CN114387164A - Terahertz single-pixel super-resolution imaging method and system - Google Patents
Terahertz single-pixel super-resolution imaging method and system Download PDFInfo
- Publication number
- CN114387164A CN114387164A CN202111532798.0A CN202111532798A CN114387164A CN 114387164 A CN114387164 A CN 114387164A CN 202111532798 A CN202111532798 A CN 202111532798A CN 114387164 A CN114387164 A CN 114387164A
- Authority
- CN
- China
- Prior art keywords
- image
- pixel
- terahertz
- sampling
- transform domain
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a terahertz single-pixel super-resolution imaging method and system, and belongs to the field of terahertz imaging. According to the invention, a statistical mode is used for learning the fringe pattern needing to be projected of a certain pattern in a priori manner, so that under-sampled terahertz single-pixel imaging is realized, and the purpose of improving the imaging speed is achieved; meanwhile, in order to make up for interference noise caused by information loss and other factors due to insufficient sampling rate, an image super-resolution network is used for realizing image denoising and pixel reconstruction in the later period, and a high-quality imaging effect is achieved.
Description
Technical Field
The invention belongs to the field of terahertz imaging, and particularly relates to a terahertz single-pixel super-resolution imaging method and system.
Background
Terahertz waves have the advantages of low photon energy, penetration of nonpolar substances and the like, and have great potential in terahertz imaging, spectral analysis, high-speed communication and other aspects. Terahertz imaging is of great significance in biomedicine, material detection and safety monitoring. However, 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.
The novel terahertz imaging technology is a terahertz single-pixel imaging system at present, and compared with an area array terahertz imaging system, the terahertz single-pixel imaging system not only saves hardware cost, but also brings new possibility for miniaturization and commercialization of terahertz. The conventional terahertz single-pixel imaging system is realized by means of compressed sensing, Hadamard basis, Fourier basis and the like, wherein She et al utilizes a 220-micron silicon-based graphene modulator and Fourier stripes to realize sub-wavelength terahertz image reconstruction, and utilizes the characteristic of image sparsity to realize 10% modulation mask image reconstruction by acquiring low-frequency coefficients and inverse Fourier transform; rayko et al uses a silicon total internal reflection prism and a Hadamard mask to realize near real-time terahertz single-pixel video, and uses the Hadamard domain sparse feature to reduce sampling time by using an under-sampling technique.
Due to the particularity of terahertz wave wavelength and the limitation of single-pixel imaging, the terahertz single-pixel imaging technology has the following problems to be solved: (1) the imaging speed is slow. Because single pixel imaging sends a single detector through a modulator to receive the light intensity of a plurality of mask patterns, the imaging speed depends on the modulation time, the number of projections, the projection speed, the response time of the detector and the like; in the case where the image resolution is increased, the imaging speed is also slowed down by the increase in the number of samples. (2) Poor imaging quality: terahertz waves are extremely susceptible to interference of coherent light in the transmission process, and meanwhile, due to detection errors caused by hardware thermal noise, the imaging quality is greatly influenced. In addition, although the undersampling mode can greatly shorten the imaging time, the high-frequency information of the output image is lost to deteriorate the image; however, from an observable point of view, when the sampling rate is lower than a certain ratio, the reconstructed pattern may have serious distortion or even distortion, so that the sampling rate is not suitable for being lower than a certain lower limit.
Disclosure of Invention
Aiming at the problems of low imaging speed, poor imaging quality and the like in the related technology, the invention aims to provide a terahertz single-pixel super-resolution imaging method and system, which are used for learning fringe patterns needing to be projected for a certain pattern a priori through a statistical method, realizing undersampled terahertz single-pixel imaging and achieving the purpose of improving the imaging speed; meanwhile, in order to make up for interference noise caused by information loss and other factors due to insufficient sampling rate, an image super-resolution network is used for realizing image denoising and pixel reconstruction in the later period, and a high-quality imaging effect is achieved.
