CN107316309A - High spectrum image conspicuousness object detection method based on matrix decomposition - Google Patents
High spectrum image conspicuousness object detection method based on matrix decomposition Download PDFInfo
- Publication number
- CN107316309A CN107316309A CN201710510904.2A CN201710510904A CN107316309A CN 107316309 A CN107316309 A CN 107316309A CN 201710510904 A CN201710510904 A CN 201710510904A CN 107316309 A CN107316309 A CN 107316309A
- Authority
- CN
- China
- Prior art keywords
- spectrum
- mrow
- matrix
- image
- conspicuousness
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
- G06T7/45—Analysis of texture based on statistical description of texture using co-occurrence matrix computation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of EO-1 hyperion conspicuousness object detection method based on matrix decomposition.By calculating spectrum gradient in the spectrum of original high spectrum image dimension, the spectrum Gradient Features of image are extracted, the adverse effect that illumination is brought is eliminated, while building image characteristic matrix；Then, decomposed by entering row matrix low-rank sparse to image characteristic matrix, the low-rank matrix of correspondence background parts and the sparse matrix of correspondence conspicuousness target are obtained, so as to avoid the piecemeal heterogeneity problem of conspicuousness interior of articles, operand can also be reduced by realizing while conspicuousness target detection.
Description
Technical field
The invention belongs to technical field of image processing, it is related to the conspicuousness object detection method of high spectrum image, and in particular to
A kind of high spectrum image conspicuousness object detection method based on matrix decomposition.
Background technology
High-spectrum seems to be recorded the spectral information of the various atural objects observed in visual field using imaging spectrometer
Obtained image data.With the increasingly maturation of high light spectrum image-forming technology, imaging device is in its spectral resolution and spatial discrimination
There is very big lifting in the indexs such as rate.So that the class such as the object detection mainly carried out originally on normal image, identification and tracking
Topic is gradually able to extend on high-spectral data.At present, for the correlative study of high spectrum image conspicuousness target detection problems
Still it is in developing stage.Existing high spectrum image conspicuousness object detection method mainly uses Itti models, by color characteristic
The spectral signature of high spectrum image is replaced with, model is applied to high spectrum image.For example, document " S.L.Moan,
A.Mansouri,et al.,Saliency for Spectral Image Analysis[J].IEEE Journal of
Selected Topics in Applied Earth Observations and Remote Sensing,2013.6(6):
P.2472-2479. it is exactly " that spectrum projection is subjected to principal component analysis (Principle into CIELAB color spaces using image
Component Analysis, PCA) etc. mode spectral information is utilized.At present, existing method is used as conspicuousness using pixel
The base unit of estimation, is assessed not by means such as principal component analysis, Euclidean distance, spectrum angles (Spectral Angle)
With the difference between pixel spectrum, the conspicuousness of each pixel is weighed out whereby.It is this that full figure conspicuousness is reflected by pixel conspicuousness
The subject matter of way be in testing result that object edge response is larger and internal response very low notable figure heterogeneity shows
As.In addition, existing method all relies on single model, it is impossible to eliminate what brightness change in high spectrum image was caused to spectroscopic data
Influence, and the huge amount of calculation brought by data scale.Therefore, it is badly in need of intrinsic in the existing hyperspectral detection method of Gonna breakthrough
Thinking, proposes the problem of new high spectrum image conspicuousness object detection method is to solve existing.
The content of the invention
In order to overcome the deficiencies in the prior art, the present invention provides a kind of EO-1 hyperion conspicuousness target inspection based on matrix decomposition
Survey method.The thought of matrix decomposition is different from Itti models and regional correlation method, and the thought of matrix decomposition is to utilize EO-1 hyperion
Similitude and continuity spectrally, construction sparse matrix and low-rank matrix on image space, and conspicuousness target is precisely
Sparse matrix in matrix decomposition.When building eigenmatrix to be decomposed, brightness irregularities pair are eliminated using spectrum Gradient Features
The influence that data are caused, builds super-pixel structure to reduce operand using bottom.The method that tradition utilizes regional correlation thought,
Conspicuousness interior of articles is caused inhomogenous problem occur, and the inventive method is based on matrix decomposition, jumps out the frame of conventional method
Frame causes conspicuousness interior of articles homogeneous homogeneous.
