CN110222631B

CN110222631B - Target recognition method of tangential space arrangement SAR image based on data block and parallel

Info

Publication number: CN110222631B
Application number: CN201910480219.9A
Authority: CN
Inventors: 于雪莲; 申威; 唐永昊; 周云
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2019-06-04
Filing date: 2019-06-04
Publication date: 2022-03-15
Anticipated expiration: 2039-06-04
Also published as: CN110222631A

Abstract

The invention belongs to the field of radar image processing, and particularly relates to a data block parallel based cut space arrangement SAR image target identification method, which is used for overcoming the limitations that an original LLTSA cannot process a large data set and the matrix order and the characteristic number are the same. The method comprises the steps of searching a blocking point with the minimum loss of target information amount in a target and noise distribution graph by analyzing pixel point distribution of an original data set, carrying out data blocking on each type of single-type original data according to the blocking point, and carrying out LLTSA on the blocked data respectively; by the method, when the dimension reduction is performed on the LLTSA, the content of the original information does not contain impurities to the maximum extent, the information of the target is completely reserved, and the dimension reduction of the subsequent steps can be set to a small value, so that the influence of speckle noise is effectively reduced; meanwhile, each type of single-type original data can be subjected to a data partitioning to target identification process in parallel, and the identification speed is greatly increased.

Description

Data block parallel based cut space arrangement SAR image target identification method

Technical Field

The invention belongs to the field of Radar image processing, and particularly relates to a Synthetic Aperture Radar (SAR) image target identification method, in particular to a data block parallel-based SAR image target identification method based on spatial permutation.

Background

Feature extraction is the most important operation link in the target recognition of the SAR image, and in recent years, a plurality of classical feature extraction algorithms, such as a principal component analysis algorithm and a linear discriminant analysis, and the like, are successfully applied to the SAR image feature extraction stage. But due to a unique imaging mechanism of the SAR image, the SAR image is very sensitive to a target azimuth angle, a pitch angle and environmental noise; that is, in the case of a small change in the target pose, global features such as the contour of the SAR image, the target silhouette, and the like become extremely unstable; therefore, in a real SAR target recognition scenario, the global feature extraction method faces a significant challenge.

Manifold learning, which is a new feature mining technology, has also been applied to the field of pattern recognition such as computer vision, radar high-resolution range profile, and the like; the method can keep the inherent local structure of the high-dimensional data and can excavate the potential low-dimensional structure in the original high-dimensional data, thereby attracting the research of a plurality of scholars. The Linear Local Tangent Space Alignment (LLTSA) algorithm is one of the representatives of the Local manifold learning algorithm, and has been successfully applied to radar high-resolution range profile identification because it can extract the Local features of the target; however, facing the problem of high-dimensional SAR image recognition, the biggest disadvantages of LLTSA are: in a feature decomposition part in the algorithm, the order number of the matrix must be the same as the number of samples, so that the processing of data sets with higher data dimensions is difficult due to the larger number of samples.

Disclosure of Invention

The invention aims to provide a SAR image target identification method of Data Block Parallel Linear indexing Space Alignment (DBPLLTSA) aiming at the problems of SAR image feature extraction; the method effectively overcomes the limitations that the original LLTSA can not process a large data set and the matrix order number is the same as the characteristic number, introduces the parallel thought, and greatly improves the operation efficiency; meanwhile, inherent local features of the original SAR image are excavated by utilizing linear local tangent space arrangement, so that the major challenge of a global feature extraction method is relieved, and the target identification performance of the SAR image is improved.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a data block parallel based cut space arrangement SAR image target recognition method comprises the following steps:

step 1: vectorizing each original image of each type of target to form single type of original data, and forming an original data set A (A) by all the single type of original data₁,A₂,...,A_m,...,A_MIn which A_mSingle-class original data representing an mth class target;

step 2: analyzing the pixel point distribution of the original data set A to obtain a target and noise distribution map; searching block points with the minimum target information loss in the target and noise distribution map, namely a target distribution maximum area and a noise purest area;

and step 3: partitioning data, and calculating a dimension reduction result of each block by using the LLTSA, specifically:

the single-class original data A of the mth class target_mCarrying out data blocking according to the blocking points obtained in the step 2 to obtain a data block A'_m：A′_m＝{A_m,1,A_m,2,A_m,3In which A_m,2Representing a target area, A_m,1、A_m,3All represent noise regions;

and 4, step 4: to A_m,1、A_m,2、A_m,3Respectively executing LLTSA to obtain dimension-reduced data blocks A ″_m＝{A′_m,1,A′_m,2,A′_m,3}；

And 4, step 4: for data block A ″)_m＝{A′_m,1,A′_m,2,A′_m,3Performing linear fusion to obtain a judgment feature set;

and 5: and finishing target identification aiming at the discrimination feature set by adopting a classifier.

Further, in step 2, the search process of the block point is as follows:

for single type original data A of m type target_mPerforming pixel point analysis to search the left boundary L of the target region_mAnd the right boundary R_m(ii) a Searching all single-type original data in the original data set A to perform pixel point analysis to obtain a left boundary set and a right boundary set, taking the minimum value in the left boundary set as a left block point, and taking the maximum value in the right boundary set as a right block point.

Further, in the step 4, A_m,1、A_m,2、A_m,3The dimensionality of the dimension reduction is a, b and c respectively, wherein b is larger than a and c.