In order to achieve the above object, an aspect of the present invention provides a terahertz single-pixel super-resolution imaging method, including the following steps:
s100, carrying out prior learning on the coding positions of the similar patterns, extracting a sampling strategy stored through the prior learning, and generating mask patterns, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction;
s200, selecting a data set to perform corresponding processing based on the sampling strategy to generate a training set, and using the training set for initializing a residual error dense network;
s300, building a terahertz single-pixel imaging system, loading the mask pattern by adopting a digital micromirror array according to the sampling strategy, and reflecting the mask pattern onto a target pattern through laser; the terahertz wave irradiates a terahertz modulator, the laser intensity of a target pattern is modulated, the terahertz wave is periodically received through a terahertz detector, transform domain information is recovered from the intensity information, and a first image is obtained through decoding;
s400, reconstructing a second image through the residual error dense network based on the first image.
Further, the step S100 specifically includes the following steps:
s101, classifying the similar pictures of the collected target patterns to form a data set;
s102, carrying out full sampling on a plurality of pictures of the same type and storing transform domain data;
s103, counting the corresponding transform domain position of the module coefficient intensity under the specific sampling rate in the transform domain of the image by adopting a statistical method, thereby generating the mask pattern of the image.
Further, in step S102, inverse fourier transform is applied to the transform domain.
Further, the step S400 specifically includes the following steps:
s401, extracting a shallow information characteristic graph from an input image through a convolutional layer, and then entering a residual error dense block;
s402, generating a corresponding feature map by each residual error dense block, splicing all the feature maps to form a whole, and performing dense feature fusion;
and S403, interpolating the first image to a high-pixel second image through an upsampling module, and reconstructing the second image to restore the second image to a single-channel high-definition gray image or a three-channel high-definition color image.
Further, the first image is an image of 32 × 32 pixels, and the second image is an image of 64 × 64 pixels.
Further, the dense feature fusion process comprises global feature fusion and global residual learning, wherein the feature maps are combined into a whole by using a fusion layer, deep information is further obtained and restored into a feature map with 64 channels through convolution operation, and finally shallow information and deep information are added through residual learning.
In another aspect of the present invention, a terahertz single-pixel super-resolution imaging system is further provided, including:
the learning unit is used for carrying out prior learning on the coding positions of the similar patterns, extracting a sampling strategy stored in the prior learning and generating mask patterns, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction;
the training unit is used for selecting a corresponding data set to carry out image scaling, simultaneously selecting corresponding coefficients of a processed image transformation domain to carry out inverse transformation based on the sampling strategy to form a training set, and using the training set for initializing a residual dense network;
the acquisition unit loads a mask pattern learned in advance by adopting a digital micromirror array and reflects the mask pattern onto a target pattern through laser; irradiating the terahertz wave to a terahertz modulator, modulating the laser intensity of a target pattern, periodically receiving the terahertz wave by a terahertz detector, restoring transform domain information from the terahertz wave, and decoding to obtain a first image;
and the reconstruction unit is used for reconstructing a second image through the residual error dense network based on the first image.
Further, the learning unit includes:
the classification module is used for classifying the similar pictures of the acquired target pattern to form a data set;
the full sampling module is used for performing full sampling on a plurality of pictures of the same type and storing transform domain data;
and the generation module is used for counting the position of the corresponding transform domain in the transform domain of the image by adopting a statistical method, wherein the intensity of the module coefficient meets the requirement of the position of the corresponding transform domain under a specific sampling rate, so that the mask pattern of the image is generated.
Further, the reconstruction unit comprises:
the convolution module extracts a shallow information characteristic graph from an input image through a convolution layer and then enters a residual error dense block;
the fusion module is used for generating corresponding feature maps by each residual error dense block, splicing all the feature maps to form a whole and carrying out dense feature fusion;
and the interpolation module is used for interpolating the first image into a high-pixel second image through the up-sampling module and reconstructing the second image to restore the second image into a single-channel high-definition gray image or a three-channel high-definition color image.
Further, the first image is an image of 32 × 32 pixels, and the second image is an image of 64 × 64 pixels.
