CN116206134A - Feature coding and recognition method and system for synthetic aperture radar image - Google Patents

Feature coding and recognition method and system for synthetic aperture radar image Download PDF

Info

Publication number
CN116206134A
CN116206134A CN202310497755.6A CN202310497755A CN116206134A CN 116206134 A CN116206134 A CN 116206134A CN 202310497755 A CN202310497755 A CN 202310497755A CN 116206134 A CN116206134 A CN 116206134A
Authority
CN
China
Prior art keywords
feature
point
image
feature point
points
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
Application number
CN202310497755.6A
Other languages
Chinese (zh)
Inventor
沈梦家
张军
张文金
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.)
719th Research Institute Of China State Shipbuilding Corp
Original Assignee
719th Research Institute Of China State Shipbuilding Corp
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 719th Research Institute Of China State Shipbuilding Corp filed Critical 719th Research Institute Of China State Shipbuilding Corp
Priority to CN202310497755.6A priority Critical patent/CN116206134A/en
Publication of CN116206134A publication Critical patent/CN116206134A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention relates to a feature coding and recognition method and a feature coding and recognition system for a synthetic aperture radar image, wherein the method comprises the following steps: performing feature point detection on the ship SAR image according to a SIFT algorithm of the feature descriptors and generating feature descriptors with uniform lengths; clustering the feature descriptors according to a K-means algorithm to obtain clustering center vectors, and aggregating the feature descriptors to obtain feature vectors with uniform length; and classifying the feature vectors with uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image. The method provided by the invention does not need a large number of image sets for training in the sea area environment, and the method can uniformly encode the features to a fixed length in the feature extraction process, is convenient for the subsequent model training and testing process, and has better interpretation and stability than the prior art.

