CN111192197A - Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion - Google Patents

Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion Download PDF

Info

Publication number
CN111192197A
CN111192197A CN201911362493.2A CN201911362493A CN111192197A CN 111192197 A CN111192197 A CN 111192197A CN 201911362493 A CN201911362493 A CN 201911362493A CN 111192197 A CN111192197 A CN 111192197A
Authority
CN
China
Prior art keywords
lens
wave infrared
image
medium wave
infrared spectrum
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.)
Withdrawn
Application number
CN201911362493.2A
Other languages
Chinese (zh)
Inventor
秦翰林
姚迪
马琳
杨硕闻
乐阳
延翔
张嘉伟
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.)
Xidian University
Original Assignee
Xidian 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 Xidian University filed Critical Xidian University
Priority to CN201911362493.2A priority Critical patent/CN111192197A/en
Publication of CN111192197A publication Critical patent/CN111192197A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • 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
    • G06T2207/10048—Infrared 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20172—Image enhancement details
    • G06T2207/20192—Edge enhancement; Edge preservation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention discloses a medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion, which are used for splitting an original medium wave infrared spectrum image to obtain medium wave infrared spectrum images of different spectrum channels; converting the mid-wave infrared spectrum images of the different spectrum channels into near infrared spectrum images; coding the near infrared spectrum image through a coding template designed based on a random Gaussian matrix to obtain a spectrum image; from the spectral image by compressed sensing principles
Figure DDA0002337561190000011
Recovering an original medium wave infrared spectrum image; restoring original medium wave infrared spectrum image W through FSRCNN based on deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain high-quality near infrared spectrum image W with target medium wave characteristicsNP×NQ×l. Hair brushObviously, the deep learning-based algorithm is used for reconstructing and super-separating the spectrum, the transmission and storage pressure of data is reduced, and the spectrum image with higher resolution and signal-to-noise ratio can be obtained.