Compared with the prior art, the technical scheme provided by the invention combines the prior optimal sampling scheme and the super-resolution reconstruction of deep learning into the terahertz single-pixel imaging system, so that the reconstruction of a high-definition image is realized within the fastest imaging time. Specifically, the invention adopts a computer-aided optimization sampler based on prior statistics, is suitable for imaging and monitoring common similar targets, and is superior to most undersampling schemes; meanwhile, in order to improve the generalization degree of the class, the sampling scheme can be further counted and optimized in the process of expanding and collecting the following samples, so that the method is a sampling mode which can be learned; in addition, the method applies the residual error dense network to terahertz imaging, thereby not only eliminating image noise brought by the system, but also improving the resolution ratio without increasing the sampling number.
Drawings
FIG. 1 is a schematic diagram of a terahertz single-pixel imaging system in an embodiment of the invention;
FIG. 2(a) is a schematic diagram of an RDN network structure in the embodiment of the present invention;
FIG. 2(b) is a diagram illustrating the RDB network structure in the embodiment of the present invention;
FIG. 3 is a flowchart of the operation of the fast terahertz single-pixel super-resolution reconstruction system according to the embodiment of the present invention;
fig. 4 is a graph showing the effect of the sample of the embodiment of the present invention in different sampling modes at 8% and 10%.
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. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
First, a terahertz single-pixel imaging device is built, as shown in fig. 1. The main optical path of the imaging Device is composed of a terahertz laser and a detector, and the modulation part is composed of laser, a Digital micromirror array (DMD), a projection lens and an intrinsic semiconductor. Laser is masked with a DMD to form phi, and the phi penetrates Indium Tin Oxide (ITO) glass through a lens to be projected on a modulator, wherein the ITO can penetrate visible light and reflect terahertz light. The terahertz light penetrates through the modulator through the collimation of the lens and is focused on the detector by the lens, and the terahertz light is modulated through the mask and outputs a modulated intensity signal I on the detector.
Based on the device, the embodiment of the invention provides a terahertz single-pixel super-resolution imaging method, which comprises the following steps:
s100, priori learning is carried out on the coding positions of the similar patterns, sampling strategies stored through the priori learning are extracted, and mask patterns are generated, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction.
Any two-dimensional image can be viewed as being weighted by a complete set of orthogonal mask patterns, each mask pattern corresponding to a frequency point on the transform domain, and the relationship between the target pattern and the transform domain function is given by equation (1).
Where I (x, y) is the object target, M, N are the length and width of the object target, u, v are the point coordinates of the frequency on the transform domain, f is a two-dimensional matrix function, the size is determined by (x, y, u, v), auvIs uniquely determined by (u, v) as the weight size. Weighting all orthogonal base patterns, wherein the process of obtaining an original pattern is called full sampling, and the number of measurements is equal to K-MxN; in order to reduce the number of measurements, the sampling mode of acquiring coefficients on part of transform domains to perform weighted inversion to obtain patterns with information loss is called undersampling. Wherein the a priori statistical sampling is given by equation (2).
In the formula, | F (x, y) | represents a modulus of a certain point on the transform domain, p is a sampling rate, and a is an absolute value of a threshold corresponding to p, which indicates that coefficients at selected positions in the transform domain occupy a probability p before the modulus of the coefficients in the whole transform domain.
Wherein, the step S100 specifically includes the following steps:
s101, classifying the similar pictures of the collected target patterns to form a data set;
s102, carrying out full sampling on a plurality of pictures of the same type and storing transform domain data;
s103, counting the corresponding transform domain position of the module coefficient intensity under the specific sampling rate in the transform domain of the image by adopting a statistical method, thereby generating the mask pattern of the image.
S200, selecting a data set to perform corresponding processing to generate a training set based on the sampling strategy, and using the training set to initialize the residual error dense network.
