CN113887278A - SAR image target identification method based on nonlinear manifold modeling - Google Patents
SAR image target identification method based on nonlinear manifold modeling Download PDFInfo
- Publication number
- CN113887278A CN113887278A CN202110970783.6A CN202110970783A CN113887278A CN 113887278 A CN113887278 A CN 113887278A CN 202110970783 A CN202110970783 A CN 202110970783A CN 113887278 A CN113887278 A CN 113887278A
- Authority
- CN
- China
- Prior art keywords
- target slice
- slice image
- manifold
- target
- image
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000013507 mapping Methods 0.000 claims abstract description 61
- 230000006870 function Effects 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 15
- 230000009466 transformation Effects 0.000 claims description 15
- 238000004891 communication Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 9
- 230000001172 regenerating effect Effects 0.000 claims description 8
- 238000004141 dimensional analysis Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 5
- 230000008929 regeneration Effects 0.000 abstract description 6
- 238000011069 regeneration method Methods 0.000 abstract description 6
- 238000003384 imaging method Methods 0.000 description 11
- 238000012545 processing Methods 0.000 description 10
- 238000002474 experimental method Methods 0.000 description 9
- 238000012549 training Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000002411 adverse Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 102100024506 Bone morphogenetic protein 2 Human genes 0.000 description 1
- 241000282326 Felis catus Species 0.000 description 1
- 101000762366 Homo sapiens Bone morphogenetic protein 2 Proteins 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a SAR image target identification method based on nonlinear manifold modeling, which comprises the following steps: acquiring a target slice image to be identified; representing nonlinear mapping of a target slice image to be recognized to manifold elements of Grassmann manifold by utilizing weighted linear combination of atoms of a pre-constructed abstract space dictionary by using corresponding weight coefficients; solving each weight coefficient; judging the image category of the target slice to be identified according to the minimum reconstruction error criterion; each atom of the abstract space dictionary is nonlinear mapping of manifold elements of a target slice image group of a category; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a subspace set by obtaining a multi-scale complex signal of the target slice image and carrying out multi-dimensional modulation, and embedding the manifold into a regeneration kernel Hilbert space to generate the manifold. The invention can effectively solve the random noise pollution problem and the target variation problem, and improve the identification precision.
Description
Technical Field
The invention belongs to the technical field of target identification, and particularly relates to an SAR image target identification method based on nonlinear manifold modeling.
Background
With the continuous development of sensor technology, the resolution of Synthetic Aperture Radar (SAR) images is continuously improved, and the existing data is completely enough to support the deep research of the automatic target identification technology based on the SAR images.
Over the last two decades, scholars at home and abroad put forward many solutions around the problem of target recognition, such as early template matching target recognition, model visual recognition strategies, machine learning recognition methods proposed in recent years, and the like.
However, when the target recognition of the SAR image is performed, these methods are usually based on a complete data set constructed under an ideal imaging condition, and cannot effectively solve the problem of random noise pollution and the problem of target variation existing in the actual task scene of the target recognition of the SAR image, so that the target recognition effect for the actual task scene is not good.
Disclosure of Invention
In order to solve the above problems in the prior art, embodiments of the present invention provide a method and an apparatus for identifying an SAR image target based on nonlinear manifold modeling, an electronic device, and a storage medium. The technical problem to be solved by the invention is realized by the following technical scheme:
in a first aspect, an embodiment of the present invention provides a method for identifying an SAR image target based on nonlinear manifold modeling, where the method includes:
acquiring a target slice image to be identified; the target slice image to be recognized belongs to an SAR image with unknown target category;
representing the nonlinear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold by utilizing the weighted linear combination of atoms of a pre-constructed abstract space dictionary by using corresponding weight coefficients;
solving each weight coefficient in the weighted linear combination;
judging the category of the target slice image to be recognized according to a minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving;
wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
In an embodiment of the present invention, the process of constructing the abstract space dictionary includes:
step 1, acquiring target slice image groups of multiple categories; wherein each category of target slice image group contains a plurality of target slice images belonging to the SAR image;
for each target slice image, executing step 2 to step 8:
step 2, performing two-dimensional Riesz transformation on the target slice image to obtain a corresponding Riesz transformation result;
step 3, constructing a multi-dimensional analysis complex signal of the target slice image by using the target slice image and a corresponding Riesz transformation result;
step 4, tuning the multi-dimensional analysis complex signal by using a band-pass filter to generate a corresponding multi-scale complex signal;
step 5, carrying out amplitude modulation and phase modulation on the complex signal of each scale in the multi-scale complex signal to obtain a plurality of groups of modulation components of the complex signal of the scale;
step 6, combining a plurality of groups of modulation components obtained by the complex signals of all scales in the multi-scale complex signals to form a subspace set;
step 7, constructing a Grassmann manifold by utilizing the subspace assembly;
step 8, embedding the Grassmann manifold into a regenerative kernel Hilbert space by introducing nonlinear mapping to obtain nonlinear mapping of manifold elements generated by the target slice image;
and 9, combining the nonlinear mapping of the manifold elements obtained from the target slice image groups of the multiple categories to obtain an abstract space dictionary.
In an embodiment of the present invention, the combining multiple groups of modulation components obtained from complex signals of all scales in the multi-scale complex signal to form a subspace set includes:
respectively resetting an amplitude modulation component and a phase modulation component obtained by the complex signal of each scale in the multi-scale complex signal into column vectors;
and combining column vectors obtained by complex signals with all scales into a two-dimensional matrix to obtain a subspace set.
In an embodiment of the present invention, the constructing the Grassmann manifold by using the subspace assembly includes:
and performing orthogonalization treatment on the elements of the subspace set, and simultaneously satisfying similar equivalence to construct and obtain Grassmann manifold.
In one embodiment of the present invention, solving the weight coefficients in the weighted linear combination comprises:
constructing an objective function for solving each weight coefficient according to the idea of solving the most sparse representation coefficient;
and defining a Grassmann kernel function, and solving the objective function to obtain each solved weight coefficient.
