CN116524358B - SAR data set amplification method for target recognition - Google Patents
SAR data set amplification method for target recognition Download PDFInfo
- Publication number
- CN116524358B CN116524358B CN202310460249.XA CN202310460249A CN116524358B CN 116524358 B CN116524358 B CN 116524358B CN 202310460249 A CN202310460249 A CN 202310460249A CN 116524358 B CN116524358 B CN 116524358B
- Authority
- CN
- China
- Prior art keywords
- image
- scattering
- sar
- target
- extraction module
- 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
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000003321 amplification Effects 0.000 title claims abstract description 14
- 238000003199 nucleic acid amplification method Methods 0.000 title claims abstract description 14
- 238000000605 extraction Methods 0.000 claims abstract description 49
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000003247 decreasing effect Effects 0.000 claims description 4
- 230000003190 augmentative effect Effects 0.000 claims 1
- 238000012512 characterization method Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 17
- 230000008569 process Effects 0.000 description 7
- 238000003384 imaging method Methods 0.000 description 4
- 230000008485 antagonism Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/0475—Generative 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
- G06N3/048—Activation functions
-
- 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
- G06N3/094—Adversarial learning
-
- 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
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to an SAR data set amplification method for target identification, which comprises the following steps: acquiring and preprocessing a SAR data set containing a target; extracting core characteristic information and scattering characteristic information of the target respectively through a core characteristic extraction module and a scattering characteristic extraction module; inputting the core characteristic information into a first-stage generation countermeasure network to obtain a low-resolution image; inputting the low-resolution image and the scattering characteristic information into a second-stage generation countermeasure network to obtain a high-resolution image, and amplifying the SAR data set. Through the implementation of the scheme, the two-stage generation countermeasure network, the core feature extraction module and the scattering feature extraction module are combined, so that rough core features and fine scattering features of targets in the stage learning images of the two-stage generation countermeasure network can be generated, the learning difficulty of the single-stage network is reduced, and SAR target image slices with high quality and more real details can be obtained through amplification.
Description
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to an SAR data set amplification method for target recognition.
Background
The synthetic aperture radar has all-weather imaging capability in the whole day, and along with the rapid development of space-based remote sensing technology, SAR image target recognition becomes an important research direction, and has important value in applications such as ocean monitoring, airport and port management. In recent years, a deep learning-based SAR image target recognition method has good effect, but the SAR image target recognition task faces the problem of lack of data samples due to high cost and time and labor consumption of the labeling of the SAR image.
Currently, the mainstream image generation method is generally based on generation of an countermeasure network, however, unlike an optical remote sensing image, SAR imaging is to actively emit electromagnetic waves by a SAR sensor and perform imaging by receiving scattering of the electromagnetic waves by a target. The specific scatter imaging mechanism of SAR images enables different types of targets to have different scatter characteristics. The general generation of the countermeasure network is not good enough when applied to the generation of SAR image targets, and the obtained result is far from the scattering characteristics of the real targets. Meanwhile, the target slice for target recognition is often large in size, the generation of a target sample with high resolution is difficult to learn by a countermeasure network at one time, and the reality of the generated result is generally poor.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention aims to provide an SAR data set amplification method for target identification, which can amplify and obtain SAR target image slices with high quality and real details.
In order to achieve the above purpose, the technical scheme of the invention is as follows:
the embodiment of the invention provides an SAR data set amplification method for target identification, which comprises the following steps:
Acquiring and preprocessing a SAR data set containing a target;
Extracting core characteristic information and scattering characteristic information of the target respectively through a core characteristic extraction module and a scattering characteristic extraction module;
Inputting the core characteristic information into a first-stage generation countermeasure network to obtain a low-resolution image;
inputting the low-resolution image and the scattering characteristic information into a second-stage generation countermeasure network to obtain a high-resolution image, and amplifying the SAR data set.
According to one aspect of an embodiment of the present invention, the preprocessing includes normalizing the image in the SAR dataset and cropping into 512 x 512 size image slices that contain the entire target.
According to an aspect of the embodiment of the present invention, the extracting, by the core feature extraction module, core feature information of the target includes:
Constructing a core feature extraction module, wherein the core feature extraction module comprises 3X 3 convolution layers for carrying out convolution processing on the image slice with the size of 512X 512, batchNorm batch normalization layers, a ReLU activation layer and a MaxPool maximum pooling layer which are contained between every two convolution layers, and one 3X 3 convolution layer and a Sigmoid activation layer for outputting a core feature map with the size of 64X 64; wherein each MaxPool max-pooling layer downsamples 2 times the size 512 x 512 image slice;
Independently training the core feature extraction module by using an SAR data set, superposing two-dimensional Gaussian distribution at the position of each target center point of the image slice as a true value, and using Focal Loss as a trained Loss function;
And extracting the position and direction information of the target in any image slice by using the trained core feature extraction module.