Description

Feature coding and recognition method and system for synthetic aperture radar image
Technical Field
The invention belongs to the technical field of image recognition, and particularly relates to a feature coding and recognition method and system for a synthetic aperture radar image.
Background
The synthetic aperture radar (Synthetic Aperture Radar, SAR) has excellent performance of full-day working characteristics, no influence of climate conditions, capability of performing large-scale scene observation, certain surface penetration capacity, plays an irreplaceable role in continuous, real-time and long-term monitoring tasks of sea areas, and has been widely applied to various fields such as military national defense, scientific research and the like.
The main working principle of the SAR is that the mechanism of a synthetic aperture and pulse compression mode and a signal processing technology are fully utilized when the SAR is used, and meanwhile, the high-resolution azimuth and the high-resolution distance of a radar image are ensured by a small aperture antenna on the SAR, so that a high-resolution SAR image is obtained. The SAR image feature coding and recognition mainly aims at carrying out specific processing on an original SAR image, obtaining various types of features of the image from the original SAR image, and then applying the features to tasks such as target detection, image recognition, classification and the like, so that the information quantity and the workload of subsequent image interpretation are reduced, and meanwhile, the recognition efficiency and the task performance can be improved.
For the feature coding problem of the ship SAR image, the traditional machine learning method such as principal component analysis, wavelet transformation and the like are difficult to characterize the high-order features in the image, and the computation complexity is high, the feature extraction method based on deep learning such as convolutional neural network is mainstream in recent years, and the stable features of complex conditions can be learned under sufficient data, but the ship SAR image training data of the real scene can be obtained due to the complex sea area environment is less, and the requirement of the neural network training is difficult to meet.
Therefore, how to accurately extract features of the ship SAR image by using less training data is a technical problem to be solved by those skilled in the art.
Disclosure of Invention
In order to achieve the object of the present invention, there is provided a feature encoding and recognition method for a synthetic aperture radar image, comprising:
step S101: performing feature point detection on the ship SAR image according to a SIFT algorithm of the feature descriptors and generating feature descriptors with uniform lengths;
step S102: clustering the feature descriptors according to a K-means algorithm to obtain clustering center vectors, and aggregating the feature descriptors to obtain feature vectors with uniform length;
step S103: and classifying the feature vectors with uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
In some specific embodiments, in step S101, the feature point detection includes: detecting extreme points in the scale space, and setting
Figure SMS_1
Representation of image->
Figure SMS_2
A pixel point is marked as +.>
Figure SMS_3
Its image in the different scale space is expressed as +.>
Figure SMS_4
，/>
Figure SMS_5
，/>
wherein ,
Figure SMS_6
，
non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than a plurality of neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
In some specific embodiments, the feature point detection further includes: taking the characteristic point as the center of a circle
Figure SMS_7
Calculating a characteristic point +.for a circular region of radius>
Figure SMS_8
Wherein the gradient magnitude is calculated according to the following equation:
Figure SMS_9
，
the gradient argument is calculated according to the following formula:
Figure SMS_10
，
for confirming the feature point direction.
In some embodiments, the generating the uniform length feature descriptors includes: taking the characteristic points as the center, taking the direction of the characteristic points as the direction of a coordinate axis, selecting a 16X 16 square area around the characteristic points, equally dividing the square area into 4X 4 sub-areas, calculating the accumulated values of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4X 8 for each characteristic point.
In some embodiments, in step S102, the obtaining the feature vector with the uniform length includes:
step S1021: extracting feature points of a picture and representing the feature points as
Figure SMS_11
, wherein />
Figure SMS_12
Representing the number of feature points of the picture;
step S1022: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure SMS_13
, wherein />
Figure SMS_14
Indicate->
Figure SMS_15
Feature point set corresponding to each cluster center, +.>
Figure SMS_16
；
Step S1023: the first feature point in the feature point set
Figure SMS_17
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure SMS_18
The vectors are concatenated as the image features of the picture.
In order to achieve the above object, the present invention further provides a feature encoding and recognition system for a synthetic aperture radar image, including:
and the feature extraction module is used for: the method comprises the steps of performing feature point detection on a ship SAR image according to a SIFT algorithm of a feature descriptor and generating a feature descriptor with uniform length;