In an embodiment of the present invention, the constructing an objective function for solving the weight coefficients according to the idea of solving the sparsest representation coefficients includes:
constructing an initial function for solving each weight coefficient according to the idea of solving the sparsest expression coefficient as follows:
and rewriting the initial function into an optimization problem under an unconstrained condition to serve as an objective function for solving each weight coefficient:
wherein α represents a weight coefficient vector; x represents a set formed by all target slice image groups; y represents the target slice image to be identified; χ (·) represents a subspace set; phi (-) represents a non-linear mapping; i | · | purple windpRepresents a p-norm; ε represents the minimum constant; lambda represents the free parameters of the equalization reconstruction error and the sparsity;
wherein the content of the first and second substances,
the expansion terms obtained by the expansion are as follows:
in an embodiment of the present invention, the determining, according to a minimum reconstruction error criterion, the category of the target slice image to be recognized by using each atom of the abstract space dictionary and each weight coefficient obtained by solution includes:
judging by using a formula corresponding to the minimum reconstruction error criterion to obtain the category i of the target slice image to be identified; wherein, the formula corresponding to the minimum reconstruction error criterion is:
wherein phi (x (y)) represents the corresponding nonlinear mapping of the target slice image to be identified; phi (X (X)i) A non-linear mapping corresponding to the target slice image group representing the ith category;
the most sparse representation of the weight coefficient vector includes the solution values of the respective weight coefficients.
In a second aspect, an embodiment of the present invention provides a target recognition apparatus for an SAR image based on nonlinear manifold modeling, where the apparatus includes:
the image acquisition module is used for acquiring a target slice image to be identified; the target slice image to be recognized belongs to an SAR image with unknown target category;
the nonlinear mapping construction module is used for representing nonlinear mapping of the target slice image to be recognized to manifold elements of the Grassmann manifold by utilizing weighted linear combination of atoms of a pre-constructed abstract space dictionary by using corresponding weight coefficients;
the weight coefficient solving module is used for solving each weight coefficient in the weighted linear combination;
the category judgment module is used for judging the category of the target slice image to be recognized according to a minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving;
wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
In a third aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, wherein,
the memory is used for storing a computer program;
the processor is used for implementing the steps of the SAR image target identification method based on the nonlinear manifold modeling provided by the embodiment of the invention when executing the program stored in the memory.
In a fourth aspect, the embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the computer program implements the steps of the method for identifying a target of a SAR image based on nonlinear manifold modeling provided in the embodiment of the present invention.
In the scheme provided by the embodiment of the invention, a plurality of categories of target slice images are utilized in advance, a multi-scale complex signal of each target slice image is obtained, multi-dimensional modulation is carried out to obtain a subspace set, a Grassmann manifold is constructed, the Grassmann manifold is embedded into a regeneration kernel Hilbert space, and an abstract space dictionary of nonlinear mapping of manifold elements of the target slice image group with the atom as one category is generated. When any target slice image to be recognized is faced, the non-linear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold is expressed by utilizing the weighted linear combination of the atoms of the abstract space dictionary by using the corresponding weight coefficients, and the type of the target slice image to be recognized is judged according to the minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving.
According to the method, the target slice image does not need to be segmented and the radar shadow of the target image is extracted as in the prior art, but the whole structure and the local details are extracted from the angle of signal processing, so that the problem of low classification precision caused by errors introduced by segmentation or shadow extraction operation can be solved. The embodiment of the invention integrally deals with the problem of local disturbance of signals by virtue of the information richness of the subspace set, quantifies the similarity between the modulation component set and the set by utilizing the Grassmann manifold distance, and judges by utilizing the similarity, thereby effectively dealing with local disturbance such as random noise pollution, target variant and the like, and improving the target classification precision.
Drawings
Fig. 1 is a schematic flowchart of an SAR image target identification method based on nonlinear manifold modeling according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a construction process of an abstract space dictionary used in an SAR image target recognition method based on nonlinear manifold modeling according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of SAR imaging and variation differences of the same target provided by an embodiment of the present invention;
FIG. 4 is a diagram illustrating a sample after being subjected to different levels of random noise contamination according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an SAR image target recognition device based on nonlinear manifold modeling according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to effectively solve the problem of random noise pollution and the problem of target variation in the actual task scene of SAR image target identification and improve the identification precision, the embodiment of the invention provides a SAR image target identification method and device based on nonlinear manifold modeling, electronic equipment and a storage medium.
It should be noted that an execution subject of the SAR image target recognition method based on the nonlinear manifold modeling provided by the embodiment of the present invention may be a SAR image target recognition apparatus based on the nonlinear manifold modeling, and the apparatus may be run in an electronic device. The electronic device may be a server or a terminal device, but is not limited thereto.
In a first aspect, the embodiment of the invention provides a method for identifying an SAR image target based on nonlinear manifold modeling. As shown in fig. 1, the following steps may be included:
and S1, acquiring the target slice image to be recognized.
The target slice image to be recognized belongs to an SAR image with unknown target category.
And S2, performing weighted linear combination by using each atom of the pre-constructed abstract space dictionary and using the corresponding weight coefficient to represent the nonlinear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold.
And S3, solving each weight coefficient in the weighted linear combination.
And S4, judging the category of the target slice image to be recognized according to a minimum reconstruction error criterion by using each atom of the abstract space dictionary and each weight coefficient obtained by solving.
Wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
In order to facilitate understanding of the above-mentioned solution of the embodiment of the present invention, a construction process of the abstract space dictionary is first described, please refer to fig. 2, where fig. 2 is a schematic diagram illustrating a construction process of the abstract space dictionary used in an SAR image target recognition method based on nonlinear manifold modeling provided by the embodiment of the present invention, and the construction process of the abstract space dictionary includes steps 1 to 9:
step 1, acquiring target slice image groups of a plurality of categories.
Wherein each category of target slice image group contains a plurality of target slice images belonging to the SAR image.
In the embodiment of the invention, the target slice image refers to an image only containing a target area in the SAR image, and can be understood as a local image of the original SAR image. The embodiment of the invention can utilize the radar to collect the needed SAR image sets of different targets so as to put forward the target area image and obtain the target slice image groups of a plurality of categories, and can also directly utilize the existing various SAR data sets to obtain the target slice image groups of a plurality of categories.
