CN112525851A - Terahertz single-pixel imaging method and system - Google Patents
Terahertz single-pixel imaging method and system Download PDFInfo
- Publication number
- CN112525851A CN112525851A CN202011436254.XA CN202011436254A CN112525851A CN 112525851 A CN112525851 A CN 112525851A CN 202011436254 A CN202011436254 A CN 202011436254A CN 112525851 A CN112525851 A CN 112525851A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- mask
- coding
- terahertz
- 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.)
- Pending
Links
- 238000003384 imaging method Methods 0.000 title claims abstract description 89
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000013507 mapping Methods 0.000 claims description 75
- 230000006870 function Effects 0.000 claims description 46
- 238000012545 processing Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 8
- 238000005070 sampling Methods 0.000 abstract description 4
- 238000013527 convolutional neural network Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 3
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 2
- XEBWQGVWTUSTLN-UHFFFAOYSA-M phenylmercury acetate Chemical group CC(=O)O[Hg]C1=CC=CC=C1 XEBWQGVWTUSTLN-UHFFFAOYSA-M 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 229910052710 silicon Inorganic materials 0.000 description 2
- 239000010703 silicon Substances 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005670 electromagnetic radiation Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- AMGQUBHHOARCQH-UHFFFAOYSA-N indium;oxotin Chemical compound [In].[Sn]=O AMGQUBHHOARCQH-UHFFFAOYSA-N 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000000123 paper Substances 0.000 description 1
- 229920003023 plastic Polymers 0.000 description 1
- 239000004033 plastic Substances 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005057 refrigeration Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V8/00—Prospecting or detecting by optical means
- G01V8/10—Detecting, e.g. by using light barriers
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Physics & Mathematics (AREA)
- Toxicology (AREA)
- Geophysics (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a terahertz single-pixel imaging method and a terahertz single-pixel imaging system. The method comprises the following steps: the terahertz waves are modulated through a mask, and a reconstructed image is finally obtained. The mask is obtained by the following steps: acquiring an original image, and carrying out image coding on the original image through a preset mask to acquire a preset coding result; inputting a preset coding result into a self-coding network to obtain a preset decoding image; constructing a minimum difference network according to a preset decoded image and an original image, optimizing the minimum difference network through iterative learning, and judging whether the minimum difference network is converged: if the convergence happens, outputting the self-coding network parameters, and if the convergence does not happen, re-learning; and acquiring the mask according to the self-coding network parameters. According to the terahertz single-pixel imaging method, the problem of contradiction between the imaging time and the imaging quality of terahertz single-pixel imaging is solved through a deep learning network, and high-quality reconstruction of images under a low sampling ratio is realized while the imaging time is shortened.
Description
Technical Field
The invention relates to the field of terahertz single-pixel imaging, in particular to a terahertz single-pixel imaging method and a terahertz single-pixel imaging system.
Background
Terahertz waves (frequency 1-10THz or wavelength 30 μm-3000 μm) refer to coherent electromagnetic radiation located between the microwave and optical bands. Terahertz waves can penetrate various optically opaque materials such as semiconductors, plastics, paper and the like, and objects hidden in the terahertz waves, such as metal, internal structures of objects covered by external materials and the like, are identified with high contrast. In addition, terahertz photons are of very low energy, making such radiation safe for use by both humans and sensitive samples. Despite showing huge application potential, the terahertz imaging technology has not been put to practical use, the most important reason being the high equipment cost. Unlike the visible light field, due to the lack of suitable materials, the development of terahertz pixelated detector arrays is slow. At present, most of multi-pixel terahertz detector arrays are narrow-band or need to work in a low-temperature refrigeration environment, and practical popularization of the terahertz imaging technology is greatly restricted.
In recent years, terahertz spatial light modulators have been developed greatly, and particularly, the combined use of a single-element detector and a spatial coding mask opens a door for terahertz imaging technology. In this imaging mode, a beam of light is encoded through a time-varying spatial mask and the transmission (or reflection) of the object is recorded with a single-element detector, and then the knowledge of the spatial mask and detector values is combined to reconstruct an image of the object. The performance of the spatial light modulator directly determines the signal-to-noise ratio and the resolution of the terahertz single-pixel image.