and (3) an aggregation module: the method comprises the steps of clustering feature descriptors according to a K-means algorithm to obtain a clustering center vector, and aggregating the feature descriptors to obtain a feature vector with uniform length;
and a classification module: the method is used for classifying the feature vectors with the uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
In some specific embodiments, in the feature extraction module, the feature point detection includes: detecting extreme points in the scale space, and setting
Figure SMS_19
Representation of image->
Figure SMS_20
A pixel point is marked as +.>
Figure SMS_21
Its image in the different scale space is expressed as +.>
Figure SMS_22
，/>
Figure SMS_23
, wherein ,/>
Figure SMS_24
，
Non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than a plurality of neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
In some specific embodiments, the feature point detection further includes: taking the characteristic point as the center of a circle
Figure SMS_25
Calculating a characteristic point +.for a circular region of radius>
Figure SMS_26
Wherein the gradient magnitude is calculated according to the following equation:
Figure SMS_27
，
the gradient argument is calculated according to the following formula:
Figure SMS_28
，
for confirming the feature point direction.
In some specific embodiments, in the feature extraction module, the generating the feature descriptors of uniform length includes: taking the characteristic points as the center, taking the direction of the characteristic points as the direction of a coordinate axis, selecting a square area with the length of 16 multiplied by 16 around the characteristic points, equally dividing the area into 4 multiplied by 4 sub-areas, calculating the accumulated value of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4 multiplied by 8 for each characteristic point.
In some embodiments, in the aggregation module, the obtaining the feature vector with the uniform length includes:
feature point extraction unit: extracting feature points of a picture and representing the feature points as
Figure SMS_29
, wherein />
Figure SMS_30
Representing the number of feature points of the picture;
classification unit: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure SMS_31
, wherein />
Figure SMS_32
Indicate->
Figure SMS_33
Feature point set corresponding to each cluster center, +.>
Figure SMS_34
；
A feature determination unit: the first feature point in the feature point set
Figure SMS_35
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure SMS_36
The vectors are concatenated as the image features of the picture.
The invention has the beneficial effects that:
according to the synthetic aperture radar image feature coding and identifying method and system, the ship SAR image is identified by performing SIFT feature extraction process, K-means-based feature clustering process and SVM multi-classification model image identification on the ship SAR image. Compared with the traditional feature extraction method and the convolutional neural network-based identification method, the method provided by the invention has better interpretation and stability, does not need a large number of image sets for training in the sea area environment, and can uniformly encode the features for a fixed length in the feature extraction process, thereby being convenient for the subsequent model training and testing process.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of some embodiments of a feature encoding and recognition method for a synthetic aperture radar image according to the present invention;
FIG. 2 is a schematic flow chart of obtaining a uniform length feature vector in a feature encoding and recognition method for a synthetic aperture radar image according to the present invention;
FIG. 3 is a schematic diagram of some embodiments of a feature encoding and recognition system for synthetic aperture radar images according to the present invention;
FIG. 4 is a schematic diagram of a structure of an aggregation module in a system for feature encoding and recognition of a synthetic aperture radar image according to the present invention to obtain feature vectors of uniform length;
FIG. 5 is a schematic diagram of a feature encoding and recognition method and system for synthetic aperture radar images according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments.
Examples of the embodiments are illustrated in the accompanying drawings, wherein like or similar symbols indicate like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative and intended to explain the present invention and should not be construed as limiting the invention.
Referring to fig. 1 and 5, a method for coding and identifying image features of a synthetic aperture radar includes:
step S101: and carrying out feature point detection on the ship SAR image according to a SIFT algorithm of the feature descriptors and generating the feature descriptors with uniform length.
In some embodiments of the present invention, feature point detection includes: detecting extreme points in the scale space, and setting
Figure SMS_37
Representation of image->
Figure SMS_38
A pixel point is marked as +.>
Figure SMS_39
Its image in a different scale space is represented as
Figure SMS_40
，/>
Figure SMS_41
，
wherein ,
Figure SMS_42
，
non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than 26 neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