Description

Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion
Technical Field
The invention belongs to the field of multispectral imaging detection, and particularly relates to a medium wave infrared multispectral imaging method and device based on novel vortex light source wavelength conversion.
Background
The infrared imaging is an imaging mode which cannot be replaced by visible light imaging due to the advantages of strong anti-interference capability, strong penetrating power and the like, and the medium-wave infrared band is just positioned in an atmospheric window of 3-5 mu m, so that the infrared imaging has important significance in the fields of environmental monitoring, agriculture and the like.
At present, infrared imaging is mainly realized by using an infrared detector, but the infrared detector has the defects of low resolution, small area array size, high price, narrow spectrum band, lack of spectral information and the like, so that the conversion of medium-wave infrared light to a near-infrared spectrum band and the detection by using a silicon-based detector become a new choice.
Disclosure of Invention
In view of the above, the main objective of the present invention is to provide a medium wave infrared multispectral imaging method and device based on wavelength conversion of a novel vortex light source.
In order to achieve the purpose, the technical scheme of the invention is realized as follows:
the embodiment of the invention provides a medium wave infrared multispectral imaging method based on vortex light source wavelength conversion, which comprises the following steps:
splitting the original mid-wave infrared spectrum image to obtain mid-wave infrared spectrum images of different spectral channels;
converting the mid-wave infrared spectrum images of the different spectrum channels into near infrared spectrum images;
coding the near infrared spectrum image through a coding template designed based on a random Gaussian matrix to obtain a spectrum image;
from the spectral image by compressed sensing principles
Figure BDA0002337561170000011
Recovering an original medium wave infrared spectrum image;
restoring original medium wave infrared spectrum image W through FSRCNN based on deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain high-quality near infrared spectrum image W with target medium wave characteristicsNP×NQ×l。
In the above scheme, the splitting the original mid-wave infrared spectrum image to obtain mid-wave infrared spectrum images of different spectrum channels specifically includes: suppose an original mid-wave infrared spectrum image data cube VP×Q×LThe resolution is P multiplied by Q, and the number of spectral channels is L; will VP×Q×LMedium wave infrared spectrum image data cube V divided into different spectrum channelsP×Q×l(l＝1,2,...,L)
In the above scheme, the converting the original mid-wave infrared spectrum image data into near infrared spectrum data specifically comprises: will VP×Q×LMedium wave infrared spectrum image data cube V divided into different spectrum channelsP×Q×l(L ═ 1, 2.., L) is converted into a near infrared spectral data cube WP×Q×l(l＝1,2,...,L)。
In the above scheme, the encoding of the near infrared spectrum image by the encoding template designed based on the random gaussian matrix to obtain the spectrum image specifically includes: when the light beam passes through the coding template, a random matrix g with the size of P multiplied by Q completes coding on the near infrared spectrum cube
Figure BDA0002337561170000021
The detector collects the spectrum image after compression sampling
Figure BDA0002337561170000022
A compressed sampling ratio of
Figure BDA0002337561170000023
In the scheme, the restored original mid-wave infrared spectrum image W is subjected to the FSRCNN based on the deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain a hyper-resolution image WNP×NQ×lThe method is realized by the following steps:
(4a) directly performing convolution operation on the low-resolution image W, performing convolution kernel of 5 multiplied by 5, and extracting features from the low-resolution image W;
(4b) reducing dimension by convolution operation with convolution kernel of 1 × 1;
