CN112329792A - Hyperspectral image target feature extraction method based on spectrum angle - Google Patents

Hyperspectral image target feature extraction method based on spectrum angle Download PDF

Info

Publication number
CN112329792A
CN112329792A CN202011188705.2A CN202011188705A CN112329792A CN 112329792 A CN112329792 A CN 112329792A CN 202011188705 A CN202011188705 A CN 202011188705A CN 112329792 A CN112329792 A CN 112329792A
Authority
CN
China
Prior art keywords
spectrum
target
projection
mean
hyperspectral 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.)
Granted
Application number
CN202011188705.2A
Other languages
Chinese (zh)
Other versions
CN112329792B (en
Inventor
孙康
陈金勇
李方方
王敏
帅通
王士成
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.)
CETC 54 Research Institute
Original Assignee
CETC 54 Research Institute
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 CETC 54 Research Institute filed Critical CETC 54 Research Institute
Priority to CN202011188705.2A priority Critical patent/CN112329792B/en
Publication of CN112329792A publication Critical patent/CN112329792A/en
Application granted granted Critical
Publication of CN112329792B publication Critical patent/CN112329792B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention discloses a hyperspectral image target feature extraction method based on a spectrum angle, and belongs to the technical field of remote sensing image processing. Which comprises the following steps: calculating a covariance matrix and a mean spectrum of the hyperspectral image; calculating a projection mean spectrum and a projection target spectrum; sequentially removing each element from the projection mean spectrum and the projection target spectrum to form a new projection mean spectrum and a new projection target spectrum, and calculating the spectrum angle of the new projection mean spectrum and the new projection target spectrum; searching a projection mean spectrum and a projection target spectrum with the maximum spectrum angle, and replacing the original projection mean spectrum and the original projection target spectrum with the projection mean spectrum and the projection target spectrum; and updating the number of the current residual features until the number of the features reaches the requirement. According to the hyperspectral image target extraction method, the mean spectrum and the target spectrum are projected through introduction of the covariance matrix, and the background is suppressed while the target information is highlighted, so that the wave band which is maximally distinguished between the target and the background can be obtained, and the hyperspectral image target extraction precision is improved.