The scholars propose that the 180nm VO2 film and Hadamard mask coding are combined to realize terahertz super-resolution image reconstruction, and the compressed sensing algorithm is used for realizing image reconstruction under a 20% modulation mask, but the compressed sensing algorithm is high in complexity and large in calculation amount, so that the whole imaging time is long. The reconstruction of sub-wavelength terahertz images is realized by using a 220um silicon-based graphene modulator and Fourier stripes, the reconstruction of images under a 10% modulation mask is realized by using the characteristics of image sparsity and acquisition of low-frequency coefficients and inverse Fourier transform, but the images are fuzzy due to the lack of high-frequency information. In the prior art, a nearly real-time terahertz single-pixel video is realized by using a silicon total internal reflection prism and a Hadamard mask, and the sampling time is reduced by using an undersampling technology by using the Hadamard domain sparse characteristic, but the problem of image reconstruction deterioration also exists.
In summary, in the prior art, for the problem that the imaging speed and the imaging quality are inconsistent commonly existing in terahertz single-pixel imaging, the relationship between the imaging speed and the imaging quality cannot be balanced. Therefore, a terahertz single-pixel imaging method is needed, which can ensure the image quality while shortening the imaging time.
Disclosure of Invention
Based on the problems in the prior art, the invention provides a terahertz single-pixel imaging method and a terahertz single-pixel imaging system. The specific scheme is as follows:
a terahertz single-pixel imaging method specifically comprises the following steps: s1, modulating terahertz waves through a mask; s2, enabling the modulated terahertz waves to interact with the target object, and receiving a formed modulation signal by a detector; s3, acquiring a reconstructed image of the target image according to the modulation signal; the mask in S1 is obtained by: acquiring an original image, and carrying out image coding on the original image through a preset mask to acquire a preset coding result; inputting the preset coding result into a self-coding network to obtain a preset decoding image; constructing a difference minimum network according to the preset decoding image and the original image, wherein the difference minimum network expression is as follows:
min‖X-X1‖2
wherein X is an original image, X1Decoding the image for presetting;
optimizing the minimum difference network through iterative learning, and judging whether the minimum difference network converges: if the minimum difference network is not converged, adjusting the self-coding network by adjusting the self-coding network parameters, re-acquiring a decoded image according to the adjusted self-coding network, acquiring a new minimum difference network, and judging whether the new minimum difference network is converged, so as to circulate until the self-coding network is converged; if the difference minimum network is converged, outputting self-coding network parameters when the difference minimum network is converged; and acquiring the mask according to the self-coding network parameters.
The preset mask comprises a Fourier mask, a Hadamard mask or a wavelet mask;
wherein the self-encoding network comprises a full-link filter, an S convolution filter, an S/2 convolution filter and a convolution filter.
In particular, after the "acquiring an original image" and before the "image coding the original image through a preset mask", the method further includes: and selecting an image database to perform graying processing and normalization processing on the original image.
Specifically, after the S3, the method further includes performing mapping processing on the reconstructed image acquired at S3 through a mapping network to acquire a high-resolution reconstructed image.
More specifically, the obtaining manner of the mapping network specifically includes: acquiring a low-resolution image L and a high-resolution image, and inputting the low-resolution image and the high-resolution image into the mapping network for learning; performing feature block processing on the low-resolution image through the mapping network to obtain a first feature; carrying out nonlinear mapping on the low-resolution image through the mapping network to obtain a second characteristic; constructing a loss function network according to the first characteristic and the second characteristic; training the mapping network, and judging whether the loss function network is converged: if the loss function network is converged, acquiring a trained mapping network; and if the loss function network is not converged, training the mapping network again, acquiring the first characteristic and the second characteristic again, acquiring a new loss function network, judging whether the new loss function network is converged, and circulating until the loss function network is converged.
Further, the expression of the first feature is:
F1(L)=max(0;W1×L+B1)
wherein, F1(L) is a first feature, W1As a characteristic filter, B1Is an offset parameter;
the expression of the second characteristic is:
F2(L)=max(0;W2×F1(L)+B2)
wherein, F2(L) is a second feature, F1(L) is a first feature, W2For high-resolution feature filters, B2As an offset parameter
The expression of the loss function network is:
min‖F2(L)-H‖2
wherein, F2(L) is a second feature, and H is a high resolution image.
Further, the "acquiring a low resolution image and a high resolution image" specifically includes: acquiring a sample image, carrying out graying processing and normalization processing on the sample image, and filtering the sample image by using a frequency domain to obtain the low-resolution image; taking the sample image as the high resolution image.
In particular, the "determining whether the mapping network converges" is determined by an euclidean distance between images.
A terahertz single-pixel image reconstruction system is applicable to the method, and comprises the following steps: terahertz wave modulation unit: for modulating terahertz waves through the mask; terahertz wave detection unit: the terahertz wave detector is used for interacting the modulated terahertz wave with a target object to form a modulation signal, and the modulation signal is received by the detector; terahertz wave reconstruction unit: a reconstructed image used for obtaining a target image according to the modulation signal; the terahertz modulation unit is provided with a mask acquisition unit; the mask obtaining unit is used for obtaining a mask according to a preset mask and a self-coding network.