Objects of embodiments of the present invention may include objects of interest in areas such as land, as well as sea, such as vehicles, buildings, etc. on land, or ships, islands, etc. in the sea. The target category may be set according to the scene requirement, and is not limited herein.
Specifically, it can be assumed that there is n in the i-th category of SAR image target slice image groupiThe target slice image group of the SAR image of the ith category can be marked as
XiEach element of which is an image of a target slice. K number ofAnd the target slice image group of the SAR image in the category is recorded as X ═ X1,X2,...,XK]All the K categories of SAR image target slice image groups are shared
And each target slice image is M multiplied by N pixels in size. Wherein n isiI and K are natural numbers greater than 0.
For each target slice image, executing step 2 to step 8:
and 2, performing two-dimensional Riesz transformation on the target slice image to obtain a corresponding Riesz transformation result.
For this target slice image, its two-dimensional Riesz transform is:
in the above equation, the first term on the left represents the two-dimensional Riesz transform, fR(x, y) represents the corresponding Riesz transform result; f represents the target slice image, which is a finite-length signal of a two-dimensional continuous integrable space; (x, y) represents coordinates; h is1And h2Is the impulse response of the distance and azimuth Riesz transformation.
Wherein the content of the first and second substances,
z is (x, y) and represents a coordinate index.
For convenience of the subsequent description, note:
h1*f＝fR1
h2*f＝fR2
for a specific process of the two-dimensional Riesz transformation, reference is made to the prior art for understanding, and a detailed description is not given here.
And 3, constructing a multi-dimensional analysis complex signal of the target slice image by using the target slice image and the corresponding Riesz transformation result.
In particular, the complex signal f is resolved in multiple dimensionsmExpressed as:
fm＝f(x,y)-(i,j)fR(x,y)
＝Re{fm}(x,y)+Im{fm}(x,y)
wherein, (i, j,1) represents a unit coordinate vector; re denotes the real part of the multi-dimensional analytic complex signal and Im denotes the imaginary part of the multi-dimensional analytic complex signal.
For the related concept and specific obtaining process of the multi-dimensional analytic complex signal, please refer to the prior art for understanding, and will not be described in detail herein.
And 4, tuning the multi-dimensional analysis complex signal by using a band-pass filter to generate a corresponding multi-scale complex signal.
This step can be expressed as:
......
wherein the multi-scale complex signal is represented as
A complex signal of S scale representing the tuning of the band pass filter;
a scale factor representing the S-th scale of the band pass filter.
Here, the specific form of the band pass filter adopted in the embodiment of the present invention is not limited, and for example, the band pass filter may be constructed by using an RLC oscillation loop, a log-Gabor filter, or the like.
And 5, carrying out amplitude modulation and phase modulation on the complex signal of each scale in the multi-scale complex signal to obtain multiple groups of modulation components of the complex signal of the scale.
Specifically, in the multiple groups of modulation components of the scale complex signal, the amplitude component and the phase component are respectively:
wherein a represents an amplitude component; phi represents the signal phase; θ denotes the imaginary phase.
It will be appreciated by those skilled in the art that frequency modulation is also incorporated into the amplitude modulation and phase modulation processes described above.
According to the embodiment of the invention, amplitude modulation and phase modulation are carried out on the complex signal of each scale in the multi-scale complex signal, so that the overall structure and the local details of the signal can be simultaneously extracted under multiple scales, and a subsequently constructed subspace set has information richness and detail integrity.
And 6, combining a plurality of groups of modulation components obtained by the complex signals of all scales in the multi-scale complex signals to form a subspace set.
In an alternative embodiment, step 6 may include steps a1 and a 2:
a1, resetting the amplitude modulation component and the phase modulation component obtained by the complex signal of each scale in the multi-scale complex signal into column vectors respectively.
Specifically, for the complex signal of each scale, the amplitude modulation component and the two phase modulation components are both two-dimensional matrices, and the two-dimensional matrices of the three modulation components are respectively reset into one-dimensional column vectors.
a2, combining the column vectors obtained by the complex signals of all scales into a two-dimensional matrix to obtain a subspace set.
Specifically, for the complex signal of each scale, three one-dimensional column vectors obtained by resetting are combined to obtain a combination matrix corresponding to the complex signal of the scale, and the subspace set can be obtained by combining the combination matrices obtained for the complex signals of all scales.
The above process is specifically expressed by a formula:
wherein χ { f } represents a set of subspaces;
a complex signal of S scale representing the tuning of the band pass filter; the cat function means that various modulation components of complex signals with all scales are reset into column vectors, and then are combined along the column direction to form a linear subspace set; m x d represents the dimensions of the subspace set.
According to the embodiment of the invention, the overall structure and the local details of the signal are extracted at the same time under multiple scales, and then different modulation components are combined to form the subspace set, so that the problem of local disturbance of the signal can be solved by means of the subspace set as a whole.
And 7, constructing the Grassmann manifold by utilizing the subspace assembly.
In an alternative embodiment, the step may include:
and performing orthogonalization treatment on the elements of the subspace set, and simultaneously satisfying similar equivalence to construct and obtain Grassmann manifold.
Specifically, the generated subspace set elements are subjected to orthogonalization treatment, so that column vectors are orthogonal pairwise; simultaneously satisfying similar equivalence: f. of1～f2If and only if span (f)1)～span(f2). Wherein f is1,f2Representing two column vectors; span (·) represents the linear subspace spanned by the column vectors.
Those skilled in the art will appreciate that through the above subspace transformation, a Grassmann manifold (Grassmannian manifold) can be constructed, which is denoted herein as Grassmann manifold
It will be appreciated by those skilled in the art that conventional data analysis typically results in failures in obtaining global geometric features. The reason for this is that space is globally linear in nature, whereas the data itself tends to exhibit strong non-linear characteristics. From the geometric point of view, the manifold essentially reflects the difference and the connection between the global property and the local property, and the manifold is a space with the Euclidean space property locally and is the extension of the concept of a curve and a curved surface in a high-dimensional space. Manifold learning is to find a low-dimensional manifold structure from high-dimensional observation data and to give a mapping from a high-dimensional space to a low-dimensional embedding. The main objective is to find the intrinsic regularity of the resulting data set from the desire, i.e. to find the essential features of the data from the observed data. Therefore, the embodiment of the invention can improve the classification precision by utilizing the Grassmann manifold to discover the relevance and the discriminant rule of the targets in the target slice data sets of various categories.