A kind of high spectrum image conspicuousness object detection method based on matrix decomposition, it is characterised in that including following step
Suddenly：
Step 1：Spectrum Gradient Features are generated：To each pixel i of high spectrum image, press
Calculating obtains its spectrum Gradient Features, and the spectrum Gradient Features of all pixels are to constitute spectrum gradient feature data block X={ g1,
g2,…,gn}T；Wherein,J-th of component of pixel i spectrum Gradient Features is represented,Represent that pixel i is corresponding original
Spectral vector yiJ-th of component, i=1,2 ..., n, j=2 ..., p, n is the total number of pixels of high spectrum image, and p is bloom
The wave band number of spectrogram picture, Δ λ is adjacent band wavelength difference；
Step 2：Build image characteristic matrix：The spectrum gradient feature data block X that step 1 is obtained is expressed as a size
For m × n two dimensional image eigenmatrix F, wherein, each row are the corresponding spectrum Gradient Features of a pixel, and m is spectrum ladder
Spend the dimension of feature, m=p-1；
Step 3：Low-rank sparse matrix decomposition：Image characteristic matrix F is subjected to low-rank sparse matrix decomposition, and profit as the following formula
It is solved with alternating iteration multiplier method, sparse matrix S is obtained：
Wherein, L is low-rank matrix, represents background parts；S is sparse matrix, represents conspicuousness target；λ is weight coefficient；
||·||*It is nuclear norm, | | | |1It is 1 norm；
Step 4：Notable figure is calculated：The openness matrix S that step 3 is obtained is normalized, and obtains notable figure, i.e.,
Conspicuousness object detection results.
The beneficial effects of the invention are as follows：By calculating spectrum gradient in the spectrum of original high spectrum image dimension, figure is extracted
The spectrum Gradient Features of picture, eliminate the adverse effect that illumination is brought, while image characteristic matrix can be built；By to image
Eigenmatrix enters the decomposition of row matrix low-rank sparse, low-rank background parts and sparse conspicuousness matrix is obtained, so as to avoid aobvious
The piecemeal heterogeneity problem of work property interior of articles.
Embodiment
With reference to embodiment, the present invention is further described, and the present invention includes but are not limited to following embodiments.
High-spectrum remote sensing is a cube structure, the space dimension corresponding pixel of reflection ground diverse location it is a certain too
Reflectivity on sunlight wave band, the relation of pixel incident light and the reflected light on different-waveband of a certain position of spectrum dimension reflection.
One panel height spectrum picture can be expressed as one p × n data acquisition system Yn={ y1,y2,…,yn, wherein, yiIt is corresponding for pixel i
Original spectrum vector, i=1,2 ..., n, n is the sum of all pixels of high spectrum image.
1st, spectrum Gradient Features are generated
Spectrum gradient refers to the ratio along the difference of the vectorial each two adjacent component of original spectrum and the difference of corresponding wavelength.And
A series of vector being made up of spectrum gradients is referred to as spectrum Gradient Features.Obtained by calculating spectrum Gradient Features to each pixel
To spectrum gradient image, extracted spectrum Gradient Features can be made to maintain the spatial relationship of original image.
Wherein：J-th of component of pixel i spectrum Gradient Features is represented,Represent original spectrum vector yiJth
Individual component, i=1,2 ..., n, j=2 ..., p, p is the wave band number of high spectrum image, and Δ λ is adjacent band wavelength difference.
Its corresponding spectrum Gradient Features, all spectrum is calculated according to formula (2) to each pixel in high spectrum image
Gradient Features are to constitute spectrum gradient data block X={ g1,g2,…,gn}T.Spectrum Gradient Features can reduce to a certain extent
Because of luminance difference caused by uneven illumination, so as to also can just weaken as this species diversity to the influence caused by subsequent step.
2nd, image characteristic matrix is built
Complete after above-mentioned steps, obtained spectrum gradient feature data block X is converted into a two-dimentional characteristics of image square
Battle array F.The each of two dimensional image eigenmatrix F is classified as the corresponding spectrum Gradient Features of a pixel, and columns is number of pixels n,
Line number is the dimension p-1 of spectrum Gradient Features.