(4c) in the FSRCNN network, M convolution layers with kernel size of 3 × 3 are connected in series, and one convolution kernel of 5 × 5 is replaced by two series connected convolution kernels of 3 × 3, and the parameter 3 × 3 × 2 ═ 18 required by the two series connected convolution kernels is smaller than the parameter 5 × 5 ═ 25 required by one convolution kernel of 5 × 5;
(4d) performing dimension expansion by using a convolution kernel of 1 multiplied by 1;
(4e) performing up-sampling operation, namely amplifying the size of the image, wherein the step length is N, namely amplifying the size by N times;
(4f) the activation function adopts a PReLU function, which can be expressed as
Figure BDA0002337561170000031
i denotes the different channels;
a loss function of
Figure BDA0002337561170000032
The embodiment of the invention also provides a medium wave infrared multispectral imaging device based on novel vortex light source wavelength conversion, which comprises a collimating lens, an acousto-optic tunable filter, a pumping light source, a spiral phase plate, a first 4-f optical system, a first lens, a nonlinear crystal, a second 4-f optical system, a coding template and a silicon-based detector, wherein infrared light enters the acousto-optic tunable filter after being converted into parallel light beams by the collimating lens, and medium wave infrared light is emitted to the first lens by the acousto-optic tunable filter;
pump laser generated by the pump light source is emitted to a first lens through a spiral phase plate and a first 4-f optical system;
the medium wave infrared light and the pump laser are converted into light beams with a common optical axis through a first lens and vertically enter the nonlinear crystal;
the nonlinear crystal 7 converts the medium wave infrared into near infrared light through a nonlinear effect, and the near infrared light is converted by a second 4-f optical system and a coding template and then emitted to a silicon-based detector;
the silicon-based detector collects the coded near-infrared spectral image, recovers the spectral image with the real size through a compressed sensing algorithm, and performs spatial super-resolution on the basis of a super-resolution network (FSRCNN) of a convolutional neural network to obtain a high-quality near-infrared spectral image with target medium wave characteristics.
In the foregoing solution, the first 4-f optical system includes a second lens and a third lens, focal lengths of the second lens and the third lens are equal, the spiral phase plate is used as an object plane of the first 4-f optical system, the first lens is used as an image plane of the first 4-f optical system, distances between the spiral phase plate and the second lens are equal to distances therebetween, which are focal lengths of the second lens that are one time, and a distance between the third lens and the first lens is a focal length of the second lens that is twice as long as the distance between the third lens and the first lens.
In the scheme, the first lens is plated with an antireflection film in an infrared medium wave band and a high reflection film at the wavelength of the vortex pump laser.
In the above scheme, the second 4-f optical system includes a fourth lens and a fifth lens that are identical, the output surface of the nonlinear crystal 7 serves as the object plane of the second 4-f optical system, the receiving surface of the silicon-based detector serves as the image plane of the second 4-f optical system, and the encoding template is placed at the position of the fourier plane of the second 4-f optical system, that is, at the position of the focal planes of the fourth lens and the fifth lens.
Compared with the prior art, the method has the advantages that the medium wave infrared light is converted into the near infrared light by utilizing the wavelength conversion based on the vortex light source, the outline characteristics of the image are enhanced, the cost of infrared imaging is reduced, the spectrum is reconstructed and overdivided based on the deep learning algorithm, the data transmission and storage pressure is reduced, the spatial correlation and the inter-spectrum correlation of the spectrum image can be better extracted by the neural network based on the deep learning, and the spectrum image with higher resolution and signal-to-noise ratio is obtained.
Drawings
