A super-resolution terahertz single-pixel imaging method, system, device and medium

Abstract

The present invention relates to the field of terahertz imaging technology, and discloses a super-resolution terahertz single-pixel imaging method, system, device and medium, the method comprising: emitting pump light to a modulator, and forming an imaging area on a first side of the modulator; emitting terahertz waves to an object to be imaged, and forming a terahertz diffraction image on a second side of the modulator; encoding the terahertz diffraction image based on the imaging area to form one-dimensional under-sampling data of the object to be imaged; cyclically utilizing a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data, obtaining a reconstructed image, calculating a loss function value of the neural network, updating the parameters of the neural network, until the neural network converges, obtaining a reconstructed image when the neural network converges, and determining it as a terahertz single-pixel image of the object to be imaged. The present invention can save data costs, realize efficient image reconstruction, and thus improve the efficiency of Hertz imaging.

Description

Technical Field

[0001] The present invention relates to the field of terahertz imaging technology, and in particular to a super-resolution terahertz single-pixel imaging method, system, equipment and medium. Background Art

[0002] Terahertz waves have non-ionizing photon energy and are transparent to most non-polar materials, so they show great application potential in biomedical imaging, security inspection and non-destructive testing. However, the long wavelength of terahertz is restricted by the diffraction limit, which limits the imaging resolution to above the millimeter level and makes it difficult to capture micron-level details. Due to the very rough diffraction limit caused by the relatively long terahertz wavelength, far-field operation seems to be difficult to perform high-resolution terahertz imaging in all cases. In order to overcome these limitations and obtain near-field single-pixel encoded terahertz images, the existing technology uses very thin semiconductor wafers as photogenerated carrier-type spatial terahertz wave modulators in terahertz time-domain spectroscopy systems, and uses nonlinear crystals or nanofilms to directly encode spatial masks onto pump light to directly detect terahertz images or generate structured terahertz waves, which allows the sub-diffraction spatial information hidden in the diffraction far-field distribution to be recovered from the intensity recorded by the single-pixel detector.

[0003] However, the existing technology has the following defects: the existing technology requires very thin semiconductor crystals or nonlinear crystals to achieve super-resolution terahertz single-pixel imaging technology. However, there is always a certain propagation distance between the terahertz source plane and the object plane, and terahertz diffraction occurs in the entire volume of the crystal, causing the terahertz image of the object to be imaged to blur rapidly when propagating in the crystal, resulting in a decrease in the spatial resolution of the terahertz image. The existing technology requires the use of supervised learning to achieve terahertz single-pixel imaging based on deep learning, which requires a mapping function derived from the statistical information of a large number of sample data sets, thereby consuming a lot of computing resources and training time to train the model. Summary of the invention

[0004] The present invention provides a super-resolution terahertz single-pixel imaging method, system, device and medium, which can save data costs, achieve efficient image reconstruction, and thus improve the efficiency of Hertz imaging.

[0005] In order to solve the above technical problems, the present invention provides a super-resolution terahertz single-pixel imaging method, comprising:

[0006] emitting pump light to the modulator and forming an imaging region on a first side of the modulator;

[0007] emitting a terahertz wave toward an object to be imaged and forming a terahertz diffraction image on a second side of the modulator;

[0008] Based on the imaging area, encoding the terahertz diffraction image to form one-dimensional under-sampled data of the object to be imaged;

[0009] Recyclingly using a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculating a loss function value of the neural network according to the reconstructed image, and then updating the parameters of the neural network according to the loss function value until the neural network converges;

[0010] When the neural network converges, a reconstructed image output by the neural network is acquired, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

[0011] The present invention emits pump light and terahertz waves to both sides of the modulator according to a sampling scheme, encodes the terahertz diffraction image using the imaging area formed by the pump light on the modulator, and obtains one-dimensional under-sampling data of the object to be imaged. The present invention can complete image analysis and image reconstruction using the one-dimensional under-sampling data, saving data processing costs. According to the one-dimensional under-sampling data, the neural network is cyclically used to reconstruct the terahertz diffraction image, and the loss function value of the neural network is calculated after the reconstructed image is obtained, so as to update the parameters of the neural network according to the loss function value until the neural network converges, and the cyclic operation of the neural network is stopped, and the reconstructed image output by the neural network at this time is determined as the terahertz single-pixel image of the object to be imaged. The present invention uses a neural network that has not been trained with samples to reconstruct the terahertz diffraction image, which can effectively save model pre-training time, realize efficient image reconstruction, and thus improve the efficiency of terahertz imaging.

[0012] Further, the step of emitting pump light to the modulator according to the sampling scheme and forming an imaging area on the modulator is specifically:

[0013] determining a first wavelength of the pump light in a preset sampling scheme;

[0014] Based on the first wavelength, emitting corresponding pump light to the modulator to form a pump light image on the modulator;

[0015] The area on the modulator where the pump light image is located is determined as an imaging area.