3rd, low-rank sparse matrix decomposition
Due to the spatial simlanty in background, the pixel in particularly same conspicuousness object, all with local space phase
Like property, it therefore, it can detect conspicuousness target using low-rank sparse matrix decomposition.In addition, being existed in natural landscape big
The similar space structure of amount, this redundancy spatially means that background is the bottom of low-rank.In order to distinguish background and conspicuousness
Object, using following matrix decomposition model：
Wherein：L is low-rank matrix, and S is sparse matrix, is weight coefficient, can use λ=3 in any rational, the present invention, |
|·||*It is nuclear norm, | | | |1It is 1 norm.
Image characteristic matrix F is resolved into a low-rank background matrix L and a sparse detection matrix S, passes through nuclear norm
To limit L low-rank, the openness of S is limited by 1 norm.Because compared with whole image, conspicuousness object is only occupied
A small amount of pixel, therefore, the solution of conspicuousness target can find sparse matrix S by solving equation (3), avoid well
The limitation that Itti models are brought.
To target equation (3), because it is related to two variables S, L, they are separable targets, therefore, using friendship
Solved, had for direction multiplier method (Alternating Direction Method of Multipliers, abbreviation ADMM)
Body process is as follows：
First, an auxiliary variable H is introduced, and equation (3) is write as：
Then, introducing Lagrange multiplier is deformed into function：
Wherein：Lγ,η(L, S, H, P, Q) is Lagrange's equation, and P, Q are Lagrange multipliers, and γ, η are penalty coefficients,
||·||FIt is F norms, const is constant.
Next, being iterated renewal according to the following steps：
(1) fixed other specification, updates matrix L as the following formula：
Wherein, U Σ VTIt is Hk+Qk/ η singular value decomposition,Represent the soft-threshold of calculating matrix ∑.
(2) fixed other specification, updates matrix S as the following formula：
Wherein,Represent the operation of calculating matrix soft-threshold.
(3) fixed other specification, updates matrix H as the following formula：
Wherein, I is unit matrix.
Repeat step (1)-(3) are iterated renewal, until iterations k exceedes defined maximum times, orWhen (ε >=0), iteration update terminates, obtain sparse matrix S.
4th, notable figure is calculated
The openness matrix S that above step is obtained passes through normalization operation, has just obtained notable figure, i.e. conspicuousness target
Testing result.
By experimental verification, the inventive method is when recall ratio is 0.7, and precision ratio can improve 10%；It is in precision ratio
When 0.7, recall ratio can improve 20%.
Claims (1)
1. a kind of high spectrum image conspicuousness object detection method based on matrix decomposition, it is characterised in that comprise the following steps：
Step 1：Spectrum Gradient Features are generated：To each pixel i of high spectrum image, pressCalculate
Its spectrum Gradient Features is obtained, the spectrum Gradient Features of all pixels are to constitute spectrum gradient feature data block X={ g1,g2,…,
gn}T；Wherein,J-th of component of pixel i spectrum Gradient Features is represented,Represent the corresponding original spectrum of pixel i to
Measure yiJ-th of component, i=1,2 ..., n, j=2 ..., p, n is the total number of pixels of high spectrum image, and p is high spectrum image
Wave band number, Δ λ is adjacent band wavelength difference；
Step 2：Build image characteristic matrix：The spectrum gradient feature data block X that step 1 is obtained is expressed as a size for m
× n two dimensional image eigenmatrix F, wherein, each row are the corresponding spectrum Gradient Features of a pixel, and m is spectrum gradient
The dimension of feature, m=p-1；
Step 3：Low-rank sparse matrix decomposition：Image characteristic matrix F is subjected to low-rank sparse matrix decomposition as the following formula, and utilizes friendship
It is solved for iteration m ultiple method, sparse matrix S is obtained：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mrow>
<mi>L</mi>
<mo>,</mo>
<mi>S</mi>
</mrow>
</munder>
<mo>|</mo>
<mo>|</mo>
<mi>L</mi>
<mo>|</mo>
<msub>
<mo>|</mo>
<mo>*</mo>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>|</mo>
<mo>|</mo>
<mi>S</mi>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>1</mn>