For the relevant concepts of the Grassmann manifold, reference is made to the prior art for an understanding and a detailed description is not provided herein.
And 8, embedding the Grassmann manifold into a regeneration kernel Hilbert space by introducing nonlinear mapping to obtain the nonlinear mapping of manifold elements generated by the target slice image.
In particular, non-linear mapping is introduced
Embedding Grassmann manifold into the reprocessed nuclear Hilbert space, i.e. if the target slice image is xi,1Then obtain its non-linear mapping phi (x)i,1) ); wherein the regenerated nuclear Hilbert space is described as
And 9, combining the nonlinear mapping of the manifold elements obtained from the target slice image groups of the multiple categories to obtain an abstract space dictionary.
Specifically, the abstract space dictionary is represented as:
D＝[φ(χ(X1)),φ(χ(X2)),...,φ(χ(XK))]
wherein phi (X)i) An abstract spatial sub-dictionary formed from a non-linear mapping of the manifold elements generated by the i-th class target slice image group.
Through the above processing, an abstract space dictionary can be obtained. The abstract space dictionary is utilized to perform category identification on any target slice image to be identified subsequently, and the construction process of the abstract space dictionary is not required to be performed again before each target slice image to be identified is identified.
The following describes specific steps of the method for identifying an SAR image target based on nonlinear manifold modeling according to the embodiment of the present invention.
For S1, the size of the target slice image to be recognized is consistent with the size of each target slice image used for constructing the abstract space dictionary, that is, the target slice image to be recognized y ∈ RM×N。
The target slice image to be recognized can be extracted from the original SAR image, and the specific manner belongs to the prior art and is not described herein.
For S2, the non-linear mapping of the target slice image to be recognized to the manifold element of the Grassmann manifold is represented as:
φ(χ(y))＝φ(χ(X1))α1+φ(χ(X2))α2+...+φ(χ(XK))αK
＝φ(χ(X))α
wherein α ═ α1,α2,...,αK]∈RnRepresenting a weight coefficient vector, wherein each weight coefficient is contained; phi (X (X)i) Represents a non-linear mapping of an atom in the abstract space dictionary, specifically a manifold element of the set of target slice images of the ith class; n represents allThe K categories of SAR image target slice image groups contain the total number of target slice images.
For S3, in an alternative embodiment, the solving of each weight coefficient in the weighted linear combination may include the following steps S31 and S32:
and S31, constructing an objective function for solving each weight coefficient according to the idea of solving the sparsest expression coefficient.
Optionally, S31 may include the following steps S311 and S312:
s311, constructing an initial function for solving each weight coefficient according to the idea of solving the most sparse representation coefficient as follows:
wherein α represents a weight coefficient vector; x represents a set formed by all target slice image groups; y represents the target slice image to be identified; χ (·) represents a subspace set; phi (-) represents a non-linear mapping; i | · | purple windpRepresents a p-norm; ε represents the minimum constant; λ represents the free parameters of the equalization reconstruction error and sparsity.
Embodiments of the invention utilize1Norm optimization algorithm solving for the sparsest representation of weight coefficient vector alpha
Since the constructed nonlinear mapping φ (-) is unknown, the initial function of the above equation cannot be computed directly, and therefore, a problem rewrite is required.
S312, the initial function is rewritten into an optimization problem under an unconstrained condition and is used as an objective function for solving each weight coefficient:
where λ represents the free parameters of the equalization reconstruction error and sparsity.
Wherein the content of the first and second substances,
the expansion terms obtained by the expansion are as follows:
and S32, defining a Grassmann kernel function, and solving the objective function to obtain each solved weight coefficient.
In particular, Grassmann kernel functions are defined
Converting the inner product of the image of the regenerated kernel Hilbert space into the similarity measurement of manifold elements, and solving the sparsest expression by using the formula of the objective function in S312
I.e. to solve the respective weight coefficients.
For S4, the determining the category of the target slice image to be recognized according to the minimum reconstruction error criterion by using each atom of the abstract space dictionary and each weight coefficient obtained by solution includes:
judging by using a formula corresponding to the minimum reconstruction error criterion to obtain the category i of the target slice image to be identified; wherein, the formula corresponding to the minimum reconstruction error criterion is:
wherein phi (x (y)) represents the corresponding nonlinear mapping of the target slice image to be identified; phi (X (X)i) A non-linear mapping corresponding to the target slice image group representing the ith category;
the most sparse representation of the weight coefficient vector includes the solution values of the respective weight coefficients.
In the scheme provided by the embodiment of the invention, a plurality of categories of target slice images are utilized in advance, a multi-scale complex signal of each target slice image is obtained, multi-dimensional modulation is carried out to obtain a subspace set, a Grassmann manifold is constructed, the Grassmann manifold is embedded into a regeneration kernel Hilbert space, and an abstract space dictionary of nonlinear mapping of manifold elements of the target slice image group with the atom as one category is generated. When any target slice image to be recognized is faced, the non-linear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold is expressed by utilizing the weighted linear combination of the atoms of the abstract space dictionary by using the corresponding weight coefficients, and the type of the target slice image to be recognized is judged according to the minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving.
According to the method, the target slice image does not need to be segmented and the radar shadow of the target image is extracted as in the prior art, but the whole structure and the local details are extracted from the angle of signal processing, so that the problem of low classification precision caused by errors introduced by segmentation or shadow extraction operation can be solved. The embodiment of the invention integrally deals with the problem of local disturbance of signals by virtue of the information richness of the subspace set, quantifies the similarity between the modulation component set and the set by utilizing the Grassmann manifold distance, and judges by utilizing the similarity, thereby effectively dealing with local disturbance such as random noise pollution, target variant and the like, and improving the target classification precision.