Further, the terahertz wave detection unit is provided with a mapping unit, and the mapping unit is used for mapping the reconstructed image through a mapping network to obtain a high-resolution reconstructed image.
The invention has the following beneficial effects:
the invention provides a terahertz single-pixel imaging method and a system thereof, which solve the problem of contradiction between imaging speed and imaging quality during terahertz single-pixel imaging through a deep learning network. The number of projection masks is reduced through a self-coding network, the imaging time is shortened, the imaging speed is improved, and the rapid terahertz single-pixel image reconstruction is realized. The low-resolution and high-resolution image mapping is realized through the convolutional neural network, the problem that the image resolution and the signal-to-noise ratio are deteriorated is solved, and the image quality of terahertz single-pixel imaging is improved.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of a terahertz single-pixel imaging method of the invention;
FIG. 2 is a mask acquisition flow chart of the terahertz single-pixel imaging method of the present invention;
FIG. 3 is a schematic diagram of a self-encoding network of the terahertz single-pixel imaging method of the present invention;
FIG. 4 is a flow chart of the terahertz single-pixel imaging method mapping network acquisition of the present invention;
FIG. 5 is a schematic diagram of a convolutional neural network of the terahertz single-pixel imaging method of the present invention;
FIG. 6 is a block diagram of a terahertz single-pixel imaging system of the present invention;
fig. 7 is a specific structural schematic diagram of the terahertz single-pixel imaging system of the invention.
Reference numerals:
1-laser; 2-DMD; 3-a projection lens; 4-ITO; 5-shadow; 6-a lens; 7-a modulator; 8-a detector; 9-a chopper; 10-target.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a terahertz single-pixel imaging method and a system thereof, aiming at the problem that the prime phase speed and the imaging quality are contradictory in terahertz single-pixel imaging in the prior art, and the high-quality imaging is realized while the imaging speed is improved.
Example 1
The embodiment provides a terahertz single-pixel imaging method, which can solve the problem of contradiction between imaging speed and imaging quality in the terahertz single-pixel imaging field in the prior art, and realizes high-quality imaging while improving the imaging speed. The specific scheme is as follows:
a terahertz single-pixel imaging method is disclosed, and the steps are shown in the attached figure 1 of the specification, and specifically comprises the following steps:
s1, modulating terahertz waves through a mask;
s2, enabling the modulated terahertz waves to interact with the target object, and receiving a formed modulation signal by a detector;
and S3, acquiring a reconstructed image of the target image according to the modulation signal.
The mask in S1 is specifically obtained through the following steps: acquiring an original image, and carrying out image coding on the original image through a preset mask to acquire a preset coding result; inputting a preset coding result Y into a self-coding network to obtain a preset decoding image; constructing a difference minimum network according to a preset decoded image and an original image; optimizing a difference minimum network through iterative learning to minimize the difference between a preset decoded image and an original image, and judging whether the difference minimum network is converged: if not, adjusting the self-coding network by adjusting the self-coding network parameters, re-acquiring the decoded image according to the adjusted self-coding network, acquiring a new minimum difference network, and judging whether the new minimum difference network is converged, so as to circulate until the self-coding network is converged; and if the network convergence occurs, outputting the self-coding network parameters when the difference minimum network converges. And finally, obtaining a mask according to the self-coding network parameters when the difference minimum network converges. The specific mask acquisition is shown in figure 2 of the specification.
Specifically, the mask acquisition process includes: obtaining an original image X, firstly selecting a standard image database, and carrying out graying processing and normalization processing on the image, wherein the standard image database comprises an MNIST database, an STL-10 database and the like. And carrying out image coding on the processed original image X through a preset mask phi to obtain a sampling value Y, wherein the sampling value is a preset coding result. The predetermined mask phi includes common masks such as a fourier mask, a hadamard mask, and a wavelet mask. Obtaining a preset decoding image X through a deep learning network W1The deep learning network selected in the invention is a self-coding network. According to X1And X, constructing a minimum difference network, and judging the modulation effect of the mask by judging the difference between the original image and a preset decoding image. The terahertz single-pixel imaging process can be expressed as encoding Y ═ phix, decoding X1σ Y, minimum difference network Min | | | X-X1||2. Assuming that the parameters of the deep learning network learning are t and v, which act on phi and sigma respectively, Y ═ t phix, X is obtained1V σ Y. Iterative learning is carried out through a self-coding network so that the difference minimum network is continuously optimized until the difference minimum network converges, at the moment, the difference between the preset decoding image and the original image is minimum, optimized t and v can be obtained, the new coding mask phi is t phi, and the optimized network decoding W is v sigma, so that the upper limit of the number of codes can be set to optimize the output t, and the new coding mask is output. Assuming that the resolution of an original image is nxn, by using the self-coding network provided by the embodiment, an image with the resolution of nxn can be reconstructed only by using M (M is much smaller than nxn) coding modes, and by reducing the time for coding acquisition, the time for the whole terahertz single-pixel imaging is further shortened, and the terahertz single-pixel imaging speed is greatly improved.