In some embodiments of the present invention, the feature point detection further includes: at the center of the characteristic point
Figure SMS_43
Calculating a characteristic point +.for a circular region of radius>
Figure SMS_44
Wherein the gradient magnitude is calculated according to the following equation:
Figure SMS_45
，
the gradient argument is calculated according to the following formula:
Figure SMS_46
，
for confirming the feature point direction.
Specifically, 360 ° is equally divided into 8 directions, namely 45 ° in each direction, and gradient vectors of all pixel points in a circular region obtained by calculation of the two formulas are mapped to the 8 directions, so as to obtain a gradient vector direction histogram. The direction of the peak value of the histogram is selected as the main direction of the feature point, and if 80% of other peak values of the main direction peak value exist in the histogram, the direction is reserved and is set as the auxiliary direction of the feature point.
In some embodiments of the present invention, generating a uniform length of feature descriptors after feature point detection includes: taking the characteristic point as the center, taking the direction of the characteristic point as the direction of a coordinate axis, selecting a 16X 16 square area around the characteristic point, equally dividing each square area into 4X 4 sub-areas, calculating the accumulated value of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4X 8 for each characteristic point.
Step S102: clustering the feature descriptors according to a K-means algorithm to obtain clustering center vectors, and aggregating the feature descriptors to obtain feature vectors with uniform length.
In some embodiments of the invention, feature point cluster centers are generated based on a K-means algorithm according to a ship SAR image, and feature point vectors of the image are aggregated according to the feature centers to obtain image feature representations with uniform lengths.
Specifically, the specific steps for generating the characteristic clustering center based on the K-means algorithm are as follows:
firstly, extracting SIFT features of all SAR pictures of a training set to obtain SIFT feature vector sets of all images
Figure SMS_47
, wherein />
Figure SMS_48
Representing the number of SIFT feature points of all images; from the feature point set->
Figure SMS_49
Is selected randomly
Figure SMS_50
Personal feature vector->
Figure SMS_51
As an initial cluster center; calculate the set +.>
Figure SMS_52
Other feature points of the vector are vector to the clustering center>
Figure SMS_53
And categorizes it into the nearest cluster center.
Updating the clustering center according to the clustering result
Figure SMS_54
When clustering center->
Figure SMS_55
When no change occurs, the clustering is finished; otherwise, the clustering center is updated by the reclustering iteration.
In order to obtain a uniform-length image feature vector for classifying ship SAR images, aggregating SIFT feature descriptors of images of training and testing models to obtain the uniform-length feature vector, referring to FIG. 2, the method comprises the following steps:
step S1021: extracting feature points of a picture and representing the feature points as
Figure SMS_56
, wherein />
Figure SMS_57
Representing the number of feature points of the picture;
step S1022: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure SMS_58
, wherein />
Figure SMS_59
Indicate->
Figure SMS_60
Feature point set corresponding to each cluster center, +.>
Figure SMS_61
；
Step S1023: the first feature point in the feature point set
Figure SMS_62
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure SMS_63
The vectors are concatenated as the image features of the picture.
Step S103: and classifying the feature vectors with uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
In some embodiments of the present invention, the SVM (Support Vector Machine ) can effectively solve the problems of nonlinear high-dimensional pattern recognition and the like. Since SVM is a classifier, marine SAR image recognition is a multi-classification problem. Therefore, the invention adopts multi-classification SVM algorithm, and the method is suitable for
Figure SMS_64
Class question using->
Figure SMS_65
A classifier, and->
Figure SMS_66
Satisfy the following requirements
Figure SMS_67
. To improve the generalization ability and convergence rate of the model, the model uses a broader Radial Basis Function (RBF) kernel function that can approximate any nonlinear function.
To achieve the above object, referring to fig. 3 and 5, the present invention further proposes a feature encoding and recognition system for a synthetic aperture radar image, including:
the feature extraction module 100: and the SIFT algorithm is used for detecting characteristic points of the ship SAR image according to the characteristic descriptors and generating the characteristic descriptors with uniform length.
In some embodiments of the present invention, in the feature extraction module 100, feature point detection includes: detecting extreme points in the scale space, and setting
Figure SMS_68
Representation of image->
Figure SMS_69
A pixel point is marked as +.>
Figure SMS_70
Its image in the different scale space is expressed as +.>
Figure SMS_71
，/>
Figure SMS_72
，
wherein ,
Figure SMS_73
，
non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than 26 neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