To illustrate the effectiveness of the methods of the embodiments of the present invention, experimental data are presented below.
(I) experimental conditions:
the embodiment of the invention utilizes the MSTAR SAR actual measurement data for verification, and the MSTAR data set is an SAR image data set published by an MSTAR plan and used for scientific research and is a data set commonly used for scientifically evaluating the performance of an SAR automatic target recognition system. The MSTAR dataset comprises 10 classes of ground tactical targets, BTR70 (armored transport vehicle), D7 (bulldozer), ZSU _234 (self-propelled antiaircraft), BRDM _2 (armored reconnaissance vehicle), T72 (tank), BTR _60 (armored transport vehicle), 2S1 (self-propelled grenade), ZIL131 (freight truck), T62 (tank) and BMP2 (infantry combat vehicle), respectively, each class of targets also has a different model, and targets of the same class but different models are equipped with some differences, but with different overall scattering characteristics.
Due to the difference of models of the same type of target, or the change of the state, configuration, etc. of the same type of target, such as the rotation of the gun barrel, the presence or absence of the oil tank, or the presence or absence of the crawler, etc., a variant target situation may occur, as shown in fig. 3, where fig. 3 is a schematic diagram of the SAR imaging and the variant difference of the same target provided by the embodiment of the present invention. In fig. 3, the gun barrel states of the armored vehicles are different, and the two images on the right side in each row are SAR images of the armored vehicles in the same row under different angles. It can be seen that in an actual task scenario, variant targets are common and have variability, and how to effectively perform variant target identification is a critical issue.
The parameters of the radar sensor for experimentally acquiring data in the embodiment of the invention are shown in the following table 1, an experimental operation system is an Intel (R) core (TM) i7-8565 CPU @1.80GHz and 64-bit Windows10 operating system, and simulation software adopts MATLAB (R2016 b).
TABLE 1 SAR image imaging parameters of experiments of embodiments of the invention
| Center frequency | 9.6GHz |
| Bandwidth of signal | 0.591GHz |
| Mode of operation | Strip imaging |
| Polarization mode | HH |
| Multiplicative noise | -10dB |
| Additive noise | -32～34dB |
| Dynamic range | 64dB |
| Azimuth beam width | 8.8 degree |
| Tilt angle beamwidth | 6.8 degree |
| Resolution ratio | 0.3X 0.3 m |
| Pixel pitch | 0.2X 0.2 m |
(II) analyzing the experimental content and the results:
the experiment of the embodiment of the invention focuses on the problems of random noise pollution and target variation in an actual imaging scene, and the following two aspects are verified respectively, and the advantages of the method of the embodiment of the invention are explained by comparing with a classical method. The method involved in the comparison is shown in Table 2.
TABLE 2 methods involved in experimental comparisons
The first set of experiments: the SAR slice images of 4 types of purposes in the MSTAR data set are selected for experiments, wherein data collected with a radar pitch angle of 17 degrees are used as reference samples for training, data collected with a radar pitch angle of 15 degrees are used as unknown samples for testing, and the experimental data are shown in table 3.
TABLE 3 training and test data for random noise pollution and target Change experiments
The embodiment of the invention selects part of MSTAR SAR actual measurement data in an experiment. In table 3, the first column indicates the pitch angle degrees; the numbers in the following columns indicate the number of samples, and the characters at the beginning of the SN indicate the specific model of the object.
For multiple variant target BMPs 2 and T72, only standard configuration samples (SN _9563, SN _13) are used for training, test samples of other configurations are predicted, and the target of an actual scene is simulated to change. For the training sample, randomly selecting a part of proportional amplitude values of the image, replacing the amplitude values by random numbers obeying uniform components, and simulating random noise pollution in the actual imaging process. The results of the experiment are shown in table 4.
For example, please refer to fig. 4, and fig. 4 is a schematic diagram of the same sample subjected to random noise pollution of different levels according to an embodiment of the present invention. The noise level gradually increases from left to right in fig. 4.
TABLE 4 experimental results on the variation of target noise pollution
The first row in table 4 represents different noise levels, e.g. 5% represents changing 5% of the amplitude values in the original image to noise; each numerical value smaller than 1 under the noise level represents the classification accuracy of the target class, and the reduction degree of each method represents the reduction percentage from the maximum classification accuracy to the minimum classification accuracy after different levels of noise are added under the method. The experimental result shows that the classification accuracy of the traditional method is greatly reduced along with the continuous aggravation of the applied noise level, and compared with the traditional method, the classification accuracy of the method provided by the embodiment of the invention is reduced only slightly, so that the adverse effect caused by noise can be effectively resisted.
The second set of experiments: the SAR slice images of three types of targets, namely BRDM2, 2S1 and ZSU23/4, in an MSTAR data set are selected for experiment, data acquired by a 17-degree radar pitch angle are adopted for training, category judgment is respectively carried out on imaging samples with the radar pitch angles of 30 degrees and 45 degrees, adverse effects of radar pitch angle changes on target identification in actual scenes are mainly considered, in addition, a plurality of groups of imaging results of variant targets (similar to different SAR images of the variant targets shown in figure 3) are simultaneously added into a test sample, the experimental data are shown in a table 5, and the data in brackets are the number of the added variant targets.
TABLE 5 target variants and Radar Pitch Angle variation Experimental data
| Pitch angle | 2S1 | BRDM2 | ZSU23/4 | Total up to |
| Training sample (17 degree) | 299 | 298 | 299 | 896 |
| Training sample (30 degree) | 288 | 287(133) | 288(118) | 1114 |
| Training sample (45 degree) | 303 | 303(120) | 303(119) | 1148 |
The experimental results are shown in table 6, and it can be seen that the abstract space dictionary constructed by the 17-degree radar pitch angle training has a good effect when performing classification judgment on the imaging data of the 30-degree radar pitch angle, and the classification accuracy generally reaches more than 95%. When the classification judgment is further carried out on the variant target imaging data of the 45-degree radar pitch angle, the performance is obviously reduced, because the scattering phenomenon of the target and the formed radar shadow are obviously changed by the 28-degree pitch angle which is different from the 17-degree radar pitch angle by the 45-degree radar pitch angle. However, compared with the conventional method, it is obvious that the decision proposed by the method of the embodiment of the present invention is more reliable and more robust.