The number of masks is reduced through a self-coding network, and the imaging speed is further improved. The self-coding network consists of a series of filters and activation functions (called layers) that map the input characteristics to the output characteristics. The number and type of filters and the type of mapping constitute the structure of the network, as shown in fig. 3 in the description. The filters of the self-encoding network include a full-run filter, an S convolution filter, an S/2 convolution filter, and a convolution filter. In addition, when determining whether the network having the minimum difference converges, the determination may be performed using the euclidean distance between the images.
After the step of acquiring the reconstructed image in S3, the method further includes performing mapping processing on the reconstructed image through a mapping network to acquire a high-resolution reconstructed image. The specific mapping network acquisition mode comprises the following steps: acquiring a low-resolution image and a high-resolution image, inputting the low-resolution image and the high-resolution image into a mapping network for learning, taking the low-resolution image as input, and taking the high-resolution image as a learning target; performing feature block processing on the low-resolution image through a mapping network to obtain a first feature; carrying out nonlinear mapping on the low-resolution image through a mapping network to obtain a second characteristic; constructing a loss function network according to the first characteristic and the second characteristic; training the mapping network, and judging whether the loss function network is converged: if the loss function network is converged, acquiring a trained mapping network; if the loss function network is not converged, the mapping network is trained again, the first characteristic and the second characteristic are obtained again, a new loss function network is obtained, whether the new loss function network is converged or not is judged, and the operation is circulated until the loss function network is converged. The specific mapping process is shown in figure 4 in the specification.
Specifically, to restore a low-resolution image to a high-resolution image through a mapping network, a mapping network F needs to be learned. The mapping network comprises a convolutional neural network, end-to-end image mapping is realized through the convolutional neural network, and a specific mapping network acquisition process comprises the following steps: obtaining a sample image Y, selecting a standard image database to perform graying and normalization processing on the image, and filtering the image by using frequency domain processing to obtain a mode with low resolutionThe blurred image is a low-resolution image L, and the original image Y is a high-resolution image H. The low-resolution image L and the high-resolution image H are input into the mapping network F for learning, and the low-resolution image L is restored into the high-resolution image H through continuous iterative learning. Extracting the features of the low-resolution image L through a mapping network F to obtain first features; carrying out nonlinear mapping on the low-resolution image L through a mapping network F to obtain a second characteristic; and constructing a loss function network according to the first characteristic and the second characteristic, and gradually approaching convergence to the loss function network by continuously adjusting the correction mapping network parameters to minimize the loss function. Wherein the first characteristic F1The expression of (L) is:
F1(L)=max(0;W1×L+B1)
wherein, W1For feature filters, L is the low resolution image, B1Is an offset parameter;
second characteristic F2The expression of (L) is:
F2(L)=max(0;W2×F1(L)+B2)
wherein, F1(L) is a first feature, W2For high-resolution feature filters, B2As an offset parameter
The expression of the loss function network is:
min‖F2(L)-H‖2
wherein, F2(L) is a second feature, and H is a high resolution image.
If the loss function network is converged, acquiring a trained mapping network; and if the loss function network is not converged, retraining the mapping network, reacquiring the first characteristic and the second characteristic, constructing a new loss function network, judging whether the new loss function network is converged, and circulating until the loss function network is converged. By defining specific values, e.g. 10-5If the loss function is lower than the value, the network convergence of the loss function can be considered, and the learning is finished.
The structural principle of end-to-end image mapping realized by the convolutional neural network is shown in the attached figure 5 of the specification, low-resolution images are subjected to dimension reduction processing in a standardized batch mode, and feature extraction is performed by using an activation function. And obtaining a high-resolution image by utilizing upsampling, using the Euclidean distance of the two images as a loss function to enable the network to be converged, optimizing the layer number and weight of the network, and realizing an end-to-end image mapping function. The low-resolution image is restored into the high-resolution image through the mapping network, the reconstructed image constructed through the terahertz single-pixel imaging is restored into the high-resolution reconstructed image, and the imaging quality of the terahertz single-pixel imaging is improved. And the network does not influence the optimization of the self-coding network on the imaging speed.
