CN108460777A - A kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion - Google Patents

A kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion Download PDF

Info

Publication number
CN108460777A
CN108460777A CN201711404372.0A CN201711404372A CN108460777A CN 108460777 A CN108460777 A CN 108460777A CN 201711404372 A CN201711404372 A CN 201711404372A CN 108460777 A CN108460777 A CN 108460777A
Authority
CN
China
Prior art keywords
plant
hyperion
formula
indicate
1dsma
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
CN201711404372.0A
Other languages
Chinese (zh)
Other versions
CN108460777B (en
Inventor
徐平
陈秉强
张竞成
李永文
肖冲
马林
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.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
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 Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201711404372.0A priority Critical patent/CN108460777B/en
Publication of CN108460777A publication Critical patent/CN108460777A/en
Application granted granted Critical
Publication of CN108460777B publication Critical patent/CN108460777B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • 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
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Image Processing (AREA)

Abstract

The present invention discloses a kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion.Elder generation primordial plant EO-1 hyperion of the present invention carries out plant regional extraction, then use splits' positions data be unfolded it is formal take full advantage of pocket between height spatial coherence, so that the data to be reconstructed after expansion not only keep consistent with original monochromatic light spectral curve on data assignment, improve the length of single data to be reconstructed again simultaneously, this so that the collected data volume of single is more under on the one hand identical sample rate, substantially increases the probability successfully reconstructed；On the other hand so that having been able to obtain enough spectral informations to the sampling of primordial plant EO-1 hyperion with lower sample rate, so as to complete the compression reconfiguration to plant EO-1 hyperion with lower compression ratio, carrying cost is accordingly reduced in practical applications, while for the plant EO-1 hyperion of reconstruct provides safeguard for plant physiology parametric inversion.