FIG. 1 is a flowchart of a method for mid-wave infrared multispectral imaging based on vortex light source wavelength conversion according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a network structure of FSRCNN in a mid-wave infrared multispectral imaging method based on vortex light source wavelength conversion according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a medium-wave infrared multispectral imaging device based on vortex light source wavelength conversion according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an acousto-optic tunable filter in a mid-wave infrared multispectral imaging device based on vortex light source wavelength conversion according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The embodiment of the invention provides a medium wave infrared multispectral imaging method based on vortex light source wavelength conversion, which is realized by the following steps as shown in figure 1:
step 101: splitting the original mid-wave infrared spectrum image to obtain mid-wave infrared spectrum images of different spectral channels;
specifically, an original mid-wave infrared spectrum image data cube VP×Q×LThe resolution is P multiplied by Q, and the number of spectral channels is L; will VP×Q×LMedium wave infrared spectrum image data divided into different spectrum channelsCube VP×Q×l(l＝1,2,...,L)。
Step 102: converting the mid-wave infrared spectrum images of the different spectral channels into near infrared spectrum images;
specifically, V isP×Q×LMedium wave infrared spectrum image data cube V divided into different spectrum channelsP×Q×l(L ═ 1, 2.., L) is converted into a near infrared spectral data cube WP×Q×l(l＝1,2,...,L)。
Step 103: coding the near infrared spectrum image through a coding template designed based on a random Gaussian matrix to obtain a spectrum image;
specifically, when the light beam passes through the coding template, a random matrix g with the size of P multiplied by Q codes the near infrared spectrum cube, and the random matrix g with the size of P multiplied by Q passes through the coding template
Figure BDA0002337561170000051
The detector collects the spectrum image after compression sampling
Figure BDA0002337561170000052
A compressed sampling ratio of
Figure BDA0002337561170000053
Step 104: from the spectral image by compressed sensing principles
Figure BDA0002337561170000054
Recovering an original medium wave infrared spectrum image;
specifically, as known from the compressed sensing principle, for the sparse signal x in some sparse basis ψ, the estimation coefficient can be obtained by solving the minimization problem
Figure BDA0002337561170000055
Figure BDA0002337561170000056
Wherein
Figure BDA0002337561170000057
Is the measurement result, phi is the measurement matrix, phi psi is the sensing matrix, by finding the minimum l from the coefficient vector theta1Of norm
Figure BDA0002337561170000058
The original image is constructed using x ═ Ψ θ.
In the actual measurement process, the influence of various errors on the imaging model needs to be considered, and in this case, the above formula can be rewritten into
Figure BDA0002337561170000059
The coding template is designed based on a Hadamard matrix; the hadamard matrix is an orthogonal square matrix composed of +1 and-1 elements. By orthogonal square matrix, it is meant that any two rows (or two columns) thereof are orthogonal. Considering the rows (or columns) as a function, any two rows (or columns) are orthogonal and satisfy Hn 'nI (where Hn' is the transpose of Hn and I is the unit square) nth order square.
The invention adopts a complementary code space matching pursuit (CMP) recovery algorithm to realize the recovery of the compressed spectrum image. If phi ∈ RM×NAre independent of each other, i.e. phiTAre linearly independent. Then phiTIs an M-dimensional subspace, defined as S, and S is an N-dimensional subspace RNOf (2) is provided. Considering the original problem y ═ Φ x, the least squares solution is obtained:
xls＝ΦT(ΦΦT)-1y
wherein x islsE.g., S, assume x*Is a sparse solution required by the present invention, then x*Can be expressed as
x*＝xls+xc
Wherein xcIs a non-zero vector and satisfies Φ xc0, i.e. xcBelonging to the orthogonal complement space of S. The orthogonal complement spatial notation is Sc.
Suppose y requires only a linear representation of k elements, i.e. xThe only k elements are non-zero. The other N-k elements are all zero and are marked as
Figure BDA0002337561170000061
In the CMP algorithm, x*Is through x*＝xls+xcTo solve it.