The embodiment provides a terahertz single-pixel imaging method, and the problem of contradiction between imaging speed and imaging quality during imaging is solved by applying a deep learning network to the field of terahertz single-pixel imaging. The number of projection masks is reduced through a self-coding network, the imaging time is shortened, the imaging speed is improved, and the rapid terahertz single-pixel image reconstruction is realized. The low-resolution and high-resolution image mapping is realized through the convolutional neural network, the problem that the image resolution and the signal-to-noise ratio are deteriorated is solved, and the image quality of terahertz single-pixel imaging is improved.
Example 2
In this embodiment, on the basis of embodiment 1, a terahertz single-pixel imaging method is systematized, and a terahertz single-pixel imaging system is provided. The specific scheme is as follows:
a terahertz single-pixel image imaging system comprises a terahertz wave modulation unit, a terahertz wave detection unit and a terahertz wave reconstruction unit, and is specifically shown in figure 6 in the specification. The terahertz wave modulation unit is used for modulating terahertz waves through a mask; the terahertz wave detection unit is used for interacting the modulated terahertz wave with a target object to form a modulation signal, and receiving the modulation signal through the detector; the terahertz wave reconstruction unit is used for acquiring a reconstructed image of the target image according to the modulation signal. The terahertz wave modulation unit includes a laser 1, a DMD (Digital micromirror array) 2, a projection lens 3, and an intrinsic semiconductor. The terahertz wave detection part mainly comprises an ITO (Indium tin oxide) 4, a light source 5, a lens 6, a modulator 7, a detector 8 and a chopper 9. Laser 1 is masked by DMD2 and then penetrates ITO4 glass through a lens 3 to be projected on a modulator 7, and ITO4 can penetrate visible light and reflect terahertz light. The terahertz light source 5 is collimated by the lens 6 to penetrate through the modulator 7 and the target 10 and is focused on the detector 8 by the lens 6, the terahertz light is modulated by the mask, and the modulated intensity signal is output on the detector 8. The specific structure is shown in figure 7 in the specification.
The terahertz wave modulation unit is also provided with a mask acquisition unit, and the mask acquisition unit acquires a mask according to a preset mask and a self-coding network. The specific acquisition process comprises the following steps: acquiring an original image, and carrying out image coding on the original image through a preset mask to acquire a preset coding result; inputting a preset coding result into a self-coding network to obtain a preset decoding image; constructing a difference minimum network according to a preset decoded image and an original image; optimizing a difference minimum network through iterative learning to minimize the difference between a preset decoded image and an original image, and judging whether the difference minimum network is converged: if not, adjusting the self-coding network by adjusting the self-coding network parameters, re-acquiring the decoded image according to the adjusted self-coding network, re-constructing the minimum difference network, and judging whether the reconstructed minimum difference network is converged, so as to circulate until the self-coding network is converged; and if the network convergence occurs, outputting the self-coding network parameters when the difference minimum network converges. And finally, obtaining a mask according to the self-coding network parameters when the difference minimum network converges.
The terahertz wave reconstruction unit is provided with a mapping unit, and the mapping unit is used for mapping the reconstructed image through a mapping network to obtain a high-resolution reconstructed image. The specific acquiring process of the mapping unit comprises the following steps: acquiring a low-resolution image and a high-resolution image, and inputting the low-resolution image and the high-resolution image into a mapping network for learning; performing feature block processing on the low-resolution image through a mapping network to obtain a first feature; carrying out nonlinear mapping on the low-resolution image through a mapping network to obtain a second characteristic; constructing a loss function network according to the first characteristic and the second characteristic; training the mapping network to make the loss function minimum or less than a certain preset value, and judging whether the loss function network is converged: if the loss function network is not converged, the mapping network is trained again, the first characteristic and the second characteristic are obtained again, a new loss function network is constructed, whether the new loss function network is converged or not is judged, and the circulation is carried out until the loss function network is converged; and if the loss function network is converged, acquiring the trained mapping network.
When the terahertz single-pixel imaging device is used, firstly, a small amount of coding stripes are used for modulating terahertz waves by a DMD, and then a terahertz image is reconstructed by utilizing a network, so that the purpose of rapid terahertz single-pixel imaging is achieved. Secondly, a target image is placed close to a modulator by utilizing the principle of near-field imaging to obtain a terahertz high-resolution image, then the target image and the modulator are placed at a certain distance, the image has the problem of low resolution due to scattering, and the high-resolution image and the low-resolution image are mapped by utilizing a mapping network, so that the problem of poor target imaging quality in an imaging system is solved.