Description

Hyperspectral image target feature extraction method based on spectrum angle
Technical Field
The invention belongs to the technical field of remote sensing image processing, and particularly relates to a hyperspectral image target feature extraction method based on a spectrum angle.
Background
The hyperspectral remote sensing technology can realize continuous spectrum sampling of ground objects from visible light to infrared, and extremely abundant spectrum information is formed. Compared with the common remote sensing image, the hyperspectral image has the important advantages that abundant spectrum dimension information is added, and the differences of physical structure characteristics and chemical composition characteristics inside a target can be fully reflected, so that the spectral dimensions of ground objects can be finely distinguished. The fine spectral data of the hyperspectral image provides a data basis for target detection and the like.
In the target detection application of the hyperspectral image, an important technology is a target spectral feature extraction technology, which is called a target feature extraction technology for short. The method can select more than ten or less characteristic wave bands from hundreds of hyperspectral image wave bands, and the characteristic wave bands are important characteristics for distinguishing the target from other ground objects. Through research on the characteristic wave band spectral characteristics, important support can be provided for detection and identification of targets.
Most of the current target feature extraction research is conducted on the target spectrum, the background of the target is not considered, and the information of other ground objects in the image is not fully utilized, so that the feature extraction effect is not ideal.
Disclosure of Invention
In view of the above, the present invention provides a method for extracting target features of a hyperspectral image based on a spectrum angle, which simultaneously considers the feature extraction of a target and an image background, does not need complex models and parameters, is easy to implement, can be automatically executed, and is suitable for various targets and backgrounds.
In order to achieve the purpose, the invention adopts the technical scheme that:
a hyperspectral image target feature extraction method based on spectrum angles is used for extracting target features of a certain target from a hyperspectral image with the total wave band number L; the method comprises the following steps:
step 1, calculating a covariance matrix K and a mean spectrum of a hyperspectral image
Figure BDA0002752164450000011
Step 2, reading a target spectrum d of a target to be processed and the number n of features to be extracted aiming at the target, wherein n is less than L;
step 3, utilizing covariance matrix K and mean value spectrum
Figure BDA0002752164450000012
And a target spectrum d, calculating a projection mean spectrum
Figure BDA0002752164450000013
And projecting the target spectrum
Figure BDA0002752164450000014
Projection mean spectrum
Figure BDA0002752164450000015
And projecting the target spectrum
Figure BDA0002752164450000016
The method comprises m elements, namely m characteristics, wherein the initial value of m is L;
step 4, from the projection mean spectrum
Figure BDA0002752164450000017
And projecting the target spectrum
Figure BDA0002752164450000018
Removing the ith characteristic, i is more than or equal to 1 and less than or equal to m, and forming a new projection mean value spectrum
Figure BDA0002752164450000019
And projecting the target spectrum
Figure BDA00027521644500000110
Step 5, calculating corresponding projection mean value spectrum for each value of i
Figure BDA0002752164450000021
And projecting the target spectrum
Figure BDA0002752164450000022
The spectral angle of (d);
step 6, searching the projection mean spectrum with the maximum spectrum angle
Figure BDA0002752164450000023
And projecting the target spectrum
Figure BDA0002752164450000024
Using it as new projection mean spectrum
Figure BDA0002752164450000025
And projecting the target spectrum
Figure BDA0002752164450000026
Step 7, subtracting 1 from m, if m is more than n, repeating the step 4 to the step 7, otherwise, projecting the target spectrum at the moment
Figure BDA0002752164450000027
Namely the target characteristic extraction result.
Further, in step 1, the covariance matrix K is:
Figure BDA0002752164450000028
wherein k isijThe inner product of the ith wave band and the jth wave band of the hyperspectral image is obtained;
mean spectrum
Figure BDA0002752164450000029
Comprises the following steps:
Figure BDA00027521644500000210
wherein x ispqThe pixel spectra of the p-th row and the q-th column of the hyperspectral image are obtained, N is the total row number of the hyperspectral image, and M is the total column number of the hyperspectral image.
Further, the specific manner of step 3 is as follows:
step 3a, calculating an inverse matrix K of the covariance matrix K-1；
Step 3b, calculating a projection mean spectrum
Figure BDA00027521644500000211
Figure BDA00027521644500000212
Wherein the content of the first and second substances,
Figure BDA00027521644500000213
is composed of
Figure BDA00027521644500000214
Transposing;
step 3c, calculating the spectrum of the projection target
Figure BDA00027521644500000215
Figure BDA00027521644500000216
Wherein d isTIs the transpose of d.
Further, in the step 5, a mean spectrum is projected
Figure BDA00027521644500000217
And projecting the target spectrum
Figure BDA00027521644500000218
Angle of spectrum of
Figure BDA00027521644500000219
Comprises the following steps:
Figure BDA0002752164450000031
compared with the prior art, the invention has the following advantages:
(1) the method projects the mean spectrum and the target spectrum by introducing the covariance matrix, and suppresses the background while highlighting target information, so that the wave band which is maximally distinguished between the target and the background can be obtained.
(2) The method uses the angle information of the target spectrum and the mean spectrum as discrimination measurement, does not need a model and parameters of complexity, and is easy to realize.
(3) The method is suitable for extracting any target information of the hyperspectral image.
In a word, the mean spectrum and the target spectrum are projected by introducing the covariance matrix, and the background is suppressed while the target information is highlighted, so that the wave band which is maximally distinguished between the target and the background can be obtained, and the precision of extracting the hyperspectral image target is improved.
Drawings
Fig. 1 is an overall flowchart of a target feature extraction method in an embodiment of the present invention.