In some embodiments of the present invention, the feature point detection further includes: at the center of the characteristic point
Figure SMS_74
Calculating a characteristic point +.for a circular region of radius>
Figure SMS_75
Wherein the gradient magnitude is calculated according to the following equation:
Figure SMS_76
，
the gradient argument is calculated according to the following formula:
Figure SMS_77
，
for confirming the feature point direction.
Specifically, 360 ° is equally divided into 8 directions, namely 45 ° in each direction, and gradient vectors of all pixel points in a circular region obtained by calculation of the two formulas are mapped to the 8 directions, so as to obtain a gradient vector direction histogram. The direction of the peak value of the histogram is selected as the main direction of the feature point, if 80% of other peak values of the main direction peak value exist in the histogram, the direction is reserved, and the direction is set as the auxiliary direction of the feature point for confirming the direction of the feature point.
In some embodiments of the present invention, in the feature extraction module 100, generating the feature descriptors of uniform length includes: taking the characteristic point as the center, taking the direction of the characteristic point as the direction of a coordinate axis, selecting a 16X 16 square area around the characteristic point, equally dividing each area into 4X 4 sub-areas, calculating the accumulated value of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4X 8 for each characteristic point.
Aggregation module 200: the method is used for clustering the feature descriptors according to the K-means algorithm to obtain clustering center vectors, and aggregating the feature descriptors to obtain feature vectors with uniform length.
In some embodiments of the invention, feature point cluster centers are generated based on a K-means algorithm according to a ship SAR image, and feature point vectors of the image are aggregated according to the feature centers to obtain image feature representations with uniform lengths.
Specifically, in the aggregation module 200, the specific steps of generating the feature cluster center based on the K-means algorithm are as follows:
firstly, extracting SIFT features of all SAR pictures of a training set to obtain SIFT feature vector sets of all images
Figure SMS_78
, wherein />
Figure SMS_79
Representing the number of SIFT feature points of all images; from the feature point set->
Figure SMS_80
Is selected randomly
Figure SMS_81
Personal feature vector->
Figure SMS_82
As an initial cluster center; calculate the set +.>
Figure SMS_83
Other feature points of the vector are vector to the clustering center>
Figure SMS_84
And categorizes it into the nearest cluster center.
Updating the clustering center according to the clustering result
Figure SMS_85
When clustering center->
Figure SMS_86
When no change occurs, the clustering is finished; otherwise, the clustering center is updated by the reclustering iteration.
In order to obtain a uniform-length image feature vector for classifying ship SAR images, aggregating SIFT feature descriptors of images of training and testing models to obtain the uniform-length feature vector, referring to FIG. 4, the method comprises the following steps:
the feature point extraction unit 210: extracting feature points of a picture and representing the feature points as
Figure SMS_87
, wherein
Figure SMS_88
Representing the number of feature points of the picture;
the classifying unit 220: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure SMS_89
, wherein />
Figure SMS_90
Indicate->
Figure SMS_91
Feature point set corresponding to each cluster center, +.>
Figure SMS_92
；
The feature determination unit 230: the first feature point in the feature point set
Figure SMS_93
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure SMS_94
The vectors are concatenated as the image features of the picture.
Classification module 300: the method is used for classifying the feature vectors with uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
In some embodiments of the invention, a multi-classification SVM algorithm is employed to perform the following steps
Figure SMS_95
Class question using->
Figure SMS_96
A classifier, and
Figure SMS_97
satisfy->
Figure SMS_98
. To improve the generalization ability and convergence rate of the model, the model uses a broader Radial Basis Function (RBF) kernel function that can approximate any nonlinear function.
Compared with the traditional feature extraction method and the convolutional neural network-based identification method, the method provided by the invention has better interpretation and stability, does not need a large number of image sets for training in the sea area environment, and can uniformly encode the features for a fixed length in the feature extraction process, thereby being convenient for the subsequent model training and testing process.
In the description of the present specification, reference to the terms "one embodiment," "some embodiments," "examples," "particular examples," "one particular embodiment," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application, and are not limiting thereof; although the present application has been described in detail with reference to the foregoing embodiments, one of ordinary skill in the art will appreciate that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not drive the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (10)