The embodiment provides a terahertz single-pixel imaging system, which adopts a cascade double network formed by a self-coding network and a mapping network to realize the reconstruction of an undersampled high-resolution image and solves the problems of long terahertz single-pixel imaging time and poor imaging quality.
In summary, the invention provides a terahertz single-pixel imaging method and a system thereof, which solve the problem of contradiction between imaging time and imaging quality in the imaging process through a deep learning network. The image coding process is learned through the deep learning network, image reconstruction under undersampling is given, universality is achieved compared with the prior art based on random coding, Hadamard coding, Fourier coding and the like, terahertz single-pixel imaging speed is greatly improved, and imaging time is shortened. The method and the device have the advantages that end-to-end network mapping is realized through the convolutional neural network, the image quality is improved, the method and the device are different from an imaging system which utilizes near-field imaging and a high-performance modulator in the prior art, a hardware system with harsh use conditions can be avoided, high-quality image reconstruction is realized through learning low-resolution-to-high-resolution image mapping, the adaptability is stronger, the real-time performance is higher, and the method and the device are more in line with the actual application requirements.
It will be understood by those skilled in the art that the modules or steps of the invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of computing devices, and optionally they may be implemented by program code executable by a computing device, such that it may be stored in a memory device and executed by a computing device, or it may be separately fabricated into various integrated circuit modules, or it may be fabricated by fabricating a plurality of modules or steps thereof into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments illustrated herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
The above disclosure is only a few specific implementation scenarios of the present invention, however, the present invention is not limited thereto, and any variations that can be made by those skilled in the art are intended to fall within the scope of the present invention.
Claims (11)
1. A terahertz single-pixel imaging method specifically comprises the following steps:
s1, modulating terahertz waves through a mask;
s2, enabling the modulated terahertz waves to interact with the target object, and receiving a formed modulation signal by a detector;
s3, acquiring a reconstructed image of the target image according to the modulation signal;
the method is characterized in that the mask in the S1 is obtained through the following steps:
acquiring an original image, and carrying out image coding on the original image through a preset mask to acquire a preset coding result;
inputting the preset coding result into a self-coding network to obtain a preset decoding image;
constructing a difference minimum network according to the preset decoding image and the original image, wherein the difference minimum network expression is as follows:
min||X-X1||2
wherein X is an original image, X1Decoding the image for presetting;
optimizing the minimum difference network through iterative learning, and judging whether the minimum difference network converges:
if the minimum difference network is not converged, adjusting the self-coding network by adjusting the self-coding network parameters, re-acquiring a decoded image according to the adjusted self-coding network, acquiring a new minimum difference network, and judging whether the new minimum difference network is converged, so as to circulate until the self-coding network is converged;
if the difference minimum network is converged, outputting self-coding network parameters when the difference minimum network is converged;
and acquiring the mask according to the self-coding network parameters.
2. The method of claim 1, wherein the predetermined mask is a fourier mask, a hadamard mask, or a wavelet mask.
3. The method of claim 1, wherein the self-encoding network comprises a full-run filter, an S convolution filter, an S/2 convolution filter, and a convolution filter.
4. The method according to claim 1, wherein after the "acquiring an original image" and before the "image coding the original image through a preset mask", further comprising:
and selecting an image database to perform graying processing and normalization processing on the original image.
5. The method according to claim 1, further comprising, after the S3, performing a mapping process on the reconstructed image obtained at the S3 through a mapping network to obtain a high resolution reconstructed image.
6. The method according to claim 6, wherein the obtaining of the mapping network specifically comprises:
acquiring a low-resolution image and a high-resolution image, taking the low-resolution image as input and the high-resolution image as a learning target, and learning the mapping network;
the mapping network carries out feature block processing on the low-resolution image to obtain a first feature;
the mapping network carries out nonlinear mapping on the low-resolution image to obtain a second characteristic;
constructing a loss function network according to the first characteristic and the second characteristic;
training the mapping network, and judging whether the loss function network is converged:
if the loss function network is converged, acquiring a trained mapping network;
and if the loss function network is not converged, training the mapping network again, acquiring the first characteristic and the second characteristic again, acquiring a new loss function network, judging whether the new loss function network is converged, and circulating until the loss function network is converged.
7. The method of claim 6, wherein the first characteristic is expressed by:
F1(L)=max(0;W1×L+B1)
wherein, F1(L) is a first feature, W1Is a characteristic filter, L isLow resolution image, B1Is an offset parameter;
the expression of the second characteristic is:
F2(L)=max(0;W2×F1(L)+B2)
wherein, F2(L) is a second feature, F1(L) is a first feature, W2For high-resolution feature filters, B2As an offset parameter
The expression of the loss function network is:
min||F2(L)-H||2
wherein, F2(L) is a second feature, and H is a high resolution image.