FIG. 2 is a graph comparing recall of test data in examples of the present invention.
FIG. 3 is a comparison of false alarm rates of test data in an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, a hyperspectral image target feature extraction method based on a spectrum angle is used for extracting a target feature from a hyperspectral image with a total number of bands L; the method comprises the following steps:
step 1, calculating covariance matrix K and mean of hyperspectral imagesValue spectrum
Figure BDA0002752164450000032
The covariance matrix K is:
Figure BDA0002752164450000033
wherein k isijThe inner product of the ith wave band and the jth wave band of the hyperspectral image is obtained;
mean spectrum
Figure BDA0002752164450000034
Comprises the following steps:
Figure BDA0002752164450000035
wherein x ispqThe pixel spectra of the p-th row and the q-th column of the hyperspectral image are obtained, N is the total row number of the hyperspectral image, and M is the total column number of the hyperspectral image.
Step 2, reading a target spectrum d of a target to be processed and the number n of features to be extracted aiming at the target, wherein n is less than L;
step 3, utilizing covariance matrix K and mean value spectrum
Figure BDA0002752164450000041
And a target spectrum d, calculating a projection mean spectrum
Figure BDA0002752164450000042
And projecting the target spectrum
Figure BDA0002752164450000043
Projection mean spectrum
Figure BDA0002752164450000044
And projecting the target spectrum
Figure BDA0002752164450000045
All contain mElements, namely m features, wherein the initial value of m is L; the concrete mode is as follows:
step 3a, calculating an inverse matrix K of the covariance matrix K-1；
Step 3b, calculating a projection mean spectrum
Figure BDA0002752164450000046
Figure BDA0002752164450000047
Wherein the content of the first and second substances,
Figure BDA0002752164450000048
is composed of
Figure BDA0002752164450000049
Transposing;
step 3c, calculating the spectrum of the projection target
Figure BDA00027521644500000410
Figure BDA00027521644500000411
Wherein d isTIs the transpose of d.
Step 4, from the projection mean spectrum
Figure BDA00027521644500000412
And projecting the target spectrum
Figure BDA00027521644500000413
Removing the ith characteristic, i is more than or equal to 1 and less than or equal to m, and forming a new projection mean value spectrum
Figure BDA00027521644500000414
And projecting the target spectrum
Figure BDA00027521644500000415
Step 5, calculating corresponding projection mean value spectrum for each value of i
Figure BDA00027521644500000416
And projecting the target spectrum
Figure BDA00027521644500000417
Angle of spectrum of
Figure BDA00027521644500000418
Figure BDA00027521644500000419
Step 6, searching the projection mean spectrum with the maximum spectrum angle
Figure BDA00027521644500000420
And projecting the target spectrum
Figure BDA00027521644500000421
Using it as new projection mean spectrum
Figure BDA00027521644500000422
And projecting the target spectrum
Figure BDA00027521644500000423
Step 7, subtracting 1 from m, if m is more than n, repeating the step 4 to the step 7, otherwise, projecting the target spectrum at the moment
Figure BDA00027521644500000424
Namely the target characteristic extraction result.
It is well known to those skilled in the art that the target features are actually against the background, and the characteristic bands of the target features are often different for different backgrounds. In view of the above, the present invention takes background information of the target into account, thereby improving the accuracy of feature extraction. The method has the main idea that the projection mean spectrum and the target spectrum of the image are constructed by calculating the covariance matrix of the image, the background information of the image is combined while the target information is highlighted, and then the characteristic combination with the maximum degree of separation between the target and the background is selected in an iterative mode by utilizing a maximum angle method.
The principle of the invention is as follows: the image mean spectrum and the target spectrum are projected to an inverse matrix of the covariance matrix by calculating the covariance matrix of the image, and then non-characteristic wave bands are removed one by one in an iterative mode until the number of the remaining characteristics meets the requirement. And when the target features are considered, the background information is also taken into consideration of feature extraction, and the extraction of the target features is completed.
The effects of the present invention can be further illustrated by the following tests:
1. test conditions.
The computer is configured with Intel Core i7-3770 CPU 3.4Ghz, 4GB memory, and the software environment is Matlab R2013 and ENVI 5.1 platform.
2. Test methods.
The test carries out comparative analysis on the three conditions, namely, feature extraction is not carried out, MVPCA feature extraction and the feature extraction of the method of the invention are not carried out, the three processing results are respectively used for target detection, and the target detection recall rate and the false alarm rate under the three conditions are compared to verify the effectiveness of the method.
3. Test contents and results.
The experimental selection was made of hyperspectral data sandiog acquired by the airborne visible/infrared imaging spectrometer (AVIRIS) of the united states aeronautics and space administration (NASA) in 2002 at the military airport of san diego, usa. The sandiog data is a preprocessed reflectivity image with a size of 400 pixels × 224 bands, wherein 189 active bands are provided, and 1-6, 33-35, 97, 107-. The target spectrum selects the spectrum of the airplane, and the number of feature extractions is set to 15.
The tests were carried out using the CEM method for object detection using the raw features (189, respectively), 15 features extracted from MVCPA, and 15 features extracted from the present invention. In order to verify the effectiveness of the method of the present invention, the target detection results of the three methods are quantitatively analyzed, and the corresponding recall rate and false alarm rate are calculated, respectively, and the results are shown in fig. 2 and fig. 3, respectively.
The test result shows that the features extracted by the method have higher recall rate which reaches 89.96 percent and are superior to the original features and the features extracted by MVPCA, and simultaneously have lower false alarm rate which is 15.59 percent. The result shows that the method can effectively select the characteristics of the hyperspectral image target, and can realize target detection with higher performance based on the target characteristics extracted by the method.

