CN110717949A

CN110717949A - Interference hyperspectral image sparse reconstruction based on TROMP

Info

Publication number: CN110717949A
Application number: CN201810767130.6A
Authority: CN
Inventors: 温佳; 刘明威; 崔军; 闫淑霞
Original assignee: Tianjin Polytechnic University
Current assignee: Tianjin Polytechnic University
Priority date: 2018-07-11
Filing date: 2018-07-11
Publication date: 2020-01-21

Abstract

The invention relates to a compressed sensing-based interference hyperspectral image sparse reconstruction algorithm, which aims to achieve the aim of accurately reconstructing interference hyperspectrum. Each frame of interference hyperspectral data has vertical interference fringes with large amplitude and fixed positions, and background images have displacement between frames, so that the traditional algorithm is difficult to obtain an ideal sparse reconstruction effect. Because the vertical information of the interference fringes of the interference hyperspectral image is rich, the interference fringes in the image are extracted according to the horizontal full-variation value, and self-adaptive sampling is carried out. Due to the characteristics of interference of hyperspectral images, the ROMP algorithm selects the subset with the highest energy from the sequence number set, and occupies too much resources and time, and the maximum energy vector is quickly selected by adopting a relevant threshold value in the algorithm. Simulation results show that compared with the ROMP algorithm, the reconstruction precision of the algorithm is obviously improved under the condition of the same sparsity.

Description

Interference hyperspectral image sparse reconstruction based on TROMP

Technical Field

The invention belongs to the field of image processing, and relates to a compressed sensing image reconstruction algorithm improved by adopting an optimal threshold thought aiming at the characteristics of interference hyperspectrum.

Background

The compressed sensing technology is a valuable practical technology in the field of image processing, the signal acquisition speed can be greatly reduced on the premise that signals are compressible or sparse, the simultaneous signal sampling and compression are realized, the acquisition of a large amount of useless data is avoided, and resources and time are saved.

The existing algorithm has the following defects: the conventional ROMP algorithm has good reconstruction effect on the traditional image, but has poor reconstruction effect on interference fringes in an interference hyperspectral image. In the ROMP algorithm, the highest average energy of the index in the atom set occupies too much resources and time, and the number of atoms selected each time in the regularization process during secondary selection may be too large, which may cause the atoms with low correlation to be selected, thereby affecting the reconstruction quality. The selection with larger vector difference of each group of interference fringe part has worse pertinence, and the reconstruction effect is influenced

Disclosure of Invention

The invention aims to overcome the defects of the ROMP algorithm, adopts a related threshold value to perform secondary selection, simplifies a large amount of calculations in the regularization process compared with the ROMP algorithm, and can properly select a plurality of groups each time, thereby avoiding useless selection, accelerating algorithm convergence and improving reconstruction accuracy.

The technical scheme adopted by the invention is as follows:

an interference hyperspectral image sparse reconstruction algorithm based on compressed sensing is characterized in that: the algorithm comprises the following steps:

inputting: measurement matrix A observation vector y sparsity K

And (3) outputting: signal sparse approximation

(1) Initialization: the input image is subjected to horizontal full-variation classification and adaptive sampling

(2) Finding J { | λ { [ max { ] { [ lambda ]_j＝(r_t-1，A_j) I, K, solving the absolute value of the inner product of the measurement matrix and the residual error and indexing the largest K

(3) Taking a threshold value and indexing a factor greater than the threshold value

(4) Updating index collections

(5) Solving the least square solution to obtain a reconstructed vector

(6) Updating residual errors

(7) Judging the selected atom set A_tI > 2K or r_tξ, stopping iteration if true, otherwise returning to step (2)

r_tDenotes the residual, t denotes the number of iterations, J indexes (column numbers) found per iteration, A_tRepresenting the set of columns of matrix a selected by index J.

The invention has the advantages and positive effects that:

according to the method, the interference hyperspectral image and the interference fringe characteristics are adopted, and the interference hyperspectral image sparse reconstruction algorithm based on compressed sensing is deduced. The TROMP algorithm has good stability in the reconstruction process, and the traditional algorithm is unstable in reconstruction of the interference hyperspectral image, so that a good solution is provided for reconstruction of the interference hyperspectral image.

Detailed Description

The present invention is further illustrated by the following examples, which are intended to be illustrative, not limiting and are not intended to limit the scope of the invention.