</msub>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>F</mi>
<mo>=</mo>
<mi>S</mi>
<mo>+</mo>
<mi>L</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, L is low-rank matrix, represents background parts；S is sparse matrix, represents conspicuousness target；λ is weight coefficient；||·
||*It is nuclear norm, | | | |1It is 1 norm；
Step 4：Notable figure is calculated：The openness matrix S that step 3 is obtained is normalized, and obtains notable figure, i.e., significantly
Property object detection results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710510904.2A CN107316309B (en) | 2017-06-29 | 2017-06-29 | Hyperspectral image saliency target detection method based on matrix decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710510904.2A CN107316309B (en) | 2017-06-29 | 2017-06-29 | Hyperspectral image saliency target detection method based on matrix decomposition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107316309A true CN107316309A (en) | 2017-11-03 |
| CN107316309B CN107316309B (en) | 2020-04-03 |
Family
ID=60180808
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710510904.2A Active CN107316309B (en) | 2017-06-29 | 2017-06-29 | Hyperspectral image saliency target detection method based on matrix decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107316309B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107782748A (en) * | 2017-11-20 | 2018-03-09 | 福建师范大学福清分校 | Microwave thermal imaging non-destructive detecting system and detection method based on matrix decomposition |
| CN108562541A (en) * | 2018-04-23 | 2018-09-21 | 南京大学 | Based on matrix decomposition without lens holographic microphotography speckle noise minimizing technology and device |
| CN108765313A (en) * | 2018-05-02 | 2018-11-06 | 西北工业大学 | High spectrum image denoising method based on low-rank representation in class |
| CN108986060A (en) * | 2018-06-25 | 2018-12-11 | 南京大学 | Based on sparse several picture reflected light suppressing methods decomposed with low-rank matrix |
| CN110275842A (en) * | 2018-07-09 | 2019-09-24 | 西北工业大学 | EO-1 hyperion Target Tracking System and method based on FPGA |
| CN111160300A (en) * | 2019-12-31 | 2020-05-15 | 北京理工大学重庆创新中心 | Deep learning hyperspectral image saliency detection algorithm combined with global prior |
| CN112784747A (en) * | 2021-01-22 | 2021-05-11 | 哈尔滨工业大学 | Multi-scale eigen decomposition method for hyperspectral remote sensing image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104281855A (en) * | 2014-09-30 | 2015-01-14 | 西安电子科技大学 | Hyperspectral image classification method based on multi-task low rank |
| CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
| CN106203523A (en) * | 2016-07-17 | 2016-12-07 | 西安电子科技大学 | The classification hyperspectral imagery of the semi-supervised algorithm fusion of decision tree is promoted based on gradient |
| CN106204613A (en) * | 2016-07-20 | 2016-12-07 | 中国科学院自动化研究所 | The display foreground object detecting method represented based on low-rank matrix and detecting system |
| CN106447667A (en) * | 2016-10-31 | 2017-02-22 | 郑州轻工业学院 | Visual significance detection method based on self-learning characteristics and matrix low-rank recovery |
-
2017
- 2017-06-29 CN CN201710510904.2A patent/CN107316309B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104281855A (en) * | 2014-09-30 | 2015-01-14 | 西安电子科技大学 | Hyperspectral image classification method based on multi-task low rank |
| CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
| CN106203523A (en) * | 2016-07-17 | 2016-12-07 | 西安电子科技大学 | The classification hyperspectral imagery of the semi-supervised algorithm fusion of decision tree is promoted based on gradient |
| CN106204613A (en) * | 2016-07-20 | 2016-12-07 | 中国科学院自动化研究所 | The display foreground object detecting method represented based on low-rank matrix and detecting system |
| CN106447667A (en) * | 2016-10-31 | 2017-02-22 | 郑州轻工业学院 | Visual significance detection method based on self-learning characteristics and matrix low-rank recovery |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107782748A (en) * | 2017-11-20 | 2018-03-09 | 福建师范大学福清分校 | Microwave thermal imaging non-destructive detecting system and detection method based on matrix decomposition |
| CN107782748B (en) * | 2017-11-20 | 2023-12-19 | 福建技术师范学院 | Microwave thermal imaging nondestructive detection system and detection method based on matrix decomposition |