According to an aspect of the embodiment of the present invention, the extracting, by the scattering feature extracting module, scattering feature information of the target includes:
constructing a scattering feature extraction module, wherein the scattering feature extraction module comprises 3X 3 convolution layers for carrying out convolution processing on the image slice with the size of 512X 512, batchNorm batch normalization layers and ReLU activation layers contained between every two convolution layers, and one 3X 3 convolution layer and Sigmoid activation layer for outputting a strong scattering point prediction graph with the size of 512X 512;
the SAR data set is utilized to carry out independent training on the scattering feature extraction module, the Harris-Laplace algorithm is utilized to calculate an image slice to obtain a strong scattering point distribution map of the target as a true value, and the training loss function is as follows:
Wherein i and j are strong scattering point prediction graphs H is a strong scattering point distribution truth-value diagram;
and extracting strong scattering point distribution information of the target in any image slice by using a trained scattering feature extraction module.
According to an aspect of the embodiment of the present invention, the inputting the core feature information into the first-stage generation countermeasure network, to obtain a low resolution image, includes:
constructing a first-level generation countermeasure network, wherein the first-level generation countermeasure network comprises a generator G 1 and a discriminator D 1;
Jointly training the generator G 1 and the arbiter D 1 with a SAR dataset, the generator G 1 decreasing the objective function by successive iterations, the arbiter D 1 increasing the objective function by successive iterations;
And inputting the core characteristic information and random noise into the generator G 1 to obtain a low-resolution image with the size of 64 multiplied by 64 and containing target core characteristic information in the corresponding original SAR image, and distinguishing the low-resolution image and the image obtained by directly downsampling the original SAR image by 8 times by the discriminator D 1.
According to one aspect of the embodiment of the present invention, the objective function of the joint training of the generator G 1 and the arbiter D 1 using SAR data sets is:
where f c denotes core feature information, z 1 denotes random noise, And the image obtained by directly downsampling the original SAR image corresponding to the low-resolution image by 8 times is represented.
According to an aspect of the embodiment of the present invention, the inputting the low resolution image and the scattering feature information into a second stage to generate an countermeasure network, to obtain a high resolution image, and amplifying the SAR data set, includes:
constructing a second-level generation countermeasure network, the second-level generation countermeasure network including a generator G 2 and a discriminator D 2;
Jointly training the generator G 2 and the arbiter D 2 with a SAR dataset, the generator G 2 decreasing the objective function by successive iterations, the arbiter D 2 increasing the objective function by successive iterations;
And inputting the low-resolution image, the scattering characteristic information and random noise into the generator G 2 to obtain a high-resolution image with the size of 512 multiplied by 512, which accords with the target scattering characteristic information in the corresponding SAR image, and distinguishing the high-resolution image and the original SAR image corresponding to the high-resolution image by the discriminator D 2.
According to one aspect of the embodiment of the present invention, the objective function of the joint training of the generator G 2 and the arbiter D 2 using SAR data sets is:
Wherein f s represents scattering feature information, z 2 represents random noise, I 1 represents a low resolution image, and I represents an original SAR image corresponding to a high resolution image.
Compared with the prior art, the invention has the following beneficial effects:
according to the scheme provided by the embodiment of the invention, the coarse core characteristics and the fine scattering characteristics of the targets in the network hierarchical learning image can be realized by constructing the two-stage generation countermeasure network, and the first-stage generation countermeasure network and the second-stage generation countermeasure network are connected by adopting a series structure, so that the learning difficulty of the single-stage network is reduced.
According to the scheme of the embodiment of the invention, the core feature extraction module and the scattering feature extraction module are constructed, the rough core information and the fine scattering feature of the target in the obtained image are input into the two-stage generation countermeasure network to further guide the generator, so that SAR target slices with high quality and more real details are generated, and the amplification of a high-quality data set is realized.
The scheme of the embodiment of the invention provides a feasible technical approach for amplifying the data set for identifying the targets such as the airplane, the ship and the like in the SAR image, and has a larger practical application value in the field of remote sensing target identification.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments will be briefly described below. It is apparent that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained from these drawings without inventive effort for a person of ordinary skill in the art.