8. The method according to claim 6, wherein said "acquiring a low resolution image and a high resolution image" comprises in particular:
acquiring a sample image, carrying out graying processing and normalization processing on the sample image, and filtering the sample image by using a frequency domain to obtain the low-resolution image;
taking the sample image as the high resolution image.
9. The method of claim 5, wherein the determining whether the mapping network converges is performed by determining Euclidean distance between images.
10. A terahertz single-pixel image reconstruction system suitable for use in the method of any one of claims 1-9, comprising:
terahertz wave modulation unit: for modulating terahertz waves through the mask;
terahertz wave detection unit: the terahertz wave detector is used for interacting the modulated terahertz wave with a target object to form a modulation signal which is received by the detector;
terahertz wave reconstruction unit: a reconstructed image used for obtaining a target image according to the modulation signal;
the terahertz modulation device is characterized in that the terahertz modulation unit is provided with a mask acquisition unit, and the mask acquisition unit is used for acquiring a mask according to a preset mask and a self-coding network.
11. The method according to claim 10, wherein the terahertz wave detection unit is provided with a mapping unit for performing mapping processing on the reconstructed image through a mapping network to obtain a high-resolution reconstructed image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011436254.XA CN112525851A (en) | 2020-12-10 | 2020-12-10 | Terahertz single-pixel imaging method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011436254.XA CN112525851A (en) | 2020-12-10 | 2020-12-10 | Terahertz single-pixel imaging method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112525851A true CN112525851A (en) | 2021-03-19 |
Family
ID=74999096
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011436254.XA Pending CN112525851A (en) | 2020-12-10 | 2020-12-10 | Terahertz single-pixel imaging method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112525851A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114037609A (en) * | 2021-10-21 | 2022-02-11 | 中国科学院沈阳自动化研究所 | Terahertz image super-resolution algorithm based on learning terahertz imaging inverse process |
CN114219843A (en) * | 2021-12-16 | 2022-03-22 | 河南工业大学 | Method and system for constructing terahertz spectrum image reconstruction model and application |
WO2023082162A1 (en) * | 2021-11-12 | 2023-05-19 | 华为技术有限公司 | Image processing method and apparatus |
WO2023109719A1 (en) * | 2021-12-15 | 2023-06-22 | 深圳先进技术研究院 | Terahertz single-pixel super-resolution imaging method and system |
WO2023193461A1 (en) * | 2022-04-06 | 2023-10-12 | 深圳先进技术研究院 | Terahertz single-pixel imaging method and system |
WO2024109836A1 (en) * | 2022-11-25 | 2024-05-30 | 中国科学院深圳先进技术研究院 | Underwater single-pixel imaging method and system based on deep learning network |
WO2024109842A1 (en) * | 2022-11-25 | 2024-05-30 | 中国科学院深圳先进技术研究院 | Terahertz single-pixel real-time imaging method and system based on physical model |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105527242A (en) * | 2014-10-27 | 2016-04-27 | 深圳先进技术研究院 | Terahertz compressive imaging system and method |
WO2018059135A1 (en) * | 2016-09-27 | 2018-04-05 | 深圳市太赫兹科技创新研究院 | Method for measuring parameters of terahertz beam |
CN109297925A (en) * | 2018-10-09 | 2019-02-01 | 天津大学 | A kind of Terahertz high-resolution fast imaging device based on splits' positions perception |
CN109444085A (en) * | 2018-12-18 | 2019-03-08 | 深圳先进技术研究院 | A kind of near field THz wave spectrum imaging system and method |
CN110231292A (en) * | 2019-06-28 | 2019-09-13 | 深圳先进技术研究院 | A kind of single pixel THz wave imaging method and system |
CN110425986A (en) * | 2019-07-17 | 2019-11-08 | 北京理工大学 | Three-dimensional computations imaging method and device based on single pixel sensor |
-
2020
- 2020-12-10 CN CN202011436254.XA patent/CN112525851A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105527242A (en) * | 2014-10-27 | 2016-04-27 | 深圳先进技术研究院 | Terahertz compressive imaging system and method |
WO2018059135A1 (en) * | 2016-09-27 | 2018-04-05 | 深圳市太赫兹科技创新研究院 | Method for measuring parameters of terahertz beam |
