CN110222631A - Based on deblocking it is parallel cut space arrangement SAR image target recognition method - Google Patents

Based on deblocking it is parallel cut space arrangement SAR image target recognition method Download PDF

Info

Publication number
CN110222631A
CN110222631A CN201910480219.9A CN201910480219A CN110222631A CN 110222631 A CN110222631 A CN 110222631A CN 201910480219 A CN201910480219 A CN 201910480219A CN 110222631 A CN110222631 A CN 110222631A
Authority
CN
China
Prior art keywords
target
deblocking
data
lltsa
piecemeal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910480219.9A
Other languages
Chinese (zh)
Other versions
CN110222631B (en
Inventor
于雪莲
申威
唐永昊
周云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201910480219.9A priority Critical patent/CN110222631B/en
Publication of CN110222631A publication Critical patent/CN110222631A/en
Application granted granted Critical
Publication of CN110222631B publication Critical patent/CN110222631B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to radar image process field, it is specially a kind of based on deblocking it is parallel cut space arrangement SAR image target recognition method, to overcome original LLTSA that can not handle large data sets, matrix order limitation identical with characteristic.The present invention searches for target information amount in target and noise profile figure and loses the smallest piecemeal point, and carry out deblocking to every a kind of single class initial data according to piecemeal point, then carry out LLTSA respectively to data after piecemeal by the pixel distribution of analysis raw data set；It does so and guarantees when LLTSA does Dimensionality Reduction, the content of raw information does not include impurity maximumlly, the information of target is retained completely, the value of very little can be set in the dimensionality reduction of subsequent step, to effectively reduce the influence of coherent speckle noise；Meanwhile every a kind of single class initial data can carry out parallel deblocking to target identification process, the recognition speeds such as propose significantly.