FIG. 1 schematically illustrates a flowchart for implementing a SAR dataset augmentation method for target recognition according to an embodiment of the present disclosure;
FIG. 2 schematically illustrates a block diagram of a two-level generation countermeasure network disclosed by an embodiment of the invention;
FIG. 3 schematically illustrates a block diagram of a core feature extraction module disclosed in an embodiment of the invention;
fig. 4 schematically shows a block diagram of a scattering feature extraction module according to an embodiment of the invention.
Detailed Description
The description of the embodiments of this specification should be taken in conjunction with the accompanying drawings, which are a complete description of the embodiments. In the drawings, the shape or thickness of the embodiments may be enlarged and indicated simply or conveniently. Furthermore, portions of the structures in the drawings will be described in terms of separate descriptions, and it should be noted that elements not shown or described in the drawings are in a form known to those of ordinary skill in the art.
Any references to directions and orientations in the description of the embodiments herein are for convenience only and should not be construed as limiting the scope of the invention in any way. The following description of the preferred embodiments will refer to combinations of features, which may be present alone or in combination, and the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims.
Fig. 1 schematically illustrates a specific implementation flow of a SAR data set amplification method for target identification according to an embodiment of the present invention. According to the method, the resolution of the generated image is improved step by constructing a two-stage generation countermeasure network, so that the learning difficulty of the single-stage network can be reduced. Meanwhile, a core feature extraction module and a scattering feature extraction module are constructed, coarse core information and fine scattering features of a target in an image are respectively obtained, the coarse core information and the fine scattering features are input into a two-stage generation countermeasure network to further guide a generator, and therefore SAR target slices with high quality and real details are obtained, and amplification of a high-quality data set is achieved. The amplified data set can be used for the subsequent fine recognition of targets such as airplanes, ships and the like.
As shown in fig. 1, the SAR data set amplification method disclosed in the present embodiment mainly includes the following steps:
S110, acquiring and preprocessing an SAR data set containing a target;
S120, respectively extracting core characteristic information and scattering characteristic information of the target through a core characteristic extraction module and a scattering characteristic extraction module;
s130, inputting core characteristic information into a first-stage generation countermeasure network to obtain a low-resolution image;
s140, inputting the low-resolution image and the scattering characteristic information into a second-stage generation countermeasure network to obtain a high-resolution image, and amplifying the SAR data set.
In the step S110, the preprocessing includes normalizing the image in the SAR data set, and cutting the image into 512×512 image slices including the whole target, that is, the SAR data set includes a plurality of 512×512 image slices after preprocessing. The size is suitable for various types of targets such as airplanes, ships and the like, can comprise the whole target, and is suitable for the input size of a subsequent network module.
That is, in the SAR data set amplification scheme for target recognition disclosed in this embodiment, an arbitrary type of target image slice is selected from the acquired SAR data set, core feature information in the target image slice is extracted by using the core feature extraction module, the core feature information of the target is input into the first stage in the constructed two-stage generation antagonism network, and a low-resolution image slice containing the core feature information of the target is output. And extracting target scattering feature information of the original slice by using a scattering feature extraction module, inputting the low-resolution image slice and the scattering feature information which are obtained in the previous process and correspond to the target into a second stage in a two-stage generation countermeasure network, and outputting a high-resolution image slice which accords with the scattering feature of the original slice. The type of the finally generated high resolution image slice is consistent with the original slice type. Selecting other target image slices in the SAR data set, and repeating the processes and steps to realize the fine amplification of the original SAR data set.
In the step S120, a specific implementation process of extracting the core feature information of the target by the core feature extraction module includes the following steps:
Firstly, constructing a core feature extraction module, as shown in fig. 3, wherein the core feature extraction module comprises 3×3 convolution layers for carrying out convolution processing on an image slice with the size of 512×512, batchNorm batch normalization layers, a ReLU activation layer and a MaxPool maximum pooling layer which are contained between every two convolution layers, and one 3×3 convolution layer and Sigmoid activation layer for outputting a core feature map with the size of 64×64; wherein each MaxPool max-pooling layer downsamples 2 times an image slice of size 512 x 512.
And then, independently training the core feature extraction module by utilizing the SAR data set, superposing two-dimensional Gaussian distribution at the position of each target center point of the image slice as a true value, and taking Focal Loss as a Loss function of training, so that the core feature extraction module has the capability of accurately extracting the target core features.