Description

A kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion
Technical field
The invention belongs to plant Hyperspectral imagery processing technical fields, are related to a kind of extraction piecemeal towards plant EO-1 hyperion Compression reconfiguration method.
Background technology
The mode that tradition obtains plant physiology parameter is that plant sample is taken to carry out a series of experiment in the lab, this A process needs a large amount of manpower and the experimental facilities of costliness.Remote sensing technology is as a kind of quick, macroscopic view earth's surface resource prison Survey technology means have objective, the lossless and real-time advantage for obtaining information, especially bloom compared with traditional ground investigation The appearance and development for composing remote sensing technology, new opportunity is brought to the quantification inverting of vegetation physiology parameter：High-spectral data In comprising the more rich spectral information of vegetation, which greatly improves to vegetation object physiological parameter (such as leaf area index and biology Amount) inverting precision, more importantly, the depigmentation (such as chlorophyll) for also making original difficulty larger outside other vegetation biology The remote-sensing inversion of chemical parameters is possibly realized.
However, existing main problem is that plant high-spectral data amount is huge during obtaining plant EO-1 hyperion, this A large amount of memory space is intended to be stored.With the development of high spectrum resolution remote sensing technique, the resolution of plant EO-1 hyperion Rate is higher and higher, and also just to storage hardware, more stringent requirements are proposed for this.To reduce carrying cost, there is an urgent need to a kind of new reasons By improving compression ratio, and these a small amount of collected data can be utilized accurately to reconstruct plant EO-1 hyperion.
Compressive sensing theory is theoretical as a kind of novel data acquisition, and the sampling of data and compression process are dexterously tied Altogether, it realizes that the gathered data with sample rate far below conventional Nyquist rate carries out Accurate Reconstruction, reduces to sensor With the requirement of storage hardware, effectively prevent pursuing the hardware and software cost problem that high-resolution is brought.Compressive sensing theory by Candes is equal to 2006 and proposes that the theory is at present to be applied to many popular domains.Such as in 3D imaging technique field, Li Jun Etc. proposing a kind of compression optics Method of Steganography；In optical imagery field, Liu is imitated the one kind that proposes such as brave and is felt based on compression The optical image encryption technology known；In medical imaging field, Zhou Chongbin etc.；Plant EO-1 hyperion field has scholar and feels compression Know that theory is applied to plant leaf blade radiation transmission data model, the experimental results showed that the theory to be applied to the pressure of plant EO-1 hyperion Influence of the contracting reconstruct to plant physiology parametric inversion is little.
To the compression reconfiguration of practical plant EO-1 hyperion, existing method is：From image area, monochromatic light spectral curve and piecemeal Form carries out compression reconfiguration to plant EO-1 hyperion, such as Liu Junfeng proposes a kind of adapter distribution towards plant EO-1 hyperion Compression reconfiguration.However it is for plant physiology parametric inversion, from image area pair for one of important application of plant EO-1 hyperion Plant EO-1 hyperion progress compression reconfiguration has ignored the high correlation between plant EO-1 hyperion spectrum, this to reconstruct in low compression ratio Plant EO-1 hyperion can lose information between the spectrum of primordial plant EO-1 hyperion, carry out physiology parameter so as to cause reconstruction result is utilized Inverting can have larger error.Single spectrum curve compression reconstruct quality reconstruction in low compression ratio is bad, cannot achieve low pressure The Accurate Reconstruction of contracting ratio.And splits' positions are disadvantageous in that relative to single spectrum curve compression reduces reconstruct efficiency, because Wait for that reconstruction signal is longer for single, it is meant that mutually increase again in requisition for the calculation matrix of construction and sparse basis exponentially, greatly greatly The big complexity of algorithm so that reconstruct efficiency substantially reduces.In fact, plant EO-1 hyperion there is we and it is uninterested Background area, the reconstruct of previous image domain compression, the reconstruct of single spectrum curve compression and splits' positions reconstruct all include to background Etc. extraneous areas compression reconfiguration.If background area can be removed, compression reconfiguration only is carried out to plant regional, no Reconstruct efficiency can be only improved, and reconstruct effect can be further increased since all bit rates are all assigned to plant regional Fruit.Based on this present invention it is further proposed that a kind of extraction towards plant EO-1 hyperion point on the basis of splits' positions reconstructing method Block compression reconfiguration method.