Description

Based on deblocking it is parallel cut space arrangement SAR image target recognition method
Technical field
The invention belongs to radar image process field, in particular to a kind of synthetic aperture radar (Synthetic Aperture Radar, SAR) images steganalysis method, it is specially a kind of based on deblocking it is parallel cut space arrangement SAR Images steganalysis method.
Background technique
Feature extraction is most important operation link in SAR image target identification, in recent years, the feature extraction of many classics Algorithm, such as Principal Component Analysis Algorithm and linear discriminant analysis these global characteristics extracting methods have been successfully applied to SAR The image characteristics extraction stage.But due to the unique image-forming mechanism of SAR image, SAR image is to azimuth of target, pitch angle and ring Border noise is very sensitive；That is, in the case where the target carriage change of very little, profile, the target back of SAR image These global characteristics such as shadow become extremely unstable；Therefore, in true SAR target identification scene, global characteristics extracting method Face significant challenge.
Manifold learning also has been applied to computer vision, radar high-resolution as a kind of new feature mining technology Range Profile isotype identifies field；Due to the partial structurtes that it can keep high dimensional data intrinsic, and original high dimension can be excavated The potential low dimensional structures in, therefore attracted the research of numerous scholars.Linear local tangent space alignment (Linear Local Tangent Space Alignment, LLTSA) algorithm be local manifolds learning algorithm one of representative, since it can be extracted The local feature of target out, therefore it has been successfully applied to high Resolution Range Profile Identification of Radar；But face the SAR image of higher-dimension Identification problem, LLTSA maximum the disadvantage is that: the feature decomposition part in algorithm, order of matrix number must be identical as sample number, This is more to sample number, and the higher data set processing of data dimension is more difficult.
Summary of the invention
It is an object of the invention to above-mentioned SAR image feature extraction there are aiming at the problem that, propose a kind of deblocking simultaneously Line local tangent space alignment (Data Block Parallel Linear Local Tangent Space Alignment, DBPLLTSA) SAR image target recognition method；Effectively overcome original LLTSA that can not handle large data sets, square Order of matrix number limitation identical with characteristic, present invention introduces parallel ideas, greatly promote operation efficiency；Meanwhile passing through utilization Linear local tangent space alignment excavates the intrinsic local feature of original SAR image, alleviates global characteristics extracting method and faces Significant challenge, improve SAR image target identification performance.
To achieve the above object, The technical solution adopted by the invention is as follows:
Based on deblocking it is parallel cut space arrangement SAR image target recognition method, comprising the following steps:
Step 1: vectorization processing being done to each width original image of each classification target, single class initial data is constituted, by institute There is single class initial data to constitute raw data set A={ A1,A2,...,Am,...,AM, wherein AmIndicate m classification target list class Initial data；
Step 2: the pixel distribution of analysis raw data set A obtains target and noise profile figure；Search for target and noise Target information amount loses the smallest piecemeal point, the i.e. most pure region of target distribution maximum region and noise in distribution map；
Step 3: deblocking, with supply LLTSA calculate each piece dimensionality reduction as a result, specifically:
By m classification target list class initial data AmDeblocking is carried out according to the piecemeal point that step 2 obtains, obtains data Block A 'm: A 'm={ Am,1,Am,2,Am,3, wherein Am,2Indicate target area, Am,1、Am,3Indicate noise region；
Step 4: to Am,1、Am,2、Am,3LLTSA is executed respectively, obtains the data block A " of dimensionality reductionm={ A 'm,1,A′m,2,A ′m,3}；
Step 4: to data block A "m={ A 'm,1,A′m,2,A′m,3Linear fusion is carried out, it obtains differentiating feature set；
Step 5: target identification is completed for differentiation feature set using classifier.
Further, in the step 2, the search process of piecemeal point are as follows:
To m classification target list class initial data AmMake pixel point analysis, searches for the left margin L of target areamWith right margin Rm；All single class initial data do pixel point analysis in search raw data set A, obtain left margin set and right margin set, Minimum value is taken in left margin set as left piecemeal point, takes in right margin set maximum value as right piecemeal point.
Further, in the step 4, Am,1、Am,2、Am,3The dimension of dimensionality reduction be respectively a, b, c, wherein b be greater than a, c。
The beneficial effects of the present invention are:
The present invention provide it is a kind of based on deblocking it is parallel cut space arrangement SAR image target recognition method, by point The pixel distribution for analysing raw data set A searches for target information amount in target and noise profile figure and loses the smallest piecemeal point, and Deblocking is carried out to every a kind of single class initial data according to piecemeal point, then LLTSA is carried out respectively to data after piecemeal；It does so Guarantee when LLTSA does Dimensionality Reduction, the content of raw information, does not include impurity maximumlly, the information of target has been retained Whole, the less region of possible some target distributions can carry certain noise, but enabling to noise region is pure noise range The value of very little can be set in the dimensionality reduction in domain, subsequent step, to effectively reduce the influence of coherent speckle noise；Meanwhile it is each Class list class initial data can carry out parallel deblocking to target identification process, the recognition speeds such as propose significantly.
To sum up, the present invention provide it is a kind of based on deblocking it is parallel cut space arrangement SAR image target recognition method, While guaranteeing algorithm discrimination, target identification speed can be greatly promoted.
Detailed description of the invention
Fig. 1 be the present invention is based on deblocking it is parallel cut space arrangement SAR image target recognition method flow diagram.
Fig. 2 is DBPLLTSA deblocking strategy in the embodiment of the present invention.
Fig. 3 is target and noise profile 3-D view in the embodiment of the present invention.
Specific embodiment
It is described in detail below for the implementation method of the disclosure herein content, in order to preferably embody this hair Bright technical essential.
The present embodiment provides a kind of SAR image recognition methods based on deblocking parallel linear local tangent space alignment, Its process as shown in Figure 1, specifically includes the following steps:
Step 1: vectorization processing being done to each width original image of each classification target, single class initial data is constituted, by institute There is single class initial data to constitute raw data set A={ A1,A2,...,Am,...,AM, wherein AmIndicate m classification target list class Initial data；
Step 2: the pixel distribution of analysis raw data set A obtains target and noise profile figure, as shown in Figure 3；Search Target information amount loses the smallest piecemeal point, the i.e. most pure area of target distribution maximum region and noise in target and noise profile figure Domain；
Step 3: deblocking, with supply LLTSA calculate each piece dimensionality reduction as a result, specifically:
By m classification target list class initial data AmDeblocking is carried out according to the piecemeal point that step 2 obtains, obtains data Block A 'm: A 'm={ Am,1,Am,2,Am,3, wherein Am,2Indicate target area, Am,1、Am,3Indicate noise region；
Step 4: to Am,1、Am,2、Am,3LLTSA is executed respectively, obtains the data block A " of dimensionality reductionm={ A 'm,1,A′m,2,A ′m,3}；
Step 4: to data block A "m={ A 'm,1,A′m,2,A′m,3Linear fusion is carried out, it obtains differentiating feature set；
Step 5: passing through k nearest neighbor (K-Nearest Neighbor, KNN) classifier and logistic regression (Logistic Regression, LR) classifier verifying calculate accuracy of identification, complete target identification and operating index record.
In the present embodiment, experimental data set is as shown in table 1；Verification method uses MSTAR standard data set, not using ten Generic ground target: BMP2, BRDM_2, BTR70, BTR60, T72,2S1, D7, T62, ZIL131, ZSU23_4 make respectively With A-J alphabet；Image size is used uniformly 128 × 128 pixels, parameter a=10, b=30, c=10；In step 2, target area Domain is [5001,11385], i.e. two separations are as follows: x1It is 5001, x2It is 11385.
Table 1
By emulation testing, the algorithm performance comparison of BPLLTSA and LLTSA of the present invention are as shown in table 2, as seen from table, In the case where guaranteeing that discrimination is constant, the speed of service of the present invention promotes about 2.8 times, and effect is substantially better than LLTSA algorithm.
Table 2
Algorithm Individual recognition time (s) KNN discrimination (%) LR discrimination (%)
LLTSA 2.37s 85.65 84.78
DBPLLTSA 0.84s 86.56 84.82
The above description is merely a specific embodiment, any feature disclosed in this specification, except non-specifically Narration, can be replaced by other alternative features that are equivalent or have similar purpose；Disclosed all features or all sides Method or in the process the step of, other than mutually exclusive feature and/or step, can be combined in any way.