In fact, equation x is knownls＝ΦT(ΦΦT)-1y and
Figure BDA0002337561170000062
can obtain the product
Figure BDA0002337561170000063
Therefore, the key to solving the equation is to find the k-order element xc[k]。
Assuming that the size of G is N x (N-M) and that the N-M columns are independent of each other, the column vectors constitute an N-M dimensional subspace. Any N-1 column is an N-M dimensional subspace. By QR decomposition of phiTG is obtained. Knowing xc∈ScSo xcA linear combination of column vectors that can be represented as G. If the coefficient is an N-M dimensional vector zcObtaining xc＝GzcTherefore, it is
Figure BDA0002337561170000064
Is equivalent to
Figure BDA0002337561170000065
Due to the fact that
Figure BDA0002337561170000066
And
Figure BDA0002337561170000067
we obtain by least squares
Figure BDA0002337561170000068
Simultaneous upper formula and xc＝GzcTo obtain
Figure BDA0002337561170000069
Push out
Figure BDA0002337561170000071
Where G is the kth row of matrix G. Simultaneous equations and equations x*＝xls+xcTo obtain
Figure BDA0002337561170000072
From this, a coefficient solution x is obtained*。
In the derivation of CMP, first find the satisfaction
Figure BDA0002337561170000073
N-1 elements of (i), i.e.
Figure BDA0002337561170000074
In actual calculations, perhaps N-1 elements are not all zero, but are close to zero. Thus, in CMP, the method of selecting elements during each iteration is
Figure BDA0002337561170000075
Suppose that
Figure BDA0002337561170000076
And
Figure BDA0002337561170000077
we select elements using the following equation
Figure BDA0002337561170000078
System of sparsenessNumber is
Figure BDA0002337561170000079
In the whole process, the only useful element is gk. Through gkThe desired non-zero elements can be obtained.
Consider a more general form of iteration. In the process of the t-th iteration, if the last iteration margin is rt-1To select useful elements, equations need to be solved
Figure BDA00023375611700000710
The basis of the calculation is
Figure BDA00023375611700000711
Selecting matching elements, i.e. the k-th line G in Gk. The corresponding coefficient is
Figure BDA00023375611700000712
Calculating the margin of
Figure BDA00023375611700000713
Finally, the approximation error after the t-th iteration
Figure BDA00023375611700000714
Step 105: restoring original medium wave infrared spectrum image W through FSRCNN based on deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain high-quality near infrared spectrum image W with target medium wave characteristicsNP×NQ×l。
Specifically, (4a) feature extraction, as shown in fig. 2. Directly, the convolution operation is carried out on the low-resolution image W, the convolution kernel is 5 multiplied by 5, and features are extracted from the low-resolution image W.
(4b) And (4) compressing. In order to reduce the parameters of the network and the computational complexity and dimension reduction, the dimension reduction is carried out by using convolution operation, and the convolution kernel is 1 multiplied by 1.
(4c) And (4) nonlinear mapping. The FSRCNN network is connected in series by M convolutional layers with the core size of 3 multiplied by 3. Two series-connected, 3 × 3 convolution kernels replace one 5 × 5 convolution kernel, and the parameters 3 × 3 × 2 ═ 18 required for two series-connected small convolution kernels are smaller than those for one large convolution kernel 5 × 5 ═ 25.
(4d) Expansion, the inverse of compression. Dimension expansion was performed with a 1 × 1 convolution kernel.
(4e) And (4) deconvoluting. And performing an upsampling operation to enlarge the image size. The step size is N, i.e. the size is enlarged N times.
(4f) The activation function adopts a PReLU function. The PReLU function can be expressed as
Figure BDA0002337561170000081
i denotes the different channels.
A loss function of
Figure BDA0002337561170000082
The embodiment of the invention also provides a medium wave infrared multispectral imaging device based on the novel vortex light source wavelength conversion, which comprises a collimating lens 1, an acousto-optic tunable filter 2, a pumping light source 3, a spiral phase plate 4, a first 4-f optical system 5, a first lens 6, a nonlinear crystal 7, a second 4-f optical system 8, an encoding template 9 and a silicon-based detector 10 as shown in figure 3,