Claims (4)

1. A hyperspectral image target feature extraction method based on a spectrum angle is characterized by being used for extracting a target feature from a hyperspectral image with the total wave band number L; the method comprises the following steps:
step 1, calculating a covariance matrix K and a mean spectrum of a hyperspectral image
Figure FDA0002752164440000011
Step 2, reading a target spectrum d of a target to be processed and the number n of features to be extracted aiming at the target, wherein n is less than L;
step 3, utilizing covariance matrix K and mean value spectrum
Figure FDA0002752164440000012
And a target spectrum d, calculating a projection mean spectrum
Figure FDA0002752164440000013
And projecting the target spectrum
Figure FDA0002752164440000014
Projection mean spectrum
Figure FDA0002752164440000015
And projecting the target spectrum
Figure FDA0002752164440000016
The method comprises m elements, namely m characteristics, wherein the initial value of m is L;
step 4, from the projection mean spectrum
Figure FDA0002752164440000017
And projecting the target spectrum
Figure FDA0002752164440000018
Removing the ith characteristic, i is more than or equal to 1 and less than or equal to m, and forming a new projection mean value spectrum
Figure FDA0002752164440000019
And projecting the target spectrum
Figure FDA00027521644400000110
Step 5, calculating corresponding projection mean value spectrum for each value of i
Figure FDA00027521644400000111
And projecting the target spectrum
Figure FDA00027521644400000112
The spectral angle of (d);
step 6, searching the projection mean spectrum with the maximum spectrum angle
Figure FDA00027521644400000113
And projecting the target spectrum
Figure FDA00027521644400000114
Using it as new projection mean spectrum
Figure FDA00027521644400000115
And projecting the target spectrum
Figure FDA00027521644400000116
Step 7, subtracting 1 from m, if m is more than n, repeating the step 4 to the step 7, otherwise, projecting the target spectrum at the moment
Figure FDA00027521644400000117
Namely the target characteristic extraction result.
2. The method for extracting the target features of the hyperspectral images based on the spectral angles according to claim 1, wherein in the step 1, the covariance matrix K is:
Figure FDA00027521644400000118
wherein k isijThe inner product of the ith wave band and the jth wave band of the hyperspectral image is obtained;
mean spectrum
Figure FDA00027521644400000119
Comprises the following steps:
Figure FDA00027521644400000120
wherein x ispqThe pixel spectra of the p-th row and the q-th column of the hyperspectral image are obtained, N is the total row number of the hyperspectral image, and M is the total column number of the hyperspectral image.
3. The method for extracting the target features of the hyperspectral images based on the spectrum angles according to claim 1, wherein the specific manner of the step 3 is as follows:
step 3a, calculating an inverse matrix K of the covariance matrix K-1；
Step 3b, calculating a projection mean spectrum
Figure FDA00027521644400000121
Figure FDA0002752164440000021
Wherein the content of the first and second substances,
Figure FDA0002752164440000022
is composed of
Figure FDA0002752164440000023
Transposing;
step 3c, calculating the spectrum of the projection target
Figure FDA0002752164440000024
Figure FDA0002752164440000025
Wherein d isTIs the transpose of d.
4. The method for extracting the target features of the hyperspectral images based on the spectral angles as claimed in claim 1, wherein in the step 5, a projection mean spectrum is used
Figure FDA0002752164440000026
And projecting the target spectrum
Figure FDA0002752164440000027
Angle of spectrum of
Figure FDA0002752164440000028
Comprises the following steps:
Figure FDA0002752164440000029
CN202011188705.2A 2020-10-30 2020-10-30 Hyperspectral image target feature extraction method based on spectrum angle Active CN112329792B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011188705.2A CN112329792B (en) 2020-10-30 2020-10-30 Hyperspectral image target feature extraction method based on spectrum angle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011188705.2A CN112329792B (en) 2020-10-30 2020-10-30 Hyperspectral image target feature extraction method based on spectrum angle

Publications (2)

Publication Number Publication Date
CN112329792A true CN112329792A (en) 2021-02-05
CN112329792B CN112329792B (en) 2022-12-09

Family

ID=74296734

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011188705.2A Active CN112329792B (en) 2020-10-30 2020-10-30 Hyperspectral image target feature extraction method based on spectrum angle

Country Status (1)