Invention content
The splits' positions reconstruct towards plant EO-1 hyperion that in view of the deficiencies of the prior art, it is an object of the present invention to provide a kind of Method.
The method of the present invention specifically includes following steps：
Step (1) carries out plant regional extraction to primordial plant EO-1 hyperion
Primordial plant EO-1 hyperion is the data of three-dimensional, establishes three-dimensional cartesian coordinate system, the figure of plant EO-1 hyperion as shown in Figure 1 As tieing up respectively as x and y-axis, spectrum dimension is used as z-axis.
With X3Dori ∈ RA×B×CIndicate primordial plant EO-1 hyperion, A, B, C indicate primordial plant EO-1 hyperion in three-dimensional straight respectively Along the length of x, y, z axis direction under angular coordinate system；
With X3Dtq ∈ RA×B×CIndicate the EO-1 hyperion of extraction plant regional；With X3Dym ∈RA×B×CIndicate two-value mask bloom Spectrum.X3DtqAcquisition process it is as follows：
The curve of spectrum in 1.1 analysis primordial plant EO-1 hyperion different type regions, chooses plant regional and background area is poor The z values (corresponding wavelength value also can be used) that value is maximum and its neighbouring curve of spectrum is more smooth, using the corresponding image of the value as Benchmark image obtains X according to formula (1) and formula (2)3Dym；More smooth judgement herein is provided with for those of ordinary skill.
X3Dym(x, y, z)=0s.t.X3Dori(x,y,z)≥Thr (1)
X3Dym(x, y, z)=1s.t.X3Dori(x,y,z)<Thr (2)
Wherein x ∈ (0, A-1), y ∈ (0, B-1), z ∈ (0, C-1)；Thr is threshold value, image plant regional on the basis of value The average value of marginal point.
1.2 obtain X according to formula (3)3Dtq
X3Dtq(x, y, z)=X3Dori(x,y,z)×X3Dym(x,y,z) (3)
Step (2) carries out piecemeal to the plant EO-1 hyperion of extraction, and sub-block is expressed under coordinate system
With X3Dsma ∈ Ra×b×cIndicate that sub-block, a, b, c indicate sub-block under three-dimensional cartesian coordinate system along x, y, z-axis respectively The length in direction；The plant EO-1 hyperion of extraction is subjected to piecemeal according to formula (4):
Wherein Q indicates that the total number of sub-block, M, N, H are indicated respectively along x, y, the block number of z-axis direction sub-block.
To retain the spectral domain information of plant EO-1 hyperion, in z-axis direction without segmentation, that is, c=C is taken, then formula (4) can be with It is reduced to formula (5).
After piecemeal, each sub-block can be indicated with formula (6)：
X3Dsma=X3Dtq(x,y,z) (6)
Wherein x ∈ (a × i-a, a × i-1), y ∈ (b × j-b, b × j-1), z ∈ (0, C-1) indicate primordial plant respectively Each coordinate of the data under three-dimensional cartesian coordinate system in EO-1 hyperion；I=1,2 ..., M, j=1,2 ..., N indicate each respectively Index of the fritter under three-dimensional cartesian coordinate system along the x-axis direction and along the y-axis direction, when initialization, take i=1, j=1.
Step (3) takes sub-block and is launched into one-dimensional data
Take sub-block X3Dsma=X3Dtq(x, y, z), X3DsmaZ-axis direction correspond to C it is big it is small be a × b image.From One image starts, and first fetches successively evidence further along y-axis direction along the x-axis direction, takes one every time, judge whether the data are more than Zero, the element is incorporated to X if it is greater than zero1Dsma, then the coordinate of the element is incorporated in Mask, after an image procossing It takes remaining image to be repeated in the above operation again, obtains the one-dimensional data X that the fritter is finally unfolded1Dsma ∈ RNr×1, wherein Nr Indicate sub-block X3DsmaThe number (i.e. plant regional data amount check) of middle nonzero element；
Step (4) constructs calculation matrix and obtains measured value
4.1 construction random Gaussian calculation matrix Φ ∈ Rm×Nr, wherein m is number of samples, has m=Nr × Bpp, the Bpp to be Sample rate (i.e. compression ratio)；
4.2 sample the one-dimensional data being unfolded in step (3) further according to formula (7), obtain measured value Y1Dsma ∈ Rm ×1。
Y1DSma=Φ X1Dsma (7)
Step (5) constructs discrete cosine transformation matrix Ψ ∈ R according to formula (8)Nr×Nr；Then in conjunction with random in step (4) Gauss measurement matrix Φ obtains sensing matrix A according to formula (9)sma ∈ Rm×Nr；
Asma=Φ Ψ (9)
Step (6), using segmentation orthogonal matching pursuit algorithm to the measured value Y that obtains in step (4)1DSmaIt is reconstructed, Obtain X1DsmaApproximationSpecifically：
6.1 initialization：T=1, rt-1=Y1Dsma,Wherein t indicates iterations, rt-1Table Show residual error, ΛtIndicate the index set of the t times iteration, AtIt indicates by index ΛtFrom AsmaIn the row set selected.