Table 6 pitch angle change experimental results for variants
Therefore, the synthetic experimental data show that compared with the existing method, the SAR image target identification method based on the nonlinear manifold modeling provided by the embodiment of the invention can effectively cope with local disturbances such as random noise pollution and target variants, and the classification accuracy is improved.
In a second aspect, corresponding to the foregoing method embodiment, an embodiment of the present invention further provides a target identification apparatus for an SAR image based on nonlinear manifold modeling, as shown in fig. 5, where the apparatus includes:
an image obtaining module 501, configured to obtain a target slice image to be identified; the target slice image to be recognized belongs to an SAR image with unknown target category;
a nonlinear mapping construction module 502, configured to represent a nonlinear mapping of the target slice image to be recognized to a manifold element of a Grassmann manifold by using a weighted linear combination performed by using corresponding weight coefficients for each atom of a pre-constructed abstract space dictionary;
a weight coefficient solving module 503, configured to solve each weight coefficient in the weighted linear combination;
a category judgment module 504, configured to judge a category of the target slice image to be recognized according to a minimum reconstruction error criterion by using each atom of the abstract space dictionary and each weight coefficient obtained through solution;
wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
Further, the target identification device for the SAR image based on the nonlinear manifold modeling further includes: an abstract space dictionary building module, the abstract space dictionary building module comprising:
the image acquisition sub-module is used for acquiring target slice image groups of a plurality of categories; wherein each category of target slice image group contains a plurality of target slice images belonging to the SAR image;
a single target slice image processing module for executing, for each target slice image:
performing two-dimensional Riesz transformation on the target slice image to obtain a corresponding Riesz transformation result;
constructing a multi-dimensional analysis complex signal of the target slice image by using the target slice image and a corresponding Riesz transformation result;
tuning the multi-dimensional analytic complex signal by using a band-pass filter to generate a corresponding multi-scale complex signal;
carrying out amplitude modulation and phase modulation on the complex signal of each scale in the multi-scale complex signal to obtain a plurality of groups of modulation components of the complex signal of the scale;
combining a plurality of groups of modulation components obtained by complex signals of all scales in the multi-scale complex signal to form a subspace set;
constructing a Grassmann manifold by utilizing the subspace assembly;
and embedding the Grassmann manifold into a regeneration kernel Hilbert space by introducing nonlinear mapping to obtain the nonlinear mapping of manifold elements generated by the target slice image.
And the abstract space dictionary generation submodule is used for carrying out nonlinear mapping on the manifold elements obtained from the target slice image groups of the multiple categories and combining the manifold elements to obtain the abstract space dictionary.
Further, when the single-target slice image processing module combines multiple groups of modulation components obtained from complex signals of all scales in the multi-scale complex signal to form a subspace set, the single-target slice image processing module is specifically configured to:
respectively resetting an amplitude modulation component and a phase modulation component obtained by the complex signal of each scale in the multi-scale complex signal into column vectors;
and combining column vectors obtained by complex signals with all scales into a two-dimensional matrix to obtain a subspace set.
Further, when the single-target slice image processing module constructs the Grassmann manifold by using the subspace assembly, the single-target slice image processing module is specifically configured to:
and performing orthogonalization treatment on the elements of the subspace set, and simultaneously satisfying similar equivalence to construct and obtain Grassmann manifold.
Further, the weight coefficient solving module 503 is specifically configured to, when solving each weight coefficient in the weighted linear combination:
constructing an objective function for solving each weight coefficient according to the idea of solving the most sparse representation coefficient;
and defining a Grassmann kernel function, and solving the objective function to obtain each solved weight coefficient.
Further, the weight coefficient solving module 503 is specifically configured to, when constructing an objective function for solving each weight coefficient according to an idea of solving the sparsest representation coefficient:
constructing an initial function for solving each weight coefficient according to the idea of solving the sparsest expression coefficient as follows:
and rewriting the initial function into an optimization problem under an unconstrained condition to serve as an objective function for solving each weight coefficient:
wherein α represents a weight coefficient vector; x represents a set formed by all target slice image groups; y represents the target slice image to be identified; χ (·) represents a subspace set; phi (-) represents a non-linear mapping; i | · | purple windpRepresents a p-norm; ε represents the minimum constant; lambda represents the free parameters of the equalization reconstruction error and the sparsity;
wherein the content of the first and second substances,
the expansion terms obtained by the expansion are as follows:
further, when the class determination module 504 determines the class of the target slice image to be recognized according to the minimum reconstruction error criterion by using each atom of the abstract space dictionary and each weight coefficient obtained by solving, the class determination module is specifically configured to:
judging by using a formula corresponding to the minimum reconstruction error criterion to obtain the category i of the target slice image to be identified; wherein, the formula corresponding to the minimum reconstruction error criterion is:
wherein phi (x (y)) represents the corresponding nonlinear mapping of the target slice image to be identified; phi (X (X)i) A non-linear mapping corresponding to the target slice image group representing the ith category;
the most sparse representation of the weight coefficient vector includes the solution values of the respective weight coefficients.
For details, please refer to the method of the first aspect, which is not described herein.
In the scheme provided by the embodiment of the invention, a plurality of categories of target slice images are utilized in advance, a multi-scale complex signal of each target slice image is obtained, multi-dimensional modulation is carried out to obtain a subspace set, a Grassmann manifold is constructed, the Grassmann manifold is embedded into a regeneration kernel Hilbert space, and an abstract space dictionary of nonlinear mapping of manifold elements of the target slice image group with the atom as one category is generated. When any target slice image to be recognized is faced, the non-linear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold is expressed by utilizing the weighted linear combination of the atoms of the abstract space dictionary by using the corresponding weight coefficients, and the type of the target slice image to be recognized is judged according to the minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving.