And extracting the position and direction information of the target in any image slice by using the trained core feature extraction module, and obtaining the core features of the target.
In the step S120, a specific implementation process of extracting the scattering feature information of the target by the scattering feature extraction module includes the following steps:
First, a scattering feature extraction module is constructed, as shown in fig. 4, and the scattering feature extraction module includes 3×3 convolution layers for performing convolution processing on an image slice of 512×512 in size, batchNorm batch normalization layers and ReLU activation layers included between each two convolution layers, and one 3×3 convolution layer and Sigmoid activation layer for outputting a strong scattering point prediction map of 512×512 in size.
Secondly, the SAR data set is utilized to train the scattering feature extraction module independently, and as the number and the spatial position distribution of the strong scattering points of different types of targets, such as airplanes, ships and the like, are different, the Harris-Laplace algorithm is utilized to calculate an image slice to obtain a strong scattering point distribution map of the target as a true value during training, so that the scattering feature extraction module has the capability of accurately extracting the scattering features of the target. The trained loss function is:
Wherein i and j are strong scattering point prediction graphs H is a strong scattering point distribution truth-value diagram;
and then, the strong scattering point distribution information of the target in any image slice is extracted by using the trained scattering feature extraction module, and the fine scattering feature of the target is obtained.
In the step S130, the specific implementation process of inputting the core feature information into the first-stage generation countermeasure network to obtain the low-resolution image includes the following steps:
A first level generation countermeasure network is constructed, as shown in fig. 2, including a generator G 1 and a discriminator D 1. The generator G 1 is a CNN network including one full connection layer and 45×5 deconvolution layers, and the arbiter D 1 is a CNN network including 45×5 deconvolution layers and one full connection layer, where each two convolution layers in the network include a BatchNorm batch normalization layer and a ReLU activation layer.
The SAR data set is utilized to carry out combined training on the generator G 1 and the discriminator D 1, the generator G 1 reduces the objective function through continuous iteration, and the discriminator D 1 increases the objective function through continuous iteration, wherein the objective function is as follows:
where f c denotes core feature information, z 1 denotes random noise, And the image obtained by directly downsampling the original SAR image corresponding to the low-resolution image by 8 times is represented.
The core characteristic information and random noise are input into a generator G 1 to obtain a low-resolution image with the size of 64 multiplied by 64 and containing target core characteristic information in the corresponding original SAR image, and a discriminator D 1 distinguishes the low-resolution image and the image obtained by directly downsampling the corresponding original SAR image by 8 times.
In the step S140, the specific implementation process of inputting the low-resolution image and the scattering feature information into the second-stage generation countermeasure network to obtain the high-resolution image and amplify the SAR data set includes the following steps:
A second level generation countermeasure network is constructed, as shown in fig. 2, including a generator G 2 and a discriminator D 2. The generator G 2 is a CNN network including one full connection layer and 45×5 deconvolution layers, and the arbiter D 2 is a CNN network including 45×5 deconvolution layers and one full connection layer, where each two convolution layers in the network include a BatchNorm batch normalization layer and a ReLU activation layer.
The SAR data set is utilized to carry out combined training on the generator G 2 and the discriminator D 2, the generator G 2 reduces the objective function through continuous iteration, and the discriminator D 2 increases the objective function through continuous iteration, wherein the objective function is as follows:
Wherein f s represents scattering feature information, z 2 represents random noise, I 1 represents a low resolution image, and I represents an original SAR image corresponding to a high resolution image.
The low-resolution image, the scattering characteristic information and the random noise are input into a generator G 2 to obtain a high-resolution image with the size of 512 multiplied by 512, which accords with the target scattering characteristic information in the corresponding original SAR image, and a discriminator D 2 distinguishes the high-resolution image and the corresponding original SAR image.
In this embodiment, the first-stage generation countermeasure network and the second-stage generation countermeasure network are connected by adopting a serial structure, so that coarse core features and fine scattering features of targets in the network hierarchical learning image can be achieved, and learning difficulty of a single-stage network is reduced.
The sequence numbers of the steps related to the method of the present invention do not mean the sequence of the execution sequence of the method, and the execution sequence of the steps should be determined by the functions and the internal logic, and should not limit the implementation process of the embodiment of the present invention in any way.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather to enable any modification, equivalent replacement, improvement or the like to be made within the spirit and principles of the invention.