6.2 calculate the inner product u of sensing matrix and residual error according to formula (10)；The value for being more than Th in u is selected, these values are corresponded to U in row serial number, that is, AsmaIn row serial number k constitute set K.Wherein Th is threshold value, and calculation expression is formula (11)：
Wherein abs [] indicates absolute value,Indicate AsmaTransposition.
Wherein | | | |2Indicate two norms.
6.3 gather into row element expansion the index set and row of next iteration according to formula (12) and formula (13)；If Λt=Λt-1, then stop iteration and enter step 6.7；
Λt=Λt-1 ∪ K (12)
At=At-1 ∪ ak (13)
Wherein akIndicate sensing matrix AsmaIn kth row.
6.4 seek Y according to formula (14)1Dsma=AtθtLeast square solution
6.5 update residual error r according to formula (15)t
6.6 judge whether iterations reach total iterations
T=t+1 is enabled, return to step 6.2 continues iteration if t≤S；Otherwise stop iteration, enter step 6.7.Wherein S For total iterations, value generally takes 10.
6.7 calculate X according to formula (16)1DsmaApproximation
The one-dimensional data of reconstruct is reverted to form and the storage of sub-block by step (7), inverse expansion process
It willInverse expansion process processing is carried out in conjunction with Mask, is obtainedThen according to formula (17) it is stored.
Wherein X3Drec ∈ RA×B×CTo store the matrix of reconstruction result.
Above-mentioned inverse expansion process is first along the y-axis direction further along x-axis direction.
Step (8) judges whether all sub-blocks have reconstructed, specifically：
8.1 enable i=i+1, if i≤M, return to step (3) otherwise enters step 8.2；
8.2 enable i=1, j=j+1, if j≤N, return to step (3) otherwise terminates.
The advantageous effect of the invention is：
1) form of piecemeal takes full advantage of the spatial coherence inside plant EO-1 hyperion sub-block, while the expansion side of sub-block Formula fully remains Spectral correlation again, this makes there is higher reconstruct relative to single spectrum compress mode under identical compression ratio Precision.
2) introduced plant extracted region on the basis of piecemeal only carries out compression reconfiguration to the plant regional extracted, On the one hand reconstruct efficiency can be improved, on the other hand since all bit rates are all assigned to plant regional and significantly improve reconstruct Precision.
3) extraction splits' positions are that targetedly the plant regional to plant EO-1 hyperion carries out compression reconfiguration so that are being pressed Contracting level obtains compression ratio as big as possible, this so that only needing to acquire a small amount of data in practical applications can restore well Go out primordial plant EO-1 hyperion, reduces carrying cost, while for the plant EO-1 hyperion of reconstruct is further applied to plant physiology Parametric inversion provides safeguard.
Description of the drawings
Fig. 1 is the three-dimensional cartesian coordinate system established on primordial plant EO-1 hyperion；
Fig. 2 is the flow chart of plant EO-1 hyperion extraction splits' positions reconstruct；
Fig. 3 is the averaged spectrum curve in tealeaves EO-1 hyperion different type region；
Fig. 4 is the flow chart for being segmented orthogonal matching pursuit algorithm reconstruct；
Fig. 5 is comparison of the reconstruction result in image area；(a) it is 0.05 compression ratio, be 0.10 compression ratio (c) is (b) 0.15 Compression ratio；
Fig. 6 is comparison of the reconstruction result in spectral domain；(a) it is 0.05 compression ratio, is (b) 0.10 compression ratio, is (c) 0.15 Compression ratio；
Fig. 7 is the reconstruct efficiency comparative under three kinds of method difference compression ratios.
Specific implementation mode
With reference to specific embodiment, the present invention is further analyzed with attached drawing.
In the present embodiment, the plant hyperspectral image data used is the hyperspectral image data of tealeaves.The tealeaves bloom The wave-length coverage for composing image data is 480~820nm, accumulative to share 341 wave bands, and single pixel is 12 whole without symbol Shape, single-range image resolution ratio are 128x288, i.e. X3Dori ∈ R128×288×341。
Extraction splits' positions reconstructing method towards plant EO-1 hyperion, shown in Fig. 2, including it is as follows：
Step (1) carries out plant regional extraction to primordial plant EO-1 hyperion
Primordial plant EO-1 hyperion is the data of three-dimensional, establishes three-dimensional cartesian coordinate system, the figure of plant EO-1 hyperion as shown in Figure 1 As tieing up respectively as x and y-axis, spectrum dimension is used as z-axis.
With X3Dori ∈ RA×B×CIndicate primordial plant EO-1 hyperion, A, B, C indicate primordial plant EO-1 hyperion in three-dimensional straight respectively Along the length of x, y, z axis direction under angular coordinate system；
With X3Dtq ∈ RA×B×CIndicate the EO-1 hyperion of extraction plant regional；With X3Dym ∈RA×B×CIndicate two-value mask bloom Spectrum.