The method is suitable for imaging and monitoring common similar targets, and is superior to most undersampling schemes; meanwhile, in order to improve the generalization degree of the class, the sampling scheme can be further counted and optimized in the process of expanding and collecting the following samples, so that the method is a sampling mode which can be learned. Therefore, in order to improve the sampling speed, the terahertz single-pixel imaging system adopting the prior statistical undersampling scheme is adopted, and mature inverse Fourier domain transformation is adopted on a transformation domain.
S300, building a terahertz single-pixel imaging system, loading the mask pattern by adopting a digital micromirror array according to the sampling strategy, and reflecting the mask pattern onto a target pattern through laser; the terahertz wave irradiates a terahertz modulator, the laser intensity of a target pattern is modulated, the terahertz wave is periodically received through a terahertz detector, transform domain information is recovered from the intensity information, and a first image is obtained through decoding;
s400, reconstructing a 64 × 64 pixel image through a residual dense network based on the 32 × 32 pixel image.
In order to reconstruct a high-definition image from the undersampled image with missing information, further processing of the imaging results is required. The residual dense network can extract feature information from the shallow layer and the deep layer of the image to perform self-adaptive feature fusion, and an up-sampling layer is used for interpolation to obtain a high-definition image. The network structure of the RDN is represented by fig. 2(a), and the structure is divided into: two ConV layers for extracting shallow features; a Residual Dense Block (RDB) for extracting features of each layer; dense Feature Fusion (DFF) for stitching features of each layer; for obtaining single-channel or multi-channel upsampling modules. The input image passes through two ConV layers with convolution kernel size of 3x3 and filters of 1 and 64, a shallow information feature map is extracted, and then the input image enters RDBs to acquire deep information. The RDBs are composed of a plurality of RDBs, the network structure of the RDBs is given by a figure 2(b), each RDB corresponds to different receptive fields, different local features can be extracted respectively, and the RDBs are given by a formula:
from the network structure, the RDB consists of residual blocks and dense blocks, the filter size is 64, and the convolution kernel size is 3 × 3. RDB has the following characteristics: a continuous memory mechanism, a local fusion mechanism and a local residual error mechanism. A continuous memory mechanism, wherein each RDB transmits a plurality of previous states to a current layer for processing, and each layer consists of a ConV layer and a ReLU layer; the later the number of layers, the more memory states are transferred. And the local fusion mechanism is used for connecting the transmitted state and the current input in series, extracting deep features from a plurality of features in a self-adaptive manner and reducing the output number of channels. And the local residual learning partially overlaps the output of the last RDB and the feature fusion part, thereby further improving the information flow and the expression capability of the model.
After passing through a series of RDBs, each RDB will generate a corresponding feature map, and all feature maps are spliced together to form a whole, which is called dense feature fusion. The dense feature fusion is divided into global feature fusion and global residual learning, firstly, all feature maps are combined into a whole by using a fusion layer, deep information is further obtained and restored into a feature map with 64 channels through convolution operation of 1x1 and convolution operation of 3x3, and finally, shallow information and deep information are added through residual learning, and the whole dense feature fusion can be expressed as follows by a formula:
FDF＝F0+HGEF([F0,Fd,...,FD]) (4)
finally, the 32x32 pixel image is interpolated to 64x64 pixels by an up-sampling module and reconstructed, and the image is restored to be a single-channel high-definition gray image or a three-channel high-definition color image, and the whole RDN network can be represented as follows:
in order to reconstruct an object image
The loss function of the network training adopts an L1loss function, and the function can be expressed as:
the step S400 specifically includes the following steps:
s401, extracting a shallow information characteristic graph from an input image through a convolutional layer, and then entering a residual error dense block;
s402, generating a corresponding feature map by each residual error dense block, splicing all the feature maps to form a whole, and performing dense feature fusion;
and S403, interpolating the first image to a high-pixel second image through an upsampling module, and reconstructing the second image to restore the second image to a single-channel high-definition gray image or a three-channel high-definition color image.