Claims (8)
1. A method for SAR dataset amplification for target recognition, comprising:
Acquiring and preprocessing a SAR data set containing a target;
Extracting core characteristic information and scattering characteristic information of the target respectively through a core characteristic extraction module and a scattering characteristic extraction module;
Inputting the core characteristic information into a first-stage generation countermeasure network to obtain a low-resolution image containing target core characteristic information corresponding to the SAR image; wherein the first-level generation countermeasure network includes a generator G 1 and a arbiter D 1; the discriminator D 1 distinguishes the image obtained by directly downsampling the low-resolution image and the corresponding original SAR image by 8 times;
Inputting the low-resolution image and the scattering characteristic information into a second-stage generation countermeasure network to obtain a high-resolution image which accords with the scattering characteristic information of a target in a corresponding SAR image, and amplifying an SAR data set; wherein the second-stage generation countermeasure network includes a generator G 2 and a discriminator D 2, the discriminator D 2 distinguishing the high-resolution image and its corresponding raw SAR image.
2. The method of claim 1, wherein the preprocessing includes normalizing the image in the SAR dataset and cropping into 512 x 512 size image slices containing the entire target.
3. The method according to claim 2, wherein the extracting, by the core feature extraction module, core feature information of the object includes:
Constructing a core feature extraction module, wherein the core feature extraction module comprises 3X 3 convolution layers for carrying out convolution processing on the image slice with the size of 512X 512, batchNorm batch normalization layers, a ReLU activation layer and a MaxPool maximum pooling layer which are contained between every two convolution layers, and one 3X 3 convolution layer and a Sigmoid activation layer for outputting a core feature map with the size of 64X 64; wherein each MaxPool max-pooling layer downsamples 2 times the size 512 x 512 image slice;
Independently training the core feature extraction module by using an SAR data set, superposing two-dimensional Gaussian distribution at the position of each target center point of the image slice as a true value, and using Focal Loss as a trained Loss function;
And extracting the position and direction information of the target in any image slice by using the trained core feature extraction module.
4. The method of claim 2, wherein the extracting, by the scattering feature extraction module, scattering feature information of the target comprises:
constructing a scattering feature extraction module, wherein the scattering feature extraction module comprises 3X 3 convolution layers for carrying out convolution processing on the image slice with the size of 512X 512, batchNorm batch normalization layers and ReLU activation layers contained between every two convolution layers, and one 3X 3 convolution layer and Sigmoid activation layer for outputting a strong scattering point prediction graph with the size of 512X 512;
the SAR data set is utilized to carry out independent training on the scattering feature extraction module, the Harris-Laplace algorithm is utilized to calculate an image slice to obtain a strong scattering point distribution map of the target as a true value, and the training loss function is as follows:
Wherein i and j are strong scattering point prediction graphs And the position index of the strong scattering point distribution truth diagram H;
and extracting strong scattering point distribution information of the target in any image slice by using a trained scattering feature extraction module.
5. The method of claim 1, wherein inputting the core feature information into a first level generation countermeasure network results in a low resolution image, comprising:
Constructing a first-stage generation countermeasure network;
Jointly training the generator G 1 and the arbiter D 1 with a SAR dataset, the generator G 1 decreasing the objective function by successive iterations, the arbiter D 1 increasing the objective function by successive iterations;
And inputting the core characteristic information and random noise into the generator G 1 to obtain a low-resolution image with the size of 64 multiplied by 64 and containing target core characteristic information corresponding to the original SAR image.
6. The method of claim 5, wherein the objective function of jointly training the generator G 1 and the arbiter D 1 using SAR datasets is:
where f c denotes core feature information, z 1 denotes random noise, And the image obtained by directly downsampling the original SAR image corresponding to the low-resolution image by 8 times is represented.
7. The method of claim 1, wherein said inputting the low resolution image and the scattering characterization information into a second stage generation countermeasure network results in a high resolution image, and wherein augmenting the SAR data set comprises:
Constructing a second-stage generation countermeasure network;
Jointly training the generator G 2 and the arbiter D 2 with a SAR dataset, the generator G 2 decreasing the objective function by successive iterations, the arbiter D 2 increasing the objective function by successive iterations;
and inputting the low-resolution image, the scattering characteristic information and random noise into the generator G 2 to obtain a high-resolution image with the size of 512 multiplied by 512, which accords with the scattering characteristic information of the target in the corresponding SAR image.