A bright area and a dark areas are respectively had chosen in tealeaves region and background area in the present embodiment, does average light Spectral curve is as shown in Figure 3.As can be seen from the figure tealeaves region is larger with background area grey value difference and it is more flat nearby Sliding wave-length coverage is about 650nm~670nm.The present embodiment takes the corresponding images of 670nm as benchmark image, according to formula (1) and formula (2) obtains mask EO-1 hyperion X3Dym, the EO-1 hyperion X in extraction tealeaves region is obtained further according to formula (3)3Dtq。
Step (2) carries out piecemeal to the plant high spectrum image of extraction, and sub-block is expressed under coordinate system
The block form used in the present embodiment is a × b × c=2 × 2 × 341, i.e. X3Dsma ∈ R2×2×341, then former Beginning tealeaves high spectrum image is divided into A sub-block.
Each sub-block can be expressed as X under three-dimensional cartesian coordinate system3Dsma=X3Dtq(x, y, z), wherein x ∈ (128 × i-128,128×i-1),y ∈ (288×j-288,288×j-1),z∈(0,340).(1,128) i ∈ in experiment, j ∈ (1,288) indicate respectively each fritter under three-dimensional cartesian coordinate system along the x-axis direction and along y axis directions index, when initialization Take i=1, j=1.
Step (3) takes sub-block and is launched into one-dimensional data
Take sub-block X3Dsma=X3Dtq(x, y, z), X3DsmaZ-axis direction correspond to it is 341 big it is small be 2 × 2 image.From First image starts, and first takes element successively further along y axis directions along the x-axis direction, judges whether the element is more than zero, if The element is incorporated to X more than zero1Dsma, then by the coordinate deposit Mask of the element, it is taken again after an image procossing Remaining image is repeated in the above operation, obtains the one-dimensional data X that the fritter is finally unfolded1Dsma ∈ RNr×1。
Step (4) constructs calculation matrix and obtains measured value
Construct random Gaussian calculation matrix Φ ∈ Rm×Nr, m successively takes the 5% of Nr, 10%, 15%, right further according to formula (7) The one-dimensional data X being unfolded in step 21DsmaIt is sampled, obtains measured value Y1Dsma ∈ Rm×1。
Step (5) constructs discrete cosine transformation matrix Ψ according to formula (8)；Then in conjunction with the Φ in step (4), further according to Formula (9) obtains sensing matrix Asma。
Step (6) is reconstructed measured value using segmentation orthogonal matching pursuit algorithm, obtains reconstruction valueSuch as Shown in Fig. 4.
Step (7), form and the storage that the one-dimensional data of reconstruct is reverted to in conjunction with Mask against expansion process sub-block
Step (8) judges whether all sub-blocks have reconstructed.
8.1 enable i=i+1, if i≤128, return to step (3) otherwise enters 8.2
8.2 enable i=1, j=j+1, if j≤288, return to step (3) otherwise terminates.
Fig. 5 is comparison of the reconstruction result in image domains, chooses 640nm respectively, under tri- wavelength of 720nm, 760nm Image is compared.As can be seen that there are much noises for reconstructed image under 0.05 compression ratio for single spectrum curve compression mode Point, when the above quality reconstruction is preferable for 0.10 compression；The image that splits' positions mode reconstructs under 0.05 compression ratio Slightly has difference with original image；Extraction splits' positions mode proposed by the present invention reconstructed image and original under 0.05 compression ratio Image is closely similar.
Fig. 6 is comparison of the reconstruction result in spectral domain.As can be seen that single spectrum curve compression mode 0.05,0.10 with And 0.15 compression ratio under the curve of spectrum that reconstructs there are apparent differences with original spectrum, especially 680nm~730nm it Between the part variation that is ramping up it is more apparent；The curve of spectrum and primary light reconstructed under the compression ratio of splits' positions mode 0.05 Spectral curve is there are apparent difference, but the curve of spectrum that is reconstructed at 0.10 or more of sample rate and original spectrum curve are very heavy It closes；The curve of spectrum that is reconstructed under 0.05 compression ratio of extraction splits' positions mode proposed by the present invention with original spectrum curve It overlaps very much.
Fig. 7 is the comparison that efficiency is reconstructed under three classes method difference sample rate.As can be seen that splits' positions mode reconstructs efficiency Significantly lower than single spectrum curve compression mode, and extraction splits' positions mode proposed by the present invention is relative to only splits' positions mode It has some improvement in reconstruct efficiency.
To sum up, extraction splits' positions mode proposed by the present invention, can not only improve image area under identical sample rate With the quality reconstruction of spectral domain, also make moderate progress in reconstruct efficiency relative to splits' positions mode.Further it is proposed that Extraction splits' positions mode further improves compression ratio relative to splits' positions mode, it means that can further decrease and deposit Cost is stored up, the development for high-resolution plant EO-1 hyperion in the future provides support.