CN109297925A (en) * | 2018-10-09 | 2019-02-01 | 天津大学 | A kind of Terahertz high-resolution fast imaging device based on splits' positions perception |
CN109444085A (en) * | 2018-12-18 | 2019-03-08 | 深圳先进技术研究院 | A kind of near field THz wave spectrum imaging system and method |
CN110231292A (en) * | 2019-06-28 | 2019-09-13 | 深圳先进技术研究院 | A kind of single pixel THz wave imaging method and system |
CN110425986A (en) * | 2019-07-17 | 2019-11-08 | 北京理工大学 | Three-dimensional computations imaging method and device based on single pixel sensor |
Non-Patent Citations (2)
Title |
---|
吴丝桐: ""基于深度学习的单像素成像方法研究"", 《中国优秀硕士学位论文全文数据库 计算机软件及计算机应用辑》 * |
朱韵樵;申朝阳;张豪;何雨莲;文岐业;: "基于空间型调制器的太赫兹波快速成像技术", 太赫兹科学与电子信息学报, no. 03 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114037609A (en) * | 2021-10-21 | 2022-02-11 | 中国科学院沈阳自动化研究所 | Terahertz image super-resolution algorithm based on learning terahertz imaging inverse process |
CN114037609B (en) * | 2021-10-21 | 2024-04-19 | 中国科学院沈阳自动化研究所 | Terahertz image super-resolution algorithm based on learning terahertz imaging inverse process |
WO2023082162A1 (en) * | 2021-11-12 | 2023-05-19 | 华为技术有限公司 | Image processing method and apparatus |
WO2023109719A1 (en) * | 2021-12-15 | 2023-06-22 | 深圳先进技术研究院 | Terahertz single-pixel super-resolution imaging method and system |
CN114219843A (en) * | 2021-12-16 | 2022-03-22 | 河南工业大学 | Method and system for constructing terahertz spectrum image reconstruction model and application |
WO2023193461A1 (en) * | 2022-04-06 | 2023-10-12 | 深圳先进技术研究院 | Terahertz single-pixel imaging method and system |
WO2024109836A1 (en) * | 2022-11-25 | 2024-05-30 | 中国科学院深圳先进技术研究院 | Underwater single-pixel imaging method and system based on deep learning network |
WO2024109842A1 (en) * | 2022-11-25 | 2024-05-30 | 中国科学院深圳先进技术研究院 | Terahertz single-pixel real-time imaging method and system based on physical model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112525851A (en) | Terahertz single-pixel imaging method and system | |
Zhang et al. | Image restoration: From sparse and low-rank priors to deep priors [lecture notes] | |
CN111968044A (en) | Low-illumination image enhancement method based on Retinex and deep learning | |
Fienup et al. | Comparison of reconstruction algorithms for images from sparse-aperture systems | |
WO2021110262A1 (en) | Noise reconstruction for image denoising | |
WO2023193461A1 (en) | Terahertz single-pixel imaging method and system | |
WO2023109719A1 (en) | Terahertz single-pixel super-resolution imaging method and system | |
CN110415193A (en) | The restored method of coal mine low-light (level) blurred picture | |
CN110751700A (en) | Sampling and reconstruction integrated deep learning network for single-pixel imaging and training method thereof | |
Wali et al. | Recent progress in digital image restoration techniques: a review | |
Li et al. | BPFD-Net: enhanced dehazing model based on Pix2pix framework for single image | |
Chen et al. | High-dynamic range, night vision, image-fusion algorithm based on a decomposition convolution neural network | |
Zhu et al. | Image restoration by a mixed high‐order total variation and L1 regularization model | |
Jiji et al. | Extended depth of focus imaging using optics and image processing | |
Ye et al. | Research on mechanism of joint-coding imaging based on generative adversarial neural network | |
Xiang et al. | Image DAEs based on residual entropy maximum | |
Pan et al. | DICNet: achieve low-light image enhancement with image decomposition, illumination enhancement, and color restoration | |
CN116258783A (en) | Underwater single-pixel imaging method and system based on deep learning network | |
Xiao et al. | Video denoising algorithm based on improved dual‐domain filtering and 3D block matching | |
CN115375540A (en) | Terahertz image super-resolution method based on deep learning algorithm | |
CN112907444A (en) | Terahertz image super-resolution reconstruction method based on complex field zero sample learning | |
Li et al. | Foggy image restoration using deep sub‐pixel reconstruction network | |
Ahn et al. | Data Augmentation for Low-Level Vision: CutBlur and Mixture-of-Augmentation | |
Ramya et al. | An effective super resolution image reconstruction using spatial entropy based normalized deep convolutional neural network | |
CN115103118B (en) | High dynamic range image generation method, device, equipment and readable storage medium |
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 |