after being converted into parallel beams by the collimating lens 1, infrared light enters the acousto-optic tunable filter 2 and is emitted to the first lens 6 through the acousto-optic tunable filter 2;
the pump laser generated by the pump light source 3 is emitted to a first lens 6 through a spiral phase plate 4 and a first 4-f optical system 5;
the medium wave infrared light and the pump laser are converted into light beams with a common optical axis through a first lens 6 and vertically enter a nonlinear crystal 7;
the nonlinear crystal 7 converts the medium wave infrared into near infrared light through a nonlinear effect, and the near infrared light is converted by a second 4-f optical system 8 and an encoding template 9 and then emitted to a silicon-based detector 10;
the silicon-based detector 10 collects the coded near-infrared spectral image, recovers the spectral image with the real size through a compressed sensing algorithm, and performs spatial super-resolution on the basis of a super-resolution network (FSRCNN) of a convolutional neural network to obtain a high-quality near-infrared spectral image with target medium wave characteristics.
As shown in fig. 4, the acousto-optic tunable filter 2 for splitting the original infrared light signal has an optical aperture of 10 × 10mm, a light-passing wavelength band of 3.0-5.0 μm, and a peak diffraction wavelength of 30%, when the original infrared light signal enters the AOTF, the infrared light is first converted into parallel beams through the collimating lens 1 to enter, and two beams of light are emitted from the AOTF by selecting an appropriate piezoelectric frequency: the polarization directions of the + 1-level light and the-1-level light are mutually vertical, and the propagation directions form a certain included angle. In the invention, one light beam is filtered by using a polaroid, so that only one light beam in the diffracted light beams enters the first lens 6 in parallel.
According to the invention, the vortex pump laser is used by the pump light source 3 to obtain laser field distribution containing optical vortex through a Spiral Phase Plate (SPP), and the profile characteristics of an image can be enhanced to a certain extent by using the spiral pump laser to perform wavelength conversion.
The first 4-f optical system 5 comprises a second lens 51 and a third lens 52, focal lengths of the second lens 51 and the third lens 52 are equal, the spiral phase plate 4 serves as an object plane of the first 4-f optical system 5, the first lens 6 serves as an image plane of the first 4-f optical system 5, a distance between the spiral phase plate 4 and the second lens 51 is equal to a distance between the spiral phase plate 4 and the second lens 51, the focal lengths of the second lens 51 are doubled, and a distance between the third lens 52 and the first lens 6 is doubled by the focal length of the second lens 51.
And the first lens 6 is plated with an antireflection film in an infrared medium wave band and a high reflection film for the wavelength of vortex pump laser.
In order to realize high-efficiency wavelength conversion, the invention uses a periodically poled nonlinear crystal such as a periodically poled lithium niobate crystal (pp-LN), because the reciprocal lattice vector provided by the nonlinear periodic structure can easily realize phase matching, and can effectively realize nonlinear frequency conversion, and in order to realize broadband coverage of wavelength conversion, the invention can be realized by designing a crystal chirp structure to design different poling periods for different wavelengths.
The second 4-f optical system 8 comprises a fourth lens 81 and a fifth lens 82 which are identical, the output surface of the nonlinear crystal 7 serves as the object plane of the second 4-f optical system 8, the receiving surface of the silicon-based detector 10 serves as the image plane of the second 4-f optical system 8, and the encoding template 9 is placed at the position of the Fourier plane of the second 4-f optical system 8, namely the position of the focal plane of the fourth lens 81 and the fifth lens 82.
The coding template 9 is designed based on a Hadamard matrix, and only has two code elements of +1 and-1. And (4) the mid-wave infrared spectrum of the target passes through the coding template to complete the calculation coding of the target spectrum.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention.