[0016] Furthermore, the emitting of terahertz waves to the object to be imaged according to the sampling scheme and forming a terahertz diffraction image on the modulator is specifically:

[0017] determining a second wavelength of the terahertz wave in a preset sampling scheme;

[0018] Based on the second wavelength, a corresponding terahertz wave is emitted to the object to be imaged, and the terahertz wave is transmitted through the object to be imaged to the modulator to form a terahertz diffraction image.

[0019] Furthermore, based on the imaging area, the terahertz diffraction image is encoded to form one-dimensional under-sampling data of the object to be imaged, specifically:

[0020] Acquiring a mask spatial response of the imaging area;

[0021] decomposing the mask space response into a positive mask space response and a negative mask space response;

[0022] Based on the positive mask spatial response and the negative mask spatial response, the terahertz diffraction image is encoded to obtain a single pixel intensity value of the image of the object to be imaged:

[0023] The single pixel intensity value of the image of the object to be imaged is determined as one-dimensional under-sampling data of the object to be imaged.

[0024] Furthermore, the use of a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image is specifically:

[0025] Using a single pixel imaging method, a rough intensity estimate of the terahertz diffraction image is calculated according to the mask spatial response and the single pixel intensity value of the image;

[0026] A rough intensity estimate of the terahertz diffraction image is input into a neural network framework to obtain an object light field intensity estimate of the terahertz diffraction image, and a reconstructed image is formed based on the object light field intensity estimate of the terahertz diffraction image.

[0027] Furthermore, the loss function value of the neural network is calculated according to the reconstructed image, and then the parameters of the neural network are updated according to the loss function value, specifically:

[0028] calculating the intensity distribution of the reconstructed image in combination with the terahertz field distribution of the modulator;

[0029] performing a convolution operation on the intensity distribution of the reconstructed image and the mask spatial response to obtain a one-dimensional intensity estimation of the reconstructed image;

[0030] Calculating a loss function value of the neural network based on the single pixel intensity value of the image and the one-dimensional intensity estimate;

[0031] The parameters of the neural network are updated according to the loss function value.

[0032] Furthermore, the intensity distribution of the reconstructed image is calculated in combination with the terahertz field distribution of the modulator, specifically:

[0033] Simulating the transmission of the reconstructed image to the modulator, and analyzing the terahertz field distribution of the modulator; wherein the terahertz field distribution includes a uniform plane wave field component and an evanescent wave field component;

[0034] Based on the reconstructed image and the terahertz field distribution, an intensity distribution of the reconstructed image is calculated.

[0035] Accordingly, the present invention provides a super-resolution terahertz single-pixel imaging system, comprising: a terahertz source module, a pump light module, a spatial light modulation module, an object module to be imaged, a terahertz modulation module, a terahertz detection module, a signal acquisition module and a signal processing module;

[0036] Controlling the pump light module to emit pump light to the spatial light modulation module;

[0037] The spatial light modulation module is used to project the pump light onto the terahertz modulation module through the terahertz detection module, and form an imaging area on a first side of the terahertz modulation module;

[0038] Controlling the terahertz source module to transmit terahertz waves to the object module to be imaged, so that the terahertz waves pass through the object module to be imaged to form a terahertz diffraction image on the second side of the terahertz modulation module;

[0039] The terahertz modulation module encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampled data of the object to be imaged;

[0040] The signal acquisition module acquires the one-dimensional under-sampling data through the terahertz detection module;

[0041] The signal processing module is used to cyclically use the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data, obtain a reconstructed image, and calculate the loss function value of the neural network according to the reconstructed image, and then update the parameters of the neural network according to the loss function value until the neural network converges; when the neural network converges, obtain the reconstructed image output by the neural network, and determine the reconstructed image as the terahertz single-pixel image of the object to be imaged.

[0042] The present invention also provides a super-resolution terahertz single-pixel imaging device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the super-resolution terahertz single-pixel imaging method as described above is implemented.

[0043] The present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the super-resolution terahertz single-pixel imaging method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of a process flow of an embodiment of the super-resolution terahertz single-pixel imaging method provided by the present invention;

[0045] Figure 2 A schematic structural diagram of an embodiment of a super-resolution terahertz single-pixel imaging system provided by the present invention;

[0046] Figure 3 A schematic structural diagram of another embodiment of the super-resolution terahertz single-pixel imaging system provided by the present invention;

[0047] Figure 4 A schematic structural diagram of an embodiment of a super-resolution terahertz single-pixel imaging device provided by the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0050] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0051] Example 1

[0052] like Figure 1 FIG. 1 is a flow chart of an embodiment of a super-resolution terahertz single-pixel imaging method provided by the present invention. The method includes steps 101 to 105, and each step is specifically as follows:

[0053] Step 101: emit pump light to a modulator and form an imaging region on a first side of the modulator.

[0054] Step 102: emitting a terahertz wave to the object to be imaged, and forming a terahertz diffraction image on the second side of the modulator.