In another aspect, an embodiment of the present invention further provides a terahertz single-pixel super-resolution imaging system, including:
the learning unit is used for carrying out prior learning on the coding positions of the similar patterns, extracting a sampling strategy stored in the prior learning and generating mask patterns, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction;
the training unit is used for selecting a corresponding data set to carry out image scaling, simultaneously selecting corresponding coefficients of a processed image transformation domain to carry out inverse transformation based on the sampling strategy to form a training set, and using the training set for initializing a residual dense network;
the acquisition unit loads a mask pattern learned in advance by adopting a digital micromirror array and reflects the mask pattern onto a target pattern through laser; irradiating the terahertz wave to a terahertz modulator, modulating the laser intensity of a target pattern, periodically receiving the terahertz wave by a terahertz detector, restoring transform domain information from the terahertz wave, and decoding to obtain a first image;
and the reconstruction unit is used for reconstructing a second image through the residual error dense network based on the first image.
The functions of the above units are referred to in the corresponding methods, and are not described in detail herein.
The imaging method flow chart is shown in fig. 3, firstly, a sampling strategy of a data set similar to a target pattern is learned a priori through a computer, a corresponding mask pattern is generated, and the number of generated stripes is equal to the sampling frequency of a pixel image in transformation multiplied by the sampling rate (the sampling rate of the experiment is 8%); selecting a corresponding data set for image scaling, and selecting corresponding coefficients of a processed image transformation domain for inverse transformation aiming at a sampling strategy of prior learning to form a training set for initializing a residual dense network; building a terahertz single-pixel imaging system, and loading a mask pattern on the DMD; secondly, a target image is placed at a certain distance from the DMD by utilizing the principle of near-field imaging and is tightly attached to the terahertz modulator, the mask pattern on the DMD is reflected to the target by 808nm laser, and the laser intensity of the target pattern is modulated by the terahertz light irradiated on the terahertz modulator; the light intensity value of each sample count is collected by a detector, and the transform domain information is restored by a computer and decoded into a 32x32 pixel image, which has a low resolution due to scattering. And finally, mapping the low-resolution image and the high-resolution image through a residual dense network to recover a high-definition image of 64x64 pixels, so as to realize the effect of reconstructing the under-sampled high-resolution image.
The effect of the sample under different sampling modes is compared by experiments below. In order to preserve most of the information as much as possible and avoid excessive detail loss, the experiment samples the 32 × 32 pixel sample pattern at 8% sampling rate, and introduces an effect map at 10% sampling rate and a corresponding optimal sampling mode as comparison. In fig. 4 and table 1, the corresponding sampling modes are circular, square, prior statistics, and optimal sampling in sequence. The invention adopts two evaluation indexes of Structural SIMilarity (SSIM) and Peak Signal to Noise Ratio (PSNR). As can be seen from table 1, the prior statistics at 8% sampling rate is similar to the 10% block mode, and is better than the other two sampling modes, and is slightly weaker than the optimal sampling mode of the same sampling.
TABLE 1 comparison of the effectiveness of the samples at 8% and 10% for different sampling regimes
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A terahertz single-pixel super-resolution imaging method is characterized by comprising the following steps:
s100, carrying out prior learning on the coding positions of the similar patterns, extracting a sampling strategy stored through the prior learning, and generating mask patterns, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction;
s200, selecting a data set to perform corresponding processing based on the sampling strategy to generate a training set, and using the training set for initializing a residual error dense network;
s300, building a terahertz single-pixel imaging system, loading the mask pattern by adopting a digital micromirror array according to the sampling strategy, and reflecting the mask pattern onto a target pattern through laser; the terahertz wave irradiates a terahertz modulator, the laser intensity of a target pattern is modulated, the terahertz wave is periodically received through a terahertz detector, transform domain information is recovered from the intensity information, and a first image is obtained through decoding;
s400, reconstructing a second image through the residual error dense network based on the first image.
2. The terahertz single-pixel super-resolution imaging method as claimed in claim 1, wherein the step S100 specifically comprises the following steps:
s101, classifying the similar pictures of the collected target patterns to form a data set;
s102, carrying out full sampling on a plurality of pictures of the same type and storing transform domain data;
s103, counting the corresponding transform domain position of the module coefficient intensity under the specific sampling rate in the transform domain of the image by adopting a statistical method, thereby generating the mask pattern of the image.