Claims (3)

1. based on deblocking it is parallel cut space arrangement SAR image target recognition method, comprising the following steps:
Step 1: vectorization processing being done to each width original image of each classification target, single class initial data is constituted, by all lists Class initial data constitutes raw data set A={ A1,A2,...,Am,...,AM, wherein AmIndicate that m classification target list class is original Data；
Step 2: the pixel distribution of analysis raw data set A obtains target and noise profile figure；Search for target and noise profile Target information amount loses the smallest piecemeal point, the i.e. most pure region of target distribution maximum region and noise in figure；
Step 3: deblocking, with supply LLTSA calculate each piece dimensionality reduction as a result, specifically:
By m classification target list class initial data AmDeblocking is carried out according to the piecemeal point that step 2 obtains, obtains data block A ′m: A 'm={ Am,1,Am,2,Am,3, wherein Am,2Indicate target area, Am,1、Am,3Indicate noise region；
Step 4: to Am,1、Am,2、Am,3LLTSA is executed respectively, obtains the data block A " of dimensionality reductionm={ A 'm,1,A′m,2,A′m,3}；
Step 4: to data block A "m={ A 'm,1,A′m,2,A′m,3Linear fusion is carried out, it obtains differentiating feature set；
Step 5: target identification is completed for differentiation feature set using classifier.
2. by described in claim 1 based on deblocking it is parallel cut space arrangement SAR image target recognition method, feature exists In, in the step 2, the search process of piecemeal point are as follows:
To m classification target list class initial data AmMake pixel point analysis, searches for the left margin L of target areamWith right margin Rm； All single class initial data do pixel point analysis in search raw data set A, obtain left margin set and right margin set, take a left side Minimum value takes in right margin set maximum value as right piecemeal point as left piecemeal point in the set of boundary.
3. by described in claim 1 based on deblocking it is parallel cut space arrangement SAR image target recognition method, feature exists In, in the step 4, Am,1、Am,2、Am,3The dimension of dimensionality reduction be respectively a, b, c, wherein b is greater than a, c.
CN201910480219.9A 2019-06-04 2019-06-04 Data block parallel based cut space arrangement SAR image target identification method Expired - Fee Related CN110222631B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910480219.9A CN110222631B (en) 2019-06-04 2019-06-04 Data block parallel based cut space arrangement SAR image target identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910480219.9A CN110222631B (en) 2019-06-04 2019-06-04 Data block parallel based cut space arrangement SAR image target identification method