[0055] Step 103: Based on the imaging area, encode the terahertz diffraction image to form one-dimensional under-sampling data of the object to be imaged.

[0056] Step 104: cyclically utilize the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculate the loss function value of the neural network according to the reconstructed image, and then update the parameters of the neural network according to the loss function value until the neural network converges.

[0057] Step 105: When the neural network converges, a reconstructed image output by the neural network is obtained, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

[0058] In an embodiment of the present invention, pump light is emitted to the modulator according to a sampling scheme, and an imaging area is formed on one side of the modulator, and the object to be imaged is imaged in the imaging area to realize terahertz imaging. When emitting pump light, a first wavelength of the pump light can be obtained from the sampling scheme, and then the pump light is emitted to one side of the modulator according to the first wavelength, thereby forming an imaging area of ​​the object to be imaged.

[0059] In an embodiment of the present invention, a terahertz wave is emitted so that the terahertz wave passes through the object to be imaged and maps a terahertz diffraction image on the other side of the modulator. The terahertz diffraction image is encoded according to the imaging area formed by the pump light on the modulator to obtain one-dimensional under-sampled data of the object to be imaged. The present invention can complete image analysis and image reconstruction using one-dimensional under-sampled data, and does not need to perform full-wave measurement of the terahertz pulse in the time domain, saving data sampling time.

[0060] In an embodiment of the present invention, a neural network is cyclically used to reconstruct a terahertz diffraction image based on one-dimensional under-sampling data, and the loss function value of the neural network is calculated after the reconstructed image is obtained, thereby updating the parameters of the neural network according to the loss function value until the neural network converges, and the cyclic operation of the neural network is stopped, and the reconstructed image output by the neural network at this time is determined as the terahertz single-pixel image of the object to be imaged. The present invention uses a neural network that has not been trained with samples to reconstruct the terahertz diffraction image, which can effectively save model pre-training time, achieve efficient image reconstruction, and thus improve the efficiency of terahertz imaging.

[0061] In summary, the embodiment of the present invention provides a super-resolution terahertz single-pixel imaging method, which emits pump light to a modulator according to a sampling scheme and forms an imaging area on the first side of the modulator; emits terahertz waves to an object to be imaged according to a sampling scheme and forms a terahertz diffraction image on the second side of the modulator; encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampling data of the object to be imaged; recycles the terahertz diffraction image based on the one-dimensional under-sampling data using a neural network to reconstruct the terahertz diffraction image, calculates the loss function value of the neural network, updates the parameters of the neural network until the neural network converges, obtains the reconstructed image when the neural network converges, and determines it as the terahertz single-pixel image of the object to be imaged. The present invention can save data costs, realize efficient image reconstruction, and thus improve the efficiency of Hertz imaging.

[0062] Example 2

[0063] like Figure 1 FIG. 1 is a flow chart of an embodiment of a super-resolution terahertz single-pixel imaging method provided by the present invention. The method includes steps 101 to 105, and each step is specifically as follows:

[0064] Step 101: emit pump light to a modulator and form an imaging region on a first side of the modulator.

[0065] Further, in an embodiment of the present invention, pump light is emitted to the modulator according to a sampling scheme, and an imaging area is formed on the modulator, specifically:

[0066] determining a first wavelength of the pump light in a preset sampling scheme;

[0067] Based on the first wavelength, emitting corresponding pump light to the modulator to form a pump light image on the modulator;

[0068] The area on the modulator where the pump light image is located is determined as an imaging area.

[0069] In an embodiment of the present invention, a first wavelength of the pump light is recorded in the preset sampling scheme. According to the first wavelength in the preset sampling scheme, the pump light with the first wavelength is emitted to the first side of the modulator, thereby forming an imaging area, so that the object to be imaged realizes terahertz imaging in the imaging area.

[0070] Step 102: emitting a terahertz wave to the object to be imaged, and forming a terahertz diffraction image on the second side of the modulator.

[0071] Furthermore, in an embodiment of the present invention, a terahertz wave is emitted to the object to be imaged according to a sampling scheme, and a terahertz diffraction image is formed on the modulator, specifically:

[0072] determining a second wavelength of the terahertz wave in a preset sampling scheme;

[0073] Based on the second wavelength, a corresponding terahertz wave is emitted to the object to be imaged, and the terahertz wave is transmitted through the object to be imaged to the modulator to form a terahertz diffraction image.

[0074] In the embodiment of the present invention, the second wavelength of the terahertz wave is recorded in the preset sampling scheme. According to the second wavelength in the preset sampling scheme, the terahertz field with the second wavelength is vertically incident on On the object plane of the imaged object, the complex amplitude distribution of the terahertz transmission field is expressed as ,in, Indicated in The complex amplitude distribution of the terahertz transmission field formed on the plane, and Represents the complex amplitude distribution The spatial coordinates of The initial position plane of the terahertz field is the object plane to be imaged. The object to be imaged propagates to The modulator surface leads to a completely different physical situation, which generates the terahertz diffraction image. ,in, It