According to the method, the target slice image does not need to be segmented and the radar shadow of the target image is extracted as in the prior art, but the whole structure and the local details are extracted from the angle of signal processing, so that the problem of low classification precision caused by errors introduced by segmentation or shadow extraction operation can be solved. The embodiment of the invention integrally deals with the problem of local disturbance of signals by virtue of the information richness of the subspace set, quantifies the similarity between the modulation component set and the set by utilizing the Grassmann manifold distance, and judges by utilizing the similarity, thereby effectively dealing with local disturbance such as random noise pollution, target variant and the like, and improving the target classification precision.
In a third aspect, an embodiment of the present invention further provides an electronic device, as shown in fig. 6, including a processor 601, a communication interface 602, a memory 603, and a communication bus 604, where the processor 601, the communication interface 602, and the memory 603 complete mutual communication through the communication bus 604,
a memory 603 for storing a computer program;
the processor 601 is configured to implement the steps of the method for identifying a target in a SAR image based on nonlinear manifold modeling according to the first aspect when executing the program stored in the memory 603.
The electronic device may be: desktop computers, laptop computers, intelligent mobile terminals, servers, and the like. Without limitation, any electronic device that can implement the present invention is within the scope of the present invention.
The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.
The communication interface is used for communication between the electronic equipment and other equipment.
The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.
Through above-mentioned electronic equipment, can realize: the random noise pollution problem and the target variation problem are effectively solved, and the identification precision is improved.
In a fourth aspect, corresponding to the method for identifying a target of a SAR image based on nonlinear manifold modeling provided in the first aspect, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method for identifying a target of a SAR image based on nonlinear manifold modeling provided in the embodiment of the present invention are implemented.
The computer-readable storage medium stores an application program which executes the method for identifying the target of the SAR image based on the nonlinear manifold modeling provided by the embodiment of the invention when the application program runs, so that the method can realize that: the random noise pollution problem and the target variation problem are effectively solved, and the identification precision is improved.
For the apparatus/electronic device/storage medium embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to part of the description of the method embodiment.
It should be noted that the apparatus, the electronic device, and the storage medium according to the embodiments of the present invention are respectively an apparatus, an electronic device, and a storage medium to which the above-described method for recognizing an SAR image based on nonlinear manifold modeling is applied, and all embodiments of the above-described method for recognizing an SAR image based on nonlinear manifold modeling are applicable to the apparatus, the electronic device, and the storage medium, and can achieve the same or similar beneficial effects.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.
Claims (10)
1. A SAR image target recognition method based on nonlinear manifold modeling is characterized by comprising the following steps:
acquiring a target slice image to be identified; the target slice image to be recognized belongs to an SAR image with unknown target category;
representing the nonlinear mapping of the target slice image to be recognized to the manifold elements of the Grassmann manifold by utilizing the weighted linear combination of atoms of a pre-constructed abstract space dictionary by using corresponding weight coefficients;
solving each weight coefficient in the weighted linear combination;
judging the category of the target slice image to be recognized according to a minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving;
wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
2. The SAR image target recognition method based on the nonlinear manifold modeling according to claim 1, wherein the construction process of the abstract space dictionary comprises:
step 1, acquiring target slice image groups of multiple categories; wherein each category of target slice image group contains a plurality of target slice images belonging to the SAR image;
for each target slice image, executing step 2 to step 8:
step 2, performing two-dimensional Riesz transformation on the target slice image to obtain a corresponding Riesz transformation result;
step 3, constructing a multi-dimensional analysis complex signal of the target slice image by using the target slice image and a corresponding Riesz transformation result;
step 4, tuning the multi-dimensional analysis complex signal by using a band-pass filter to generate a corresponding multi-scale complex signal;
step 5, carrying out amplitude modulation and phase modulation on the complex signal of each scale in the multi-scale complex signal to obtain a plurality of groups of modulation components of the complex signal of the scale;
step 6, combining a plurality of groups of modulation components obtained by the complex signals of all scales in the multi-scale complex signals to form a subspace set;
step 7, constructing a Grassmann manifold by utilizing the subspace assembly;
step 8, embedding the Grassmann manifold into a regenerative kernel Hilbert space by introducing nonlinear mapping to obtain nonlinear mapping of manifold elements generated by the target slice image;
and 9, combining the nonlinear mapping of the manifold elements obtained from the target slice image groups of the multiple categories to obtain an abstract space dictionary.
3. The SAR image target recognition method based on nonlinear manifold modeling according to claim 2, wherein the combining multiple groups of modulation components obtained from all scales of complex signals in the multi-scale complex signals to form a subspace set comprises:
respectively resetting an amplitude modulation component and a phase modulation component obtained by the complex signal of each scale in the multi-scale complex signal into column vectors;
and combining column vectors obtained by complex signals with all scales into a two-dimensional matrix to obtain a subspace set.
4. The SAR image target recognition method based on nonlinear manifold modeling according to claim 2 or 3, wherein the constructing Grassmann manifold by using the subspace set comprises:
and performing orthogonalization treatment on the elements of the subspace set, and simultaneously satisfying similar equivalence to construct and obtain Grassmann manifold.
5. The SAR image target recognition method based on nonlinear manifold modeling according to claim 1, wherein the solving of each weight coefficient in the weighted linear combination includes:
constructing an objective function for solving each weight coefficient according to the idea of solving the most sparse representation coefficient;
and defining a Grassmann kernel function, and solving the objective function to obtain each solved weight coefficient.
6. The SAR image target recognition method based on nonlinear manifold modeling according to claim 5, wherein the constructing an objective function for solving each weight coefficient according to an idea of solving a sparsest representation coefficient includes:
constructing an initial function for solving each weight coefficient according to the idea of solving the sparsest expression coefficient as follows:
and rewriting the initial function into an optimization problem under an unconstrained condition to serve as an objective function for solving each weight coefficient:
wherein α represents a weight coefficient vector; x represents a set formed by all target slice image groups; y represents the target slice image to be identified; χ (·) represents a subspace set; phi (-) represents a non-linear mapping; i | · | purple windpRepresents a p-norm; ε represents the minimum constant; lambda represents the free parameters of the equalization reconstruction error and the sparsity;
wherein the content of the first and second substances,
the expansion terms obtained by the expansion are as follows:
7. the SAR image target recognition method based on nonlinear manifold modeling according to claim 6, wherein the determining the category of the target slice image to be recognized according to a minimum reconstruction error criterion by using each atom of the abstract space dictionary and each weight coefficient obtained by solving comprises:
judging by using a formula corresponding to the minimum reconstruction error criterion to obtain the category i of the target slice image to be identified; wherein, the formula corresponding to the minimum reconstruction error criterion is:
wherein phi (x (y)) represents the corresponding nonlinear mapping of the target slice image to be identified; phi (X (X)i) A non-linear mapping corresponding to the target slice image group representing the ith category;
the most sparse representation of the weight coefficient vector includes the solution values of the respective weight coefficients.