A compressed sensing-based interference hyperspectral image sparse reconstruction algorithm is disclosed, a schematic diagram of a processing process is shown in figure 1, and the innovation of the invention is as follows:

1. screening interference fringe blocks by using horizontal total variation according to the image characteristics of interference hyperspectrum;

2. in the image reconstruction stage, self-adaptive sampling is adopted according to different image blocks, so that the reconstruction effect of the interference fringe block is further improved;

3. the TROMP algorithm is proposed herein according to the conventional ROMP algorithm. The algorithm simplifies the regularized computation complexity of the ROMP algorithm in secondary selection and improves the accuracy of secondary selection of related atoms.

Fig. 2 is an image used for simulation. It is obvious from fig. 3 and 4 that the invention is greatly improved compared with the conventional ROMP algorithm. Under the condition that the sampling rate is 0.35, the reconstruction effect of the invention on the interference hyperspectrum is better than that of the traditional ROMP algorithm. Even under the condition that the sampling rate of the ROM algorithm reaches 0.5, the method has great advantages and has good reconstruction effect on interference fringes in the interference hyperspectral image, and the traditional ROMP algorithm cannot be used. It can be seen more intuitively from tables 1 and 2 that the reconstruction effect of the present invention is greatly improved than that of the ROMP algorithm, but the reconstruction time is improved.

TABLE 1 SNR and reconstruction time contrast for three images under different algorithms with a sampling rate M/N of 0.35

TABLE 2 SNR and reconstruction time contrast for three images under different algorithms with a sampling rate M/N of 0.5

While the invention has been described in connection with specific embodiments thereof, it will be understood that these should not be construed as limiting the scope of the invention, which is defined in the following claims, and any variations which fall within the scope of the claims are intended to be embraced thereby.

Drawings

Fig. 1 is an algorithm flow chart, fig. 2 is a simulated original image, fig. 3 is a CS reconstructed image of Lasis01, and fig. 4 is a CS reconstructed image of Lasis 02.

Claims

1. An interference hyperspectral image sparse reconstruction algorithm based on compressed sensing is characterized in that: the algorithm comprises the following steps:

inputting: measurement matrix A observation vector y sparsity K

And (3) outputting: signal sparsenessApproximation

(1) Initialization: r is₀＝y，

t＝1

(2) To find

Solving the absolute value of the inner product of the measurement matrix and the residual error and putting the maximum K into J

(3) Selecting a threshold value, screening factors larger than the threshold value from J, and putting the factors into J₀

J₀＝{|λ_j＝<r_t-1，a_j>|≥G_t，j＝1，2，3，...，N}

(4) Update index set Λ_t＝Λ_t-1∪{J₀}，A_t＝A_t-1∪a_j

(5) Solving the least square solution to obtain a reconstructed vector

(6) Updating residual errors

(7) Judging the selected atom set | | | Λ_t||₀> 2K or r_tIf true, stopping iteration and outputting the last result

Otherwise, t is t +1 and returns to the step (2).

r_tDenotes the residual, t denotes the number of iterations, J denotes the index (column number) found per iteration, A_tRepresentation by index Λ_tThe selected column set of matrix a.

2. The sparse reconstruction algorithm of the interference hyperspectral image based on the compressed sensing according to claim 1, wherein the sparse reconstruction algorithm comprises the following steps: step (1), interference fringes are sorted according to horizontal total variation:

3. the sparse reconstruction algorithm of the interference hyperspectral image based on the compressed sensing according to claim 1, wherein the sparse reconstruction algorithm comprises the following steps: the step (1) is that the self-adaptive sampling is carried out on the image:

dividing the interference hyperspectrum into an interference fringe block and a common block according to the total variation value of the image, and respectively using M to sample₁And M₂(M₁＞M₂) Are sampled separately. The specific sampling data are as follows:

TABLE 1 adaptive sampling of each part of the data

4. The sparse reconstruction algorithm of the interference hyperspectral image based on the compressed sensing according to claim 1, wherein the sparse reconstruction algorithm comprises the following steps: the index threshold value in the step (3):

T＝0.6*max{|λ_j＝<r_t-1，A_j>|，K}

the algorithm adopts a correlation threshold value to replace secondary selection in the ROMP algorithm, and the correlation threshold value can ensure that the correlation of atoms selected each time is high enough and the atoms with the highest energy cannot be missed. When atoms are selected, a plurality of atoms can be properly selected each time to ensure enough cycle times, and the omission of atoms with higher matching degree in the follow-up process is avoided. Therefore, not only is the algorithm precision improved, but also the operation speed of each circulation is ensured.