared image based on depth transfer learning
CN116935188A (en) Model training method, image recognition method, device, equipment and medium
Chander et al. Comparative analysis on deep learning models for detection of anomalies and leaf disease prediction in cotton plant data
CN113902672A (en) Rare skin lesion classification system based on space transformation optimization element learning
Yaman et al. Image processing and machine learning‐based classification method for hyperspectral images
CN104573746A (en) Fruit fly type identification method based on magnetic resonance imaging

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