3. The terahertz single-pixel super-resolution imaging method as claimed in claim 2, wherein in step S102, inverse fourier domain transform is applied to the transform domain.
4. The terahertz single-pixel super-resolution imaging method as claimed in claim 1, wherein the step S400 specifically comprises the following steps:
s401, extracting a shallow information characteristic graph from an input image through a convolutional layer, and then entering a residual error dense block;
s402, generating a corresponding feature map by each residual error dense block, splicing all the feature maps to form a whole, and performing dense feature fusion;
and S403, interpolating the first image to a high-pixel second image through an upsampling module, and reconstructing the second image to restore the second image to a single-channel high-definition gray image or a three-channel high-definition color image.
5. The terahertz single-pixel super-resolution imaging method according to claim 4, wherein the first image is a 32x32 pixel image and the second image is a 64x64 pixel image.
6. The terahertz single-pixel super-resolution imaging method as claimed in claim 5, wherein the dense feature fusion process comprises global feature fusion and global residual learning, the feature maps are firstly combined into a whole by using fusion layers, deep information is further acquired through convolution operation and restored into a feature map of 64 channels, and finally, shallow information and deep information are added through residual learning.
7. A terahertz single-pixel super-resolution imaging system is characterized by comprising:
the learning unit is used for carrying out prior learning on the coding positions of the similar patterns, extracting a sampling strategy stored in the prior learning and generating mask patterns, wherein the number of the mask patterns is equal to the product of the sampling times and the sampling rate of the pixel image in single-pixel imaging reconstruction;
the training unit is used for selecting a corresponding data set to carry out image scaling, simultaneously selecting corresponding coefficients of a processed image transformation domain to carry out inverse transformation based on the sampling strategy to form a training set, and using the training set for initializing a residual dense network;
the acquisition unit loads a mask pattern learned in advance by adopting a digital micromirror array and reflects the mask pattern onto a target pattern through laser; irradiating the terahertz wave to a terahertz modulator, modulating the laser intensity of a target pattern, periodically receiving the terahertz wave by a terahertz detector, restoring transform domain information from the terahertz wave, and decoding to obtain a first image;
and the reconstruction unit is used for reconstructing a second image through the residual error dense network based on the first image.
8. The terahertz single-pixel super-resolution imaging system of claim 7, wherein the learning unit comprises:
the classification module is used for classifying the similar pictures of the acquired target pattern to form a data set;
the full sampling module is used for performing full sampling on a plurality of pictures of the same type and storing transform domain data;
and the generation module is used for counting the position of the corresponding transform domain in the transform domain of the image by adopting a statistical method, wherein the intensity of the module coefficient meets the requirement of the position of the corresponding transform domain under a specific sampling rate, so that the mask pattern of the image is generated.
9. The terahertz single-pixel super-resolution imaging system of claim 7, wherein the reconstruction unit comprises:
the convolution module extracts a shallow information characteristic graph from an input image through a convolution layer and then enters a residual error dense block;
the fusion module is used for generating corresponding feature maps by each residual error dense block, splicing all the feature maps to form a whole and carrying out dense feature fusion;
and the interpolation module is used for interpolating the first image into a high-pixel second image through the up-sampling module and reconstructing the second image to restore the second image into a single-channel high-definition gray image or a three-channel high-definition color image.