Publications (2)

Publication Number Publication Date
CN110222631A true CN110222631A (en) 2019-09-10
CN110222631B CN110222631B (en) 2022-03-15

Family

ID=67819220

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910480219.9A Expired - Fee Related CN110222631B (en) 2019-06-04 2019-06-04 Data block parallel based cut space arrangement SAR image target identification method

Country Status (1)

Country Link
CN (1) CN110222631B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101079104A (en) * 2007-06-14 2007-11-28 上海交通大学 Human face identification method based on information
CN104008383A (en) * 2014-06-24 2014-08-27 哈尔滨工业大学 Hyperspectral image characteristic extraction algorithm based on manifold learning linearization
CN104077594A (en) * 2013-03-29 2014-10-01 浙江大华技术股份有限公司 Image recognition method and device
CN106199544A (en) * 2016-06-24 2016-12-07 电子科技大学 The Recognition of Radar Target Using Range Profiles method of local tangent space alignment is differentiated based on core
CN108319983A (en) * 2018-01-31 2018-07-24 中山大学 A kind of nonlinear data dimension reduction method of local nonlinearity alignment

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101079104A (en) * 2007-06-14 2007-11-28 上海交通大学 Human face identification method based on information
CN104077594A (en) * 2013-03-29 2014-10-01 浙江大华技术股份有限公司 Image recognition method and device
CN104008383A (en) * 2014-06-24 2014-08-27 哈尔滨工业大学 Hyperspectral image characteristic extraction algorithm based on manifold learning linearization
CN106199544A (en) * 2016-06-24 2016-12-07 电子科技大学 The Recognition of Radar Target Using Range Profiles method of local tangent space alignment is differentiated based on core
CN108319983A (en) * 2018-01-31 2018-07-24 中山大学 A kind of nonlinear data dimension reduction method of local nonlinearity alignment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
谢晓华等: "一种半监督邻域自适应线性局部切空间排列的故障识别方法研究", 《机械强度》 *

Also Published As

Publication number Publication date
CN110222631B (en) 2022-03-15

Similar Documents

Publication Publication Date Title
Jia et al. Single-side domain generalization for face anti-spoofing
Liu et al. Sphereface: Deep hypersphere embedding for face recognition
CN108564129B (en) Trajectory data classification method based on generation countermeasure network
Xu et al. Stacked Sparse Autoencoder (SSAE) based framework for nuclei patch classification on breast cancer histopathology
CN110321967B (en) Image classification improvement method based on convolutional neural network
Lim et al. Shape classification using local and global features
CN102708364B (en) Cascade-classifier-based fingerprint image classification method
CN110222712B (en) Multi-special-item target detection algorithm based on deep learning
CN106250909A (en) A kind of based on the image classification method improving visual word bag model
Schinas et al. CERTH@ MediaEval 2012 Social Event Detection Task.
Hussain et al. A survey of traffic sign recognition systems based on convolutional neural networks
Zhong et al. Spottingnet: Learning the similarity of word images with convolutional neural network for word spotting in handwritten historical documents
Yi et al. A robust eye localization method for low quality face images
CN109446997A (en) Document code automatic identifying method
CN110969101A (en) Face detection and tracking method based on HOG and feature descriptor
Gu et al. Vision-based multi-scaled vehicle detection and distance relevant mix tracking for driver assistance system
CN106485202A (en) Unconfinement face identification system and method
Hu et al. Military object detection based on optimal gabor filtering and deep feature pyramid network
Li et al. Exploiting inductive bias in transformer for point cloud classification and segmentation
CN107146215A (en) A kind of conspicuousness detection method based on color histogram and convex closure
Zhao et al. A CCD based machine vision system for real-time text detection
CN110222631A (en) Based on deblocking it is parallel cut space arrangement SAR image target recognition method
Xiao et al. An improved siamese network model for handwritten signature verification
XiaoFan et al. Introduce GIoU into RFB net to optimize object detection bounding box
Yang et al. An improved YOLO leucocyte classification and recognition method

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220315