8. The method of claim 7, wherein the objective function of jointly training the generator G 2 and the arbiter D 2 using SAR datasets is:
Wherein f s represents scattering feature information, z 2 represents random noise, I 1 represents a low resolution image, and I represents an original SAR image corresponding to a high resolution image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310460249.XA CN116524358B (en) | 2023-04-25 | 2023-04-25 | SAR data set amplification method for target recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310460249.XA CN116524358B (en) | 2023-04-25 | 2023-04-25 | SAR data set amplification method for target recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116524358A CN116524358A (en) | 2023-08-01 |
| CN116524358B true CN116524358B (en) | 2024-05-03 |
Family
ID=87404123
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310460249.XA Active CN116524358B (en) | 2023-04-25 | 2023-04-25 | SAR data set amplification method for target recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116524358B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886335A (en) * | 2014-04-09 | 2014-06-25 | 西安电子科技大学 | Polarized SAR image classifying method based on fuzzy particle swarms and scattering entropy |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
| CN107256414A (en) * | 2017-06-27 | 2017-10-17 | 哈尔滨工业大学 | Polarimetric SAR Image convolutional neural networks sorting technique based on spatial feature collection of illustrative plates |
| CN112801968A (en) * | 2021-01-22 | 2021-05-14 | 常州市第二人民医院 | Double-layer depth network model, method and device for nuclear magnetic image segmentation |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472627B (en) * | 2019-07-02 | 2022-11-08 | 五邑大学 | End-to-end SAR image recognition method, device and storage medium |
-
2023
- 2023-04-25 CN CN202310460249.XA patent/CN116524358B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886335A (en) * | 2014-04-09 | 2014-06-25 | 西安电子科技大学 | Polarized SAR image classifying method based on fuzzy particle swarms and scattering entropy |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
| CN107256414A (en) * | 2017-06-27 | 2017-10-17 | 哈尔滨工业大学 | Polarimetric SAR Image convolutional neural networks sorting technique based on spatial feature collection of illustrative plates |
| CN112801968A (en) * | 2021-01-22 | 2021-05-14 | 常州市第二人民医院 | Double-layer depth network model, method and device for nuclear magnetic image segmentation |
Non-Patent Citations (2)
| Title |
|---|
| Yi-Ying Lu.A Machine-Learning-based Data Classifier to Reduce the Write Amplification in SSDs.RACS '20: Proceedings of the International Conference on Research in Adaptive and Convergent Systems.2020,213–218. * |
| 基于数据分布的聚类联邦学习;常黎明等;计算机应用研究;20230119;第40卷(第6期);1697-1701 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116524358A (en) | 2023-08-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cui et al. | Image data augmentation for SAR sensor via generative adversarial nets | |
| CN108230329B (en) | Semantic segmentation method based on multi-scale convolution neural network | |
| CN110135267B (en) | Large-scene SAR image fine target detection method | |
| CN107239751B (en) | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network | |
| CN111077523B (en) | Inverse synthetic aperture radar imaging method based on generation countermeasure network | |
| CN111368896A (en) | Hyperspectral remote sensing image classification method based on dense residual three-dimensional convolutional neural network | |
| CN109035172B (en) | Non-local mean ultrasonic image denoising method based on deep learning | |
| Huang et al. | An intelligent ship image/video detection and classification method with improved regressive deep convolutional neural network | |
| CN112285712A (en) | Method for improving detection precision of ship on shore in SAR image | |
| CN112465759A (en) | Convolutional neural network-based aeroengine blade defect detection method | |
| CN111680755B (en) | Medical image recognition model construction and medical image recognition method, device, medium and terminal | |
| CN113536963B (en) | SAR image airplane target detection method based on lightweight YOLO network | |
| CN113240040A (en) | Polarized SAR image classification method based on channel attention depth network | |
| CN112699717A (en) | SAR image generation method and generation device based on GAN network | |
| CN111259733A (en) | Point cloud image-based ship identification method and device | |
| CN116091764A (en) | Cloud image segmentation method based on fusion transformation network | |
| CN116152678A (en) | Marine disaster-bearing body identification method based on twin neural network under small sample condition | |
| CN116188944A (en) | Infrared dim target detection method based on Swin-transducer and multi-scale feature fusion | |
| Shi et al. | Synthetic aperture radar SAR image target recognition algorithm based on attention mechanism | |
| CN113989612A (en) | Remote sensing image target detection method based on attention and generation countermeasure network | |
| CN117523394A (en) | SAR vessel detection method based on aggregation characteristic enhancement network | |
| CN116524358B (en) | SAR data set amplification method for target recognition | |
| CN111832463A (en) | Deep learning-based traffic sign detection method | |
| CN114998749B (en) | SAR data amplification method for target detection | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system |
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 |