Claims (9)

1. A medium wave infrared multispectral imaging method based on vortex light source wavelength conversion is characterized by comprising the following steps:
splitting the original mid-wave infrared spectrum image to obtain mid-wave infrared spectrum images of different spectral channels;
converting the mid-wave infrared spectrum images of the different spectrum channels into near infrared spectrum images;
coding the near infrared spectrum image through a coding template designed based on a random Gaussian matrix to obtain a spectrum image;
from the spectral image by compressed sensing principles
Figure FDA0002337561160000011
Recovering an original medium wave infrared spectrum image;
restoring original medium wave infrared spectrum image W through FSRCNN based on deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain high-quality near infrared spectrum image W with target medium wave characteristicsNP×NQ×l。
2. The mid-wave infrared multispectral imaging method based on vortex light source wavelength conversion according to claim 1, wherein the splitting of the original mid-wave infrared spectrum image to obtain mid-wave infrared spectrum images of different spectral channels specifically comprises: suppose an original mid-wave infrared spectrum image data cube VP×Q×LThe resolution is P multiplied by Q, and the number of spectral channels is L; will VP×Q×LMedium wave infrared spectrum image data cube V divided into different spectrum channelsP×Q×l(l＝1,2,...,L) 。
3. The mid-wave infrared multispectral imaging method based on vortex light source wavelength conversion according to claim 1 or 2, wherein the converting of the original mid-wave infrared spectral image data into near-infrared spectral data comprises: will VP×Q×LMedium wave infrared spectrum image data cube V divided into different spectrum channelsP×Q×l(L ═ 1, 2.., L) is converted into a near infrared spectral data cube WP×Q×l(l＝1,2,...,L)。
4. The mid-wave infrared multispectral imaging method based on vortex light source wavelength conversion, according to claim 3, wherein the spectral image is obtained by encoding the near-infrared spectral image through an encoding template designed based on a random Gaussian matrix, specifically: when the light beam passes through the coding template, a random matrix g with the size of P multiplied by Q completes coding on the near infrared spectrum cube
Figure FDA0002337561160000012
The detector collects the spectrum image after compression sampling
Figure FDA0002337561160000021
A compressed sampling ratio of
Figure FDA0002337561160000022
5. Root of herbaceous plantThe method according to claim 4, wherein the restored original mid-wave infrared spectrum image W is subjected to FSRCNN based on deep learningP×Q×lPerforming hyper-resolution reconstruction to obtain a hyper-resolution image WNP×NQ×lThe method is realized by the following steps:
(4a) directly performing convolution operation on the low-resolution image W, performing convolution kernel of 5 multiplied by 5, and extracting features from the low-resolution image W;
(4b) reducing dimension by convolution operation with convolution kernel of 1 × 1;
(4c) in the FSRCNN network, M convolution layers with kernel size of 3 × 3 are connected in series, and one convolution kernel of 5 × 5 is replaced by two series connected convolution kernels of 3 × 3, and the parameter 3 × 3 × 2 ═ 18 required by the two series connected convolution kernels is smaller than the parameter 5 × 5 ═ 25 required by one convolution kernel of 5 × 5;
(4d) performing dimension expansion by using a convolution kernel of 1 multiplied by 1;
(4e) performing up-sampling operation, namely amplifying the size of the image, wherein the step length is N, namely amplifying the size by N times;
(4f) the activation function adopts a PReLU function, which can be expressed as
Figure FDA0002337561160000023
i denotes the different channels;
a loss function of
Figure FDA0002337561160000024
6. A mid-wave infrared multispectral imaging device based on novel vortex light source wavelength conversion is characterized by comprising a collimating lens, an acousto-optic tunable filter, a pumping light source, a spiral phase plate, a first 4-f optical system, a first lens, a nonlinear crystal, a second 4-f optical system, an encoding template and a silicon-based detector,
after the infrared light is converted into parallel light beams through the collimating lens, the parallel light beams enter the acousto-optic tunable filter, and medium wave infrared light is emitted to the first lens through the acousto-optic tunable filter;
pump laser generated by the pump light source is emitted to a first lens through a spiral phase plate and a first 4-f optical system;
the medium wave infrared light and the pump laser are converted into light beams with a common optical axis through a first lens and vertically enter the nonlinear crystal;
the nonlinear crystal 7 converts the medium wave infrared into near infrared light through a nonlinear effect, and the near infrared light is converted by a second 4-f optical system and a coding template and then emitted to a silicon-based detector;
the silicon-based detector collects the coded near-infrared spectral image, recovers the spectral image with the real size through a compressed sensing algorithm, and performs spatial super-resolution on the basis of a super-resolution network (FSRCNN) of a convolutional neural network to obtain a high-quality near-infrared spectral image with target medium wave characteristics.
7. The novel vortex light source wavelength conversion based medium wave infrared multispectral imaging device according to claim 6, wherein the first 4-f optical system comprises a second lens and a third lens, the focal lengths of the second lens and the third lens are equal, the spiral phase plate is used as an object plane of the first 4-f optical system, the first lens is used as an image plane of the first 4-f optical system, the distance between the spiral phase plate and the second lens is equal to the distance between the spiral phase plate and the second lens, the focal lengths of the second lens are doubled, and the distance between the third lens and the first lens is doubled as the focal length of the second lens.
8. The novel vortex light source wavelength conversion based medium wave infrared multispectral imaging device according to claim 7, wherein the first lens is coated with an antireflection film in the infrared medium wave band and a high reflection film for the wavelength of the vortex pump laser.
9. The mid-wave infrared multispectral imaging device based on wavelength conversion of a novel vortex light source as claimed in any one of claims 6 to 8, wherein the second 4-f optical system comprises a fourth lens and a fifth lens which are identical, the output surface of the nonlinear crystal serves as the object plane of the second 4-f optical system, the silicon-based detector receiving surface serves as the image plane of the second 4-f optical system, and the coding template is placed at the position of the Fourier plane of the second 4-f optical system, namely the position of the focal plane of the fourth lens and the position of the focal plane of the fifth lens.
CN201911362493.2A 2019-12-26 2019-12-26 Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion Withdrawn CN111192197A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911362493.2A CN111192197A (en) 2019-12-26 2019-12-26 Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911362493.2A CN111192197A (en) 2019-12-26 2019-12-26 Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion

Publications (1)

Publication Number Publication Date
CN111192197A true CN111192197A (en) 2020-05-22

Family

ID=70707597

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911362493.2A Withdrawn CN111192197A (en) 2019-12-26 2019-12-26 Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion

Country Status (1)

Country Link
CN (1) CN111192197A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113363793A (en) * 2021-05-17 2021-09-07 电子科技大学 Random laser generation, spectrum synchronization and code sharing method
CN115327771A (en) * 2022-08-15 2022-11-11 华侨大学 Device and method for compressed sensing of vortex optical information

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140133011A1 (en) * 2011-07-08 2014-05-15 Daniel Schwedt Confocal incident-light scanning microscope
WO2018007572A1 (en) * 2016-07-08 2018-01-11 Danmarks Tekniske Universitet Infrared upconversion spectrometer for the mid-ir range
CN109447891A (en) * 2019-01-09 2019-03-08 北京理工大学 A kind of high quality imaging method of the spectrum imaging system based on convolutional neural networks

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140133011A1 (en) * 2011-07-08 2014-05-15 Daniel Schwedt Confocal incident-light scanning microscope
WO2018007572A1 (en) * 2016-07-08 2018-01-11 Danmarks Tekniske Universitet Infrared upconversion spectrometer for the mid-ir range
US20190242747A1 (en) * 2016-07-08 2019-08-08 Danmarks Tekniske Universitet Infrared upconversion spectrometer for the mid-ir range
CN109447891A (en) * 2019-01-09 2019-03-08 北京理工大学 A kind of high quality imaging method of the spectrum imaging system based on convolutional neural networks

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
GAGAN RATH: "Complementary Matching Pursuit Algorithms for Sparse", 《INSTITUTE OF ELECTRICAL AND ELECTRONICS ENGINEERS》 *
李聪: "压缩感知原信号重构算法研究", 《中国优秀硕士学位论文全文数据库》 *
王晓蕊: "《光电成像系统-建模、仿真、测试与评估》", 31 October 2017 *
董超: "Accelerating the Super-Resolution Convolutional Neural Network", 《EUROPEAN CONFERENCE ON COMPUTER VISION》 *
迪可新等: "基于压缩感知的空间稀疏目标成像方法研究", 《微电子学与计算机》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113363793A (en) * 2021-05-17 2021-09-07 电子科技大学 Random laser generation, spectrum synchronization and code sharing method
CN113363793B (en) * 2021-05-17 2022-11-25 电子科技大学 Random laser generation, spectrum synchronization and code sharing method
CN115327771A (en) * 2022-08-15 2022-11-11 华侨大学 Device and method for compressed sensing of vortex optical information

Similar Documents

Publication Publication Date Title
CN111157114B (en) Long-wave infrared multispectral imaging method and device based on wavelength conversion
Yang et al. Compressive hyperspectral imaging via sparse tensor and nonlinear compressed sensing
Chen et al. Hyperspectral image compressive sensing reconstruction using subspace-based nonlocal tensor ring decomposition
Tan et al. Compressive hyperspectral imaging via approximate message passing
CN103575396B (en) The imaging spectrum system converted based on compressed sensing and Hadamard and method
CN108955882B (en) Three-dimensional data reconstruction method based on liquid crystal hyperspectral calculation imaging system
CN103632385B (en) Satellite EO-1 hyperion compressed sensing method for reconstructing based on sky spectrum joint sparse priori
Han et al. Tensor FISTA-Net for real-time snapshot compressive imaging
CN204788661U (en) Calculate many spectral imaging system based on compressed sensing
Arguello et al. Spatial super-resolution in code aperture spectral imaging
Liu et al. Fourth-order tensors with multidimensional discrete transforms
CN111192197A (en) Medium wave infrared multispectral imaging method and device based on vortex light source wavelength conversion
CN113008370B (en) Three-dimensional self-adaptive compression reconstruction method based on liquid crystal hyperspectral calculation imaging system
CN115561182A (en) Priori image guidance-based snapshot type spectral imaging system reconstruction method
CN109682476B (en) Method for performing compressed hyperspectral imaging by using self-adaptive coding aperture
Wen et al. DeepCASD: An end-to-end approach for multi-spectral image super-resolution
Lin et al. All-addition hyperspectral compressed sensing for metasurface-driven miniaturized satellite
Degraux et al. Multispectral compressive imaging strategies using Fabry–Pérot filtered sensors
De Weerdt et al. Designing Transformer networks for sparse recovery of sequential data using deep unfolding
Salman et al. Orthonets: Orthogonal channel attention networks
CN107886555B (en) Distributed image reconstruction method for radio interference array
Tsagkatakis et al. Non-negative matrix completion for the enhancement of snapshot mosaic multispectral imagery
Hunt Prospects for image restoration
Ghosh et al. Use of EM algorithm for data reduction under sparsity assumption
Comon et al. Sparse representations and low-rank tensor approximation

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20200522