Claims (2)

1. a kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion, it is characterised in that this approach includes the following steps：
Step (1) carries out plant regional extraction to primordial plant EO-1 hyperion
With X3Dori∈RA×B×CIndicate primordial plant EO-1 hyperion, A, B, C indicate primordial plant EO-1 hyperion in three-dimensional rectangular coordinate respectively Along the length of x, y, z axis direction under system；
With X3Dtq∈RA×B×CIndicate the EO-1 hyperion of extraction plant regional；With X3Dym∈RA×B×CIndicate two-value mask EO-1 hyperion, specifically Acquisition process is as follows：
The curve of spectrum in 1.1 analysis primordial plant EO-1 hyperion different type regions, chooses plant regional and background area difference most Big and more smooth its neighbouring curve of spectrum z values or wavelength value, using the corresponding image of the value as benchmark image, according to formula (1) and formula (2) obtains X3Dym；
X3Dym(x, y, z)=0s.t.X3Dori(x, y, z) >=Thr (1)
X3Dym(x, y, z)=1s.t.X3Dori(x, y, z) ＜ Thr (2)
Wherein x ∈ (0, A-1), y ∈ (0, B-1), z ∈ (0, C-1)；Thr is threshold value, image plant regional edge on the basis of value The average value of point；
1.2 obtain X according to formula (3)3Dtq；
X3Dtq(x, y, z)=X3Dori(x, y, z) × X3Dym(x, y, z) (3)；
Step (2) carries out piecemeal to the plant EO-1 hyperion of extraction
With X3Dsma∈Ra×b×cIndicate image subblock, a, b, c indicate sub-block under three-dimensional cartesian coordinate system along x, y, z axis side respectively To length；
The plant EO-1 hyperion of extraction is subjected to piecemeal according to formula (4)：
Wherein Q indicates the total number of sub-block, and M, N, H indicate the block number along x, y, z axis direction sub-block respectively；
To retain the spectral domain information of plant EO-1 hyperion, in z-axis direction without segmentation, that is, c=C is taken, then formula (4) can be reduced to Formula (5)：
After piecemeal, each sub-block can be indicated with formula (6)：
X3Dsma=X3Dtq(x, y, z) (6)
Wherein x ∈ (a × i-a, a × i-1), y ∈ (b × j-b, b × j-1), z ∈ (0, C-1) indicate primordial plant bloom respectively Each coordinate of the data under three-dimensional cartesian coordinate system in spectrum；I=1,2 ..., M, j=1,2 ..., N indicate each small respectively Index of the block under three-dimensional cartesian coordinate system along the x-axis direction and along the y-axis direction, when initialization, take i=1, j=1；
Step (3) takes sub-block and is launched into one-dimensional data
By X3DsmaAccording to first fetching successively evidence further along y-axis direction along the x-axis direction, one is taken every time, judges whether the data are big In zero, the element is incorporated to X if it is greater than zero1Dsma, then the coordinate of the element is incorporated in Mask, an image procossing finishes It takes remaining image to be repeated in the above operation again afterwards, obtains the one-dimensional data X that the fritter is finally unfolded1Dsma∈RNr×1, wherein Nr Indicate sub-block X3DsmaThe number of middle nonzero element, i.e. plant regional data amount check；
Step (4) constructs calculation matrix and obtains measured value
4.1 construction random Gaussian calculation matrix Φ ∈ Rm×Nr, wherein m is number of samples, and it is sample rate to have m=Nr × Bpp, Bpp (i.e. compression ratio)；
4.2 sample the one-dimensional data being unfolded in step (3) according to formula (7), obtain measured value Y1Dsma∈Rm×1；
Y1DSma=Φ X1Dsma (7)
Step (5) constructs discrete cosine transformation matrix Ψ ∈ R according to formula (8)Nr×Nr；It is surveyed then in conjunction with random Gaussian in step (4) Moment matrix Φ obtains sensing matrix A according to formula (9)sma∈Rm×Nr；
Asma=Φ Ψ (9)
Step (6), using segmentation orthogonal matching pursuit algorithm to the measured value Y that obtains in step (4)1DSmaIt is reconstructed, obtains X1DsmaApproximation
The one-dimensional data of reconstruct is reverted to form and the storage of sub-block by step (7), inverse expansion process
It willInverse expansion process processing is carried out in conjunction with Mask, is obtainedThen according to formula (17) into Row storage；
Wherein X3Drec∈RA×B×CTo store the matrix of reconstruction result；
Step (8) judges whether all sub-blocks have reconstructed, specifically：
8.1 enable i=i+1, if i≤M, return to step (3) otherwise enters step 8.2；
8.2 enable i=1, j=j+1, if j≤N, return to step (3) otherwise terminates.
2. a kind of extraction splits' positions reconstructing method towards plant EO-1 hyperion as described in claim 1, it is characterised in that step Suddenly (6) are specifically：
6.1 initialization：T=1, rt-1=Y1Dsma,Wherein t indicates iterations, rt-1Indicate residual Difference, ΛtIndicate the index set of the t times iteration, AtIt indicates by index ΛtFrom AsmaIn the row set selected；
6.2 calculate the inner product u of sensing matrix and residual error according to formula (10)；The value for being more than Th in u is selected, by the corresponding u of these values In row serial number, that is, AsmaIn row serial number k constitute set K；Wherein Th is threshold value, and calculation expression is formula (11)：
Wherein abs [] indicates absolute value,Indicate AsmaTransposition；
Wherein | | | |2Indicate two norms；
6.3 gather into row element expansion the index set and row of next iteration according to formula (12) and formula (13)；If Λt= Λt-1, then stop iteration and enter step 6.7；
Λt=At-1∪K (12)
At=At-1∪ak (13)
Wherein akIndicate sensing matrix AsmaIn kth row；
6.4 seek Y according to formula (14)1Dsma=AtθtLeast square solution
6.5 update residual error r according to formula (15)t
6.6 judge whether iterations reach total iterations
T=t+1 is enabled, return to step 6.2 continues iteration if t≤S；Otherwise stop iteration, enter step 6.7；Wherein S is total Iterations, value generally take 10；
6.7 calculate X according to formula (16)1DsmaApproximation
CN201711404372.0A 2017-12-22 2017-12-22 Extraction, blocking, compression and reconstruction method for hyperspectrum of plant Active CN108460777B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711404372.0A CN108460777B (en) 2017-12-22 2017-12-22 Extraction, blocking, compression and reconstruction method for hyperspectrum of plant