| CN108562541A (en) * | 2018-04-23 | 2018-09-21 | 南京大学 | Based on matrix decomposition without lens holographic microphotography speckle noise minimizing technology and device |
| CN108765313A (en) * | 2018-05-02 | 2018-11-06 | 西北工业大学 | High spectrum image denoising method based on low-rank representation in class |
| CN108986060A (en) * | 2018-06-25 | 2018-12-11 | 南京大学 | Based on sparse several picture reflected light suppressing methods decomposed with low-rank matrix |
| CN108986060B (en) * | 2018-06-25 | 2021-09-28 | 南京大学 | Multi-image reflected light suppression method based on sparse and low-rank matrix decomposition |
| CN110275842A (en) * | 2018-07-09 | 2019-09-24 | 西北工业大学 | EO-1 hyperion Target Tracking System and method based on FPGA |
| CN110275842B (en) * | 2018-07-09 | 2022-10-21 | 西北工业大学 | Hyperspectral target tracking system and method based on FPGA |
| CN111160300A (en) * | 2019-12-31 | 2020-05-15 | 北京理工大学重庆创新中心 | Deep learning hyperspectral image saliency detection algorithm combined with global prior |
| CN111160300B (en) * | 2019-12-31 | 2022-06-28 | 北京理工大学重庆创新中心 | Deep learning hyperspectral image saliency detection algorithm combined with global prior |
| CN112784747A (en) * | 2021-01-22 | 2021-05-11 | 哈尔滨工业大学 | Multi-scale eigen decomposition method for hyperspectral remote sensing image |
| CN112784747B (en) * | 2021-01-22 | 2022-03-04 | 哈尔滨工业大学 | Multi-scale eigen decomposition method for hyperspectral remote sensing image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107316309B (en) | 2020-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107316309A (en) | High spectrum image conspicuousness object detection method based on matrix decomposition | |
| CN104978573B (en) | A kind of non-negative matrix factorization method applied to Hyperspectral imagery processing | |
| CN104952050B (en) | High spectrum image adaptive de mixing method based on region segmentation | |
| Mohan et al. | Hyperspectral image processing and analysis | |
| CN107798345B (en) | High-spectrum disguised target detection method based on block diagonal and low-rank representation | |
| CN102540271B (en) | Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method | |
| CN102609944B (en) | Hyper-spectral remote sensing image mixed pixel decomposition method based on distance geometry theory | |
| CN107688003B (en) | Blade reflectivity satellite remote sensing extraction method for eliminating vegetation canopy structure and earth surface background influence | |
| CN104809471B (en) | A kind of high spectrum image residual error integrated classification method based on spatial spectral information | |
| CN108830130A (en) | A kind of polarization EO-1 hyperion low-altitude reconnaissance image typical target detection method | |
| CN103413292A (en) | Hyperspectral image nonlinear abundance estimation method based on constrained least squares | |
| Okamura et al. | Feasibility study of multi-pixel retrieval of optical thickness and droplet effective radius of inhomogeneous clouds using deep learning | |
| CN105913448A (en) | Hyperspectral image target detection method based on tensor matched subspace | |
| CN115620128A (en) | Hyperspectral anomaly detection method | |
| CN101908211B (en) | High spectral image fusion method based on variational method | |
| CN117115669B (en) | Object-level ground object sample self-adaptive generation method and system with double-condition quality constraint | |
| Meng et al. | Improving the spatial resolution of hyperspectral image using panchromatic and multispectral images: An integrated method | |
| CN112784747B (en) | Multi-scale eigen decomposition method for hyperspectral remote sensing image | |
| CN102609963B (en) | Simulation method of hyperspectral images | |
| CN107152969A (en) | A kind of offshore type Fourier imaging spectrometer data processing method | |
| Bloechl et al. | A comparison of real and simulated airborne multisensor imagery | |
| Qin et al. | An illumination estimation and compensation method for radiometric correction of UAV multispectral images | |
| CN106503754B (en) | The multidate hyperspectral image classification method of global geometry is kept based on empty spectrum signature | |
| CN109285148A (en) | Based on weighting low-rank again and enhance sparse method for detecting infrared puniness target | |
| Li et al. | A multispectral remote sensing data spectral unmixing algorithm based on variational Bayesian ICA |
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 |