The invention has the beneficial effects that:

the invention provides a cut space arrangement SAR image target identification method based on data blocking parallel, which comprises the steps of searching a blocking point with minimum target information loss in a target and noise distribution diagram by analyzing pixel point distribution of an original data set A, carrying out data blocking on each type of single original data according to the blocking point, and carrying out LLTSA on the blocked data respectively; by the method, when the dimension reduction is performed on the LLTSA, the content of the original information does not contain impurities to the maximum extent, the information of the target is completely reserved, some regions with less target distribution possibly carry certain noise, but the noise region can be a pure noise region, and the dimension reduction of the subsequent steps can be set to a small value, so that the influence of speckle noise is effectively reduced; meanwhile, each type of single-type original data can be subjected to a data partitioning to target identification process in parallel, and the identification speed is greatly increased.

In conclusion, the invention provides a data block parallel based cut space arrangement SAR image target identification method, which can greatly improve the target identification speed while ensuring the algorithm identification rate.

Drawings

FIG. 1 is a schematic flow chart of a data block parallel based cut space arrangement SAR image target recognition method of the present invention.

Fig. 2 shows a DBPLLTSA data blocking policy according to an embodiment of the present invention.

Fig. 3 is a three-dimensional view of the target and noise distributions in an embodiment of the present invention.

Detailed Description

Hereinafter, a detailed description will be given of an embodiment of the present disclosure in order to better embody the technical points of the present disclosure.

The present embodiment provides an SAR image recognition method based on data block parallel linear local tangent space arrangement, the flow of which is shown in fig. 1, and the method specifically includes the following steps:

step 2: analyzing the distribution of the pixel points of the original data set A to obtain a target and noise distribution map, as shown in FIG. 3; searching block points with the minimum target information loss in the target and noise distribution map, namely a target distribution maximum area and a noise purest area;

and 5: and verifying the calculated identification precision through a K-Nearest Neighbor (KNN) classifier and a Logistic Regression (LR) classifier, and finishing target identification and operation index recording.

In this example, the experimental data set is shown in table 1; the verification method uses an MSTAR standard data set and ten different types of ground targets: BMP2, BRDM _2, BTR70, BTR60, T72, 2S1, D7, T62, ZIL131, ZSU23_4, using the a-J alphabet respectively; the image size uniformly adopts 128 × 128 pixels, the parameter a is 10, the parameter b is 30, and the parameter c is 10; in step 2, the target region is [5001,11385 ]]I.e. the two cut points are: x is the number of₁Is 5001, x₂Is 11385.

TABLE 1

Through simulation tests, the performance comparison of the BPLLTSA and LLTSA algorithms is shown in Table 2, and the table shows that the running speed of the invention is improved by about 2.8 times under the condition of ensuring that the recognition rate is not changed, and the effect is obviously better than that of the LLTSA algorithm.

TABLE 2

Algorithm	Single sheet identification time(s)	KNN recognition rate (%)	LR recognition rate (%)
				LLTSA	2.37s	85.65	84.78
DBPLLTSA	0.84s	86.56	84.82

While the invention has been described with reference to specific embodiments, any feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise; all of the disclosed features, or all of the method or process steps, may be combined in any combination, except mutually exclusive features and/or steps.

Claims

1. A method for target recognition in SAR images based on data block and parallel tangent space arrangement, including the following steps:

Step 1: Perform vectorization processing on each original image of each type of target to form single-type original data, and form all single-type original data into the original data set A={A ₁ ,A ₂ ,...,A _m ,...,A _M }, where A _m represents the single-class raw data of the m-th target;

Step 2: Analyze the pixel distribution of the original data set A, and obtain the target and noise distribution map; search for the block points with the least loss of target information in the target and noise distribution maps, that is, the area with the largest target distribution and the purest area of noise;

Step 3: Divide the data into blocks for LLTSA to calculate the dimensionality reduction result of each block, specifically:

Divide the single-type original data Am of the _m -th target into data blocks according to the block points obtained in step 2, and obtain the data block A' _m : A' _m ={A _m,1 ,A _m,2 ,A _{m, 3} }, where _Am,2 represents the target area, and _Am,1 and _Am,3 both represent the noise area;

Step 4: Perform LLTSA on Am, ₁ , Am, ₂ , Am _,3 respectively, and obtain a dimensionality-reduced data block A″ _m ={A′ _m,1 ,A′ _m,2 ,A′ _m,3 };

Step 4: Perform linear fusion on the data block A″ _m ={A′ _m,1 ,A′ _m,2 ,A′ _m,3 } to obtain a discriminant feature set;

Step 5: Use the classifier to complete the target recognition for the discriminative feature set.

2. according to claim 1, it is characterized in that, in described step 2, the search process of block point is:

Perform pixel point analysis on the single-type original data A _m of the m-th object, search the left boundary L _m and right boundary R _m of the target area; search all single-type original data in the original data set A for pixel point analysis, and obtain the left boundary Set and right boundary set, take the minimum value in the left boundary set as the left block point, and take the maximum value in the right boundary set as the right block point.

3. according to claim 1, it is characterized in that, in the described step 4, the dimensionality reduction of Am _,1 , _Am,2 , _Am,3 is based on the tangential space arrangement SAR image target recognition method of data block and parallel, it is characterised in that The dimensions of are a, b, and c, where b is greater than a and c.