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711404372.0A CN108460777B (en) 2017-12-22 2017-12-22 Extraction, blocking, compression and reconstruction method for hyperspectrum of plant

Publications (2)

Publication Number Publication Date
CN108460777A true CN108460777A (en) 2018-08-28
CN108460777B CN108460777B (en) 2020-06-05

Family

ID=63221331

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711404372.0A Active CN108460777B (en) 2017-12-22 2017-12-22 Extraction, blocking, compression and reconstruction method for hyperspectrum of plant

Country Status (1)

Country Link
CN (1) CN108460777B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109522918A (en) * 2018-09-14 2019-03-26 广东工业大学 High spectrum image feature extracting method based on improved local singular spectrum analysis
CN110930466A (en) * 2019-10-22 2020-03-27 杭州电子科技大学 Hyperspectral adaptive compressive sensing method for BOIs (Bill of information) in any shapes

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103810755A (en) * 2014-03-04 2014-05-21 西安电子科技大学 Method for reconstructing compressively sensed spectral image based on structural clustering sparse representation
CN103903261A (en) * 2014-03-24 2014-07-02 西安电子科技大学 Spectrum image processing method based on partition compressed sensing
CN104779960A (en) * 2015-03-20 2015-07-15 南京邮电大学 A signal reconstruction method based on block compressed sensing
CN105513026A (en) * 2015-12-17 2016-04-20 浙江大学 Compressed sensing reconstruction method based on image nonlocal similarity
CN105869126A (en) * 2016-04-05 2016-08-17 天津大学 Compression perception denoising method of high-spectral images