8. A SAR image target recognition device based on nonlinear manifold modeling is characterized by comprising:
the image acquisition module is used for acquiring a target slice image to be identified; the target slice image to be recognized belongs to an SAR image with unknown target category;
the nonlinear mapping construction module is used for representing nonlinear mapping of the target slice image to be recognized to manifold elements of the Grassmann manifold by utilizing weighted linear combination of atoms of a pre-constructed abstract space dictionary by using corresponding weight coefficients;
the weight coefficient solving module is used for solving each weight coefficient in the weighted linear combination;
the category judgment module is used for judging the category of the target slice image to be recognized according to a minimum reconstruction error criterion by utilizing each atom of the abstract space dictionary and each weight coefficient obtained by solving;
wherein each atom of the abstract space dictionary is a nonlinear mapping of a manifold element of a class of target slice image group; and carrying out non-linear mapping on the manifold elements of each target slice image to obtain a multi-scale complex signal of the target slice image, carrying out multi-dimensional modulation to obtain a subspace set, and embedding the subspace set into a regenerative kernel Hilbert space to generate the Grassmann manifold by utilizing the subspace set.
9. An electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;
the memory is used for storing a computer program;
the processor, when executing the program stored in the memory, implementing the method steps of any of claims 1-7.
10. A computer-readable storage medium, characterized in that,
the computer-readable storage medium has stored therein a computer program which, when being executed by a processor, carries out the method steps of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970783.6A CN113887278A (en) | 2021-08-23 | 2021-08-23 | SAR image target identification method based on nonlinear manifold modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110970783.6A CN113887278A (en) | 2021-08-23 | 2021-08-23 | SAR image target identification method based on nonlinear manifold modeling |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113887278A true CN113887278A (en) | 2022-01-04 |
Family
ID=79011204
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110970783.6A Pending CN113887278A (en) | 2021-08-23 | 2021-08-23 | SAR image target identification method based on nonlinear manifold modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113887278A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956611A (en) * | 2016-04-25 | 2016-09-21 | 西安电子科技大学 | SAR image target identification method based on authentication non-linear dictionary learning |
| CN113093164A (en) * | 2021-03-31 | 2021-07-09 | 西安电子科技大学 | Translation-invariant and noise-robust radar image target identification method |
-
2021
- 2021-08-23 CN CN202110970783.6A patent/CN113887278A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956611A (en) * | 2016-04-25 | 2016-09-21 | 西安电子科技大学 | SAR image target identification method based on authentication non-linear dictionary learning |
| CN113093164A (en) * | 2021-03-31 | 2021-07-09 | 西安电子科技大学 | Translation-invariant and noise-robust radar image target identification method |
Non-Patent Citations (1)
| Title |
|---|
| GANGGANG DONG 等: "SAR Target Recognition Via Sparse Representation of Monogenic Signal on Grassmann Manifolds", 《IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pei et al. | SAR automatic target recognition based on multiview deep learning framework | |
| CN109086700B (en) | Radar one-dimensional range profile target identification method based on deep convolutional neural network | |
| Williams et al. | Mine classification with imbalanced data | |
| Lewis et al. | Generative adversarial networks for SAR image realism | |
| CN110163294B (en) | Remote sensing image change region detection method based on dimension reduction operation and convolution network | |
| CN112560966A (en) | Polarimetric SAR image classification method, medium and equipment based on scattergram convolution network | |
| Zhou et al. | A Two-Component $ K $–Lognormal Mixture Model and Its Parameter Estimation Method | |
| CN113111706B (en) | SAR target feature unwrapping and identifying method for azimuth continuous deletion | |
| CN112906564B (en) | Intelligent decision support system design and implementation method for automatic target recognition of unmanned airborne SAR (synthetic aperture radar) image | |
| CN115034257A (en) | Cross-modal information target identification method and device based on feature fusion | |
| CN112070151B (en) | Target classification and identification method for MSTAR data image | |
| CN113269706A (en) | Laser radar image quality evaluation method, device, equipment and storage medium | |
| CN116681927A (en) | Classification method and system for PolSAR data by combining deep learning model and traditional classifier | |
| Yin et al. | Cloud detection of high-resolution remote sensing image based on improved U-Net | |
| CN107403136B (en) | SAR target model identification method based on structure-preserving dictionary learning | |
| CN113887278A (en) | SAR image target identification method based on nonlinear manifold modeling | |
| CN116030300A (en) | Progressive domain self-adaptive recognition method for zero-sample SAR target recognition | |
| CN110135280B (en) | Multi-view SAR automatic target recognition method based on sparse representation classification | |
| CN108830290B (en) | SAR image classification method based on sparse representation and Gaussian distribution | |
| Lan et al. | Spatial-Transformer and Cross-Scale Fusion Network (STCS-Net) for Small Object Detection in Remote Sensing Images | |
| Gao et al. | Feature tracking for target identification in acoustic image sequences | |
| CN113988163A (en) | Radar high-resolution range profile identification method based on multi-scale grouping fusion convolution | |
| CN113989632A (en) | Bridge detection method and device for remote sensing image, electronic equipment and storage medium | |
| CN113762203B (en) | Cross-domain self-adaptive SAR image classification method, device and equipment based on simulation data | |
| Yu et al. | An Image Comparison Algorithm Based on Contour Similarity |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220104 |
|
| RJ01 | Rejection of invention patent application after publication |