1. A feature encoding and recognition method for a synthetic aperture radar image, comprising:
step S101: performing feature point detection on the ship SAR image according to a SIFT algorithm of the feature descriptors, and generating feature descriptors with uniform length;
step S102: clustering the feature descriptors according to a K-means algorithm to obtain clustering center vectors, and aggregating the feature descriptors to obtain feature vectors with uniform length;
step S103: and classifying the feature vectors with uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
2. The feature encoding and recognition method for a synthetic aperture radar image according to claim 1, wherein in step S101, the feature point detection includes: detecting extreme points in the scale space, and setting
Figure QLYQS_1
Representation of image->
Figure QLYQS_2
A pixel point is marked as +.>
Figure QLYQS_3
Its image in the different scale space is expressed as +.>
Figure QLYQS_4
，
Figure QLYQS_5
，
wherein ,
Figure QLYQS_6
，
non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than a plurality of neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
3. The feature encoding and recognition method for a synthetic aperture radar image according to claim 2, wherein in step S101, the feature point detection further includes: taking the characteristic point as the center of a circle
Figure QLYQS_7
Calculating a characteristic point +.for a circular region of radius>
Figure QLYQS_8
Wherein the gradient magnitude is calculated according to the following equation:
Figure QLYQS_9
，
the gradient argument is calculated according to the following formula:
Figure QLYQS_10
，
for confirming the feature point direction.
4. The method for feature encoding and recognition of a synthetic aperture radar image according to claim 1, wherein in step S101, the generating a uniform length feature descriptor includes: taking the characteristic points as the center, taking the direction of the characteristic points as the direction of a coordinate axis, selecting a 16X 16 square area around the characteristic points, equally dividing the square area into 4X 4 sub-areas, calculating the accumulated values of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4X 8 for each characteristic point.
5. The feature encoding and recognition method for a synthetic aperture radar image according to claim 1, wherein in step S102, the acquiring feature vectors of uniform length includes:
step S1021: extracting feature points of a picture and representing the feature points as
Figure QLYQS_11
, wherein />
Figure QLYQS_12
Representing the number of feature points of the picture;
step S1022: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure QLYQS_13
, wherein />
Figure QLYQS_14
Indicate->
Figure QLYQS_15
Feature point set corresponding to each cluster center, +.>
Figure QLYQS_16
；
Step S1023: the first feature point in the feature point set
Figure QLYQS_17
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure QLYQS_18
The vectors are concatenated as an image of the pictureFeatures.
6. A feature encoding and recognition system for a synthetic aperture radar image, comprising:
and the feature extraction module is used for: the method comprises the steps of performing feature point detection on a ship SAR image according to a SIFT algorithm of a feature descriptor and generating a feature descriptor with uniform length;
and (3) an aggregation module: the method comprises the steps of clustering feature descriptors according to a K-means algorithm to obtain a clustering center vector, and aggregating the feature descriptors to obtain a feature vector with uniform length;
and a classification module: the method is used for classifying the feature vectors with the uniform length by using a multi-classification SVM algorithm and using a radial basis function to obtain a classification recognition result of the image.
7. The feature encoding and recognition system for a synthetic aperture radar image of claim 6, wherein in the feature extraction module, the feature point detection comprises: detecting extreme points in the scale space, and setting
Figure QLYQS_19
Representation of image->
Figure QLYQS_20
A pixel point is marked as +.>
Figure QLYQS_21
Its image in the different scale space is expressed as +.>
Figure QLYQS_22
，
Figure QLYQS_23
，
wherein ,
Figure QLYQS_24
，
non-maximum suppression is carried out in a 3X 3 pyramid stereo neighborhood of the scale space, when a certain pixel point is larger or smaller than a plurality of neighborhood values around the previous scale, the next scale and the current scale, the pixel point is determined to be an extreme point, and a Hessian matrix is adopted to confirm the characteristic point.
8. The feature encoding and recognition system for a synthetic aperture radar image of claim 7, wherein the feature points are centered at
Figure QLYQS_25
Calculating a characteristic point +.for a circular region of radius>
Figure QLYQS_26
Wherein the gradient magnitude is calculated according to the following equation:
Figure QLYQS_27
，
the gradient argument is calculated according to the following formula:
Figure QLYQS_28
，
for confirming the feature point direction.
9. The feature encoding and recognition system for a synthetic aperture radar image of claim 6, wherein in the feature extraction module, the generating a uniform length feature descriptor comprises: taking the characteristic points as the center, taking the direction of the characteristic points as the direction of a coordinate axis, selecting a 16X 16 square area around the characteristic points, equally dividing the square area into 4X 4 sub-areas, calculating the accumulated values of 8 directions on each sub-area, and finally generating a SIFT characteristic description vector with the length of 4X 8 for each characteristic point.
10. The feature encoding and recognition system for a synthetic aperture radar image of claim 6, wherein in the aggregation module, the obtaining feature vectors of uniform length comprises:
feature point extraction unit: extracting feature points of a picture and representing the feature points as
Figure QLYQS_29
, wherein />
Figure QLYQS_30
Representing the number of feature points of the picture;
classification unit: calculating the distance between each feature point and the clustering center, classifying each feature point to the nearest clustering center, and obtaining a feature point set and representing the feature point set as
Figure QLYQS_31
, wherein />
Figure QLYQS_32
Indicate->
Figure QLYQS_33
Feature point set corresponding to each cluster center, +.>
Figure QLYQS_34
；
A feature determination unit: the first feature point in the feature point set
Figure QLYQS_35
Feature point vector cumulative averaging in each cluster center, will get +.>
Figure QLYQS_36
The vectors are concatenated as the image features of the picture. />
CN202310497755.6A 2023-05-06 2023-05-06 Feature coding and recognition method and system for synthetic aperture radar image Pending CN116206134A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310497755.6A CN116206134A (en) 2023-05-06 2023-05-06 Feature coding and recognition method and system for synthetic aperture radar image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310497755.6A CN116206134A (en) 2023-05-06 2023-05-06 Feature coding and recognition method and system for synthetic aperture radar image