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103810755A (en) * 2014-03-04 2014-05-21 西安电子科技大学 Method for reconstructing compressively sensed spectral image based on structural clustering sparse representation
CN103903261A (en) * 2014-03-24 2014-07-02 西安电子科技大学 Spectrum image processing method based on partition compressed sensing
CN104779960A (en) * 2015-03-20 2015-07-15 南京邮电大学 A signal reconstruction method based on block compressed sensing
CN105513026A (en) * 2015-12-17 2016-04-20 浙江大学 Compressed sensing reconstruction method based on image nonlocal similarity
CN105869126A (en) * 2016-04-05 2016-08-17 天津大学 Compression perception denoising method of high-spectral images

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
PING XU ET AL.: "Adaptive Grouping Distributed Compressive Sensing Reconstruction of Plant Hyperspectral Data", 《SENSORS》 *
佘青山 等: "一种新的分块压缩传感图像重建算法", 《东南大学学报（自然科学版）》 *
徐平 等: "面向植物生理生化参数反演的光谱信息压缩感知重构", 《农业工程学报》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109522918A (en) * 2018-09-14 2019-03-26 广东工业大学 High spectrum image feature extracting method based on improved local singular spectrum analysis
CN109522918B (en) * 2018-09-14 2020-05-05 广东工业大学 Hyperspectral image feature extraction method based on improved local singular spectrum analysis
CN110930466A (en) * 2019-10-22 2020-03-27 杭州电子科技大学 Hyperspectral adaptive compressive sensing method for BOIs (Bill of information) in any shapes
CN110930466B (en) * 2019-10-22 2023-08-29 杭州电子科技大学 Hyperspectral self-adaptive compressive sensing method for BOIs with arbitrary shape

Also Published As

Publication number Publication date
CN108460777B (en) 2020-06-05

Similar Documents

Publication Publication Date Title
Deng et al. Machine learning in pansharpening: A benchmark, from shallow to deep networks
Li et al. Remote sensing image fusion via sparse representations over learned dictionaries
CN110501072B (en) Reconstruction method of snapshot type spectral imaging system based on tensor low-rank constraint
CN107451956B (en) A kind of reconstructing method of code aperture spectrum imaging system
Li et al. Super-resolution for GaoFen-4 remote sensing images
CN112991288B (en) Hyperspectral remote sensing image fusion method based on abundance image sharpening reconstruction
CN114821261B (en) Image fusion method
CN108876884A (en) A kind of high spectrum image method for reconstructing based on non local tensor low-rank regularization
CN109191404A (en) A kind of high spectrum image restorative procedure based on E-3DTV canonical
CN109741407A (en) A kind of high quality reconstructing method of the spectrum imaging system based on convolutional neural networks
CN107966210B (en) Hyperspectral image-based thermal infrared fusion reconstruction method
CN107833182A (en) The infrared image super resolution ratio reconstruction method of feature based extraction
CN116703769B (en) Satellite remote sensing image full-color sharpening system
CN109087262B (en) Multi-view spectral image reconstruction method and storage medium
CN106157269A (en) Full-colour image sharpening method based on direction multiple dimensioned group low-rank decomposition
Li et al. Removal of optically thick clouds from high-resolution satellite imagery using dictionary group learning and interdictionary nonlocal joint sparse coding
CN108460777A (en) A kind of extraction splits&#39; positions reconstructing method towards plant EO-1 hyperion
CN103714534A (en) Material surface defect detection method based on compressed sensing
Chen et al. Attentive generative adversarial network for removing thin cloud from a single remote sensing image
CN102054267B (en) Enhancement method aiming at passive millimeter wave image
Liu et al. SAR image denoising based on patch ordering in nonsubsample shearlet domain
Wu et al. Neural interferometry: Image reconstruction from astronomical interferometers using transformer-conditioned neural fields
CN116202623A (en) Single photon hyperspectral imaging method and system based on spectrum compressed sampling
CN106570876A (en) Ghost imaging image edge extracting method
Xu et al. Multimodal and Multiresolution Data Fusion for High-Resolution Cloud Removal: A Novel Baseline and Benchmark

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