Country Link
CN (1) CN112329792B (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1546959A (en) * 2003-12-12 2004-11-17 上海交通大学 Nonlinear spectrum similarity measurement method
CN104463808A (en) * 2014-12-25 2015-03-25 武汉大学 High-spectral data noise reduction method and system based on spatial correlation
CN106023065A (en) * 2016-05-13 2016-10-12 中国矿业大学 Tensor hyperspectral image spectrum-space dimensionality reduction method based on deep convolutional neural network
CN108596077A (en) * 2018-04-20 2018-09-28 南阳理工学院 The adaptive spectrum focus wave section selection method of high spectrum image
WO2019194693A1 (en) * 2018-04-05 2019-10-10 Inesc Tec - Instituto De Engenharia De Sistemas E Computadores, Tecnologia E Ciência Spectrophotometry method and device for predicting a quantification of a constituent from a sample
CN111007018A (en) * 2019-12-18 2020-04-14 湖北久之洋红外系统股份有限公司 Background estimation method and system for spectrum gas detection
CN111291675A (en) * 2020-02-04 2020-06-16 北京科技大学 Hyperspectral ancient painting detection and identification method based on deep learning
CN111597951A (en) * 2020-05-12 2020-08-28 中国电子科技集团公司第五十四研究所 Hyperspectral image feature selection method for target detection

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1546959A (en) * 2003-12-12 2004-11-17 上海交通大学 Nonlinear spectrum similarity measurement method
CN104463808A (en) * 2014-12-25 2015-03-25 武汉大学 High-spectral data noise reduction method and system based on spatial correlation
CN106023065A (en) * 2016-05-13 2016-10-12 中国矿业大学 Tensor hyperspectral image spectrum-space dimensionality reduction method based on deep convolutional neural network
WO2019194693A1 (en) * 2018-04-05 2019-10-10 Inesc Tec - Instituto De Engenharia De Sistemas E Computadores, Tecnologia E Ciência Spectrophotometry method and device for predicting a quantification of a constituent from a sample
CN108596077A (en) * 2018-04-20 2018-09-28 南阳理工学院 The adaptive spectrum focus wave section selection method of high spectrum image
CN111007018A (en) * 2019-12-18 2020-04-14 湖北久之洋红外系统股份有限公司 Background estimation method and system for spectrum gas detection
CN111291675A (en) * 2020-02-04 2020-06-16 北京科技大学 Hyperspectral ancient painting detection and identification method based on deep learning
CN111597951A (en) * 2020-05-12 2020-08-28 中国电子科技集团公司第五十四研究所 Hyperspectral image feature selection method for target detection

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘娟娟: "基于线性模型的高光谱图像解混及应用", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 *

Also Published As

Publication number Publication date
CN112329792B (en) 2022-12-09

Similar Documents

Publication Publication Date Title
CN107451614B (en) Hyperspectral classification method based on fusion of space coordinates and space spectrum features
Sun et al. Low-rank and sparse matrix decomposition-based anomaly detection for hyperspectral imagery
Gu et al. A selective KPCA algorithm based on high-order statistics for anomaly detection in hyperspectral imagery
CN104463224B (en) High spectrum image solution mixing method and system based on abundance significance analysis
CN112990313B (en) Hyperspectral image anomaly detection method and device, computer equipment and storage medium
CN111160273A (en) Hyperspectral image space spectrum combined classification method and device
CN109034213B (en) Hyperspectral image classification method and system based on correlation entropy principle
Wang et al. Hyperspectral image classification based on capsule network
WO2023115682A1 (en) Hyperspectral traditional chinese medicine identification method based on adaptive random block convolutional kernel network
CN113269201A (en) Hyperspectral image band selection method and system based on potential feature fusion
CN112733736A (en) Class imbalance hyperspectral image classification method based on enhanced oversampling
CN111597951A (en) Hyperspectral image feature selection method for target detection
CN115661069A (en) Hyperspectral anomaly detection method and computer device
CN107203779A (en) The EO-1 hyperion dimension reduction method kept based on empty spectrum information
Nyasaka et al. Learning hyperspectral feature extraction and classification with resnext network
CN112329792B (en) Hyperspectral image target feature extraction method based on spectrum angle
Robila et al. A fast source separation algorithm for hyperspectral image processing
Xiang et al. Pixel-associated autoencoder for hyperspectral anomaly detection
CN114022752B (en) SAR target detection method based on attention feature refinement and alignment
CN112396066B (en) Feature extraction method suitable for hyperspectral image
CN112330622B (en) Hyperspectral image band selection method based on ground feature maximum discrimination
Lu et al. Sparse representation based hyperspectral anomaly detection via adaptive background sub-dictionaries
Vafadar et al. Hyperspectral anomaly detection using Modified Principal component analysis reconstruction error
Cheng et al. Manifold regularized low-rank representation for hyperspectral anomaly detection
Xu et al. Gabor-filtering-based probabilistic collaborative representation for hyperspectral image classification

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