Publications (1)

Publication Number Publication Date
CN116206134A true CN116206134A (en) 2023-06-02

Family

ID=86513379

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310497755.6A Pending CN116206134A (en) 2023-05-06 2023-05-06 Feature coding and recognition method and system for synthetic aperture radar image

Country Status (1)

Country Link
CN (1) CN116206134A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105335757A (en) * 2015-11-03 2016-02-17 电子科技大学 Model identification method based on local characteristic aggregation descriptor
CN107066559A (en) * 2017-03-30 2017-08-18 天津大学 A kind of method for searching three-dimension model based on deep learning
CN108491430A (en) * 2018-02-09 2018-09-04 北京邮电大学 It is a kind of based on the unsupervised Hash search method clustered to characteristic direction

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105335757A (en) * 2015-11-03 2016-02-17 电子科技大学 Model identification method based on local characteristic aggregation descriptor
CN107066559A (en) * 2017-03-30 2017-08-18 天津大学 A kind of method for searching three-dimension model based on deep learning
CN108491430A (en) * 2018-02-09 2018-09-04 北京邮电大学 It is a kind of based on the unsupervised Hash search method clustered to characteristic direction

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
应厚泽: ""基于机器学习的SAR图像分类与识别硏究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 02, pages 20 - 42 *

Similar Documents

Publication Publication Date Title
Lin et al. Convolution in the cloud: Learning deformable kernels in 3d graph convolution networks for point cloud analysis
Uçar et al. A new facial expression recognition based on curvelet transform and online sequential extreme learning machine initialized with spherical clustering
CN105574534B (en) Conspicuousness object detection method based on sparse subspace clustering and low-rank representation
CN109993748B (en) Three-dimensional grid object segmentation method based on point cloud processing network
Xu et al. Scale-space texture description on sift-like textons
CN110334762B (en) Feature matching method based on quad tree combined with ORB and SIFT
Chen et al. Local patch network with global attention for infrared small target detection
AU2020101435A4 (en) A panoramic vision system based on the uav platform
Jia et al. Flexible Gabor-based superpixel-level unsupervised LDA for hyperspectral image classification
CN111460881A (en) Traffic sign countermeasure sample detection method and classification device based on neighbor discrimination
CN103093226B (en) A kind of building method of the RATMIC descriptor for characteristics of image process
CN110633640A (en) Method for identifying complex scene by optimizing PointNet
Van Zyl et al. Unique animal identification using deep transfer learning for data fusion in siamese networks
Alsanad et al. Real-time fuel truck detection algorithm based on deep convolutional neural network
Jiang et al. Abnormal behavior detection using streak flow acceleration
Liu et al. SAR image target detection in complex environments based on improved visual attention algorithm
Li et al. 3D model classification based on nonparametric discriminant analysis with kernels
Zhang et al. Fast visual tracking with lightweight Siamese network and template-guided learning
Gu et al. Enhanced local and global learning for rotation-invariant point cloud representation
CN116206134A (en) Feature coding and recognition method and system for synthetic aperture radar image
Sima et al. Action recognition algorithm based on skeletal joint data and adaptive time pyramid
CN115272696A (en) Point cloud semantic segmentation method based on self-adaptive convolution and local geometric information
Kiran et al. Human posture classification using hybrid Particle Swarm Optimization
Abouqora et al. A hybrid CNN-CRF inference models for 3D mesh segmentation
An et al. Research review of object detection algorithms in vehicle detection

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20230602