10. The terahertz single-pixel super-resolution imaging system of claim 9, wherein the first image is a 32x32 pixel image and the second image is a 64x64 pixel image.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111532798.0A CN114387164A (en) | 2021-12-15 | 2021-12-15 | Terahertz single-pixel super-resolution imaging method and system |
| PCT/CN2022/138217 WO2023109719A1 (en) | 2021-12-15 | 2022-12-09 | Terahertz single-pixel super-resolution imaging method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111532798.0A CN114387164A (en) | 2021-12-15 | 2021-12-15 | Terahertz single-pixel super-resolution imaging method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114387164A true CN114387164A (en) | 2022-04-22 |
Family
ID=81198426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111532798.0A Pending CN114387164A (en) | 2021-12-15 | 2021-12-15 | Terahertz single-pixel super-resolution imaging method and system |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN114387164A (en) |
| WO (1) | WO2023109719A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117173503B (en) * | 2023-08-16 | 2024-04-19 | 安徽派睿太赫兹医疗器械技术开发有限公司 | Fuzzy terahertz image recognition method and device based on deep learning and electronic equipment |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013024842A (en) * | 2011-07-26 | 2013-02-04 | Hitachi High-Tech Control Systems Corp | Terahertz wave imaging device |
| CN105374020B (en) * | 2015-12-17 | 2018-04-17 | 深圳职业技术学院 | A kind of ultrasonic imaging detection method of fast hi-resolution |
| CN110231292B (en) * | 2019-06-28 | 2022-02-11 | 深圳先进技术研究院 | Single-pixel terahertz wave imaging method and system |
| CN112525851A (en) * | 2020-12-10 | 2021-03-19 | 深圳先进技术研究院 | Terahertz single-pixel imaging method and system |
| CN113506222B (en) * | 2021-07-30 | 2024-03-01 | 合肥工业大学 | Multi-mode image super-resolution method based on convolutional neural network |
| CN114387164A (en) * | 2021-12-15 | 2022-04-22 | 深圳先进技术研究院 | Terahertz single-pixel super-resolution imaging method and system |
-
2021
- 2021-12-15 CN CN202111532798.0A patent/CN114387164A/en active Pending
-
2022
- 2022-12-09 WO PCT/CN2022/138217 patent/WO2023109719A1/en unknown
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023109719A1 (en) | 2023-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023109719A1 (en) | Terahertz single-pixel super-resolution imaging method and system | |
| WO2023193461A1 (en) | Terahertz single-pixel imaging method and system | |
| CN111340183B (en) | Deep learning-based scatter imaging device and method | |
| CN111343376B (en) | Transmission-type double-slit-based aperture coding imaging system and super-resolution method thereof | |
| CN112116065A (en) | RGB image spectrum reconstruction method, system, storage medium and application | |
| CN111008936B (en) | Multispectral image panchromatic sharpening method | |
| CN111563562B (en) | Color target reconstruction method of single-frame scattering image based on convolutional neural network | |
| CN112525851A (en) | Terahertz single-pixel imaging method and system | |
| CN111028302B (en) | Compressed object imaging method and system based on deep learning | |
| Zhang et al. | Deep unfolding for singular value decomposition compressed ghost imaging | |
| CN116029902A (en) | Knowledge distillation-based unsupervised real world image super-resolution method | |
| US11609332B2 (en) | Method and apparatus for generating image using LiDAR | |
| CN112113661B (en) | Deep learning type snapshot spectrum imaging device and detection method thereof | |
| CN116091640B (en) | Remote sensing hyperspectral reconstruction method and system based on spectrum self-attention mechanism | |
| CN113705340A (en) | Deep learning change detection method based on radar remote sensing data | |
| CN102735347B (en) | Method and device for infrared gaze tomography of target tracking | |
| Yin et al. | A deep network based on wavelet transform for image compressed sensing | |
| Li et al. | A brief review on 2D and 3D image reconstruction using single-pixel imaging | |
| CN115375540A (en) | Terahertz image super-resolution method based on deep learning algorithm | |
| CN115330631A (en) | Multi-scale fusion defogging method based on stacked hourglass network | |
| Chen et al. | Improved U-Net3+ with spatial–spectral transformer for multispectral image reconstruction | |
| Song et al. | Single-pixel imaging based on deep learning | |
| CN110044474B (en) | Correlation imaging optimization method based on interpolation algorithm | |
| CN112203068A (en) | Single-pixel imaging method, system, device and medium | |
| CN116310959B (en) | Method and system for identifying low-quality camera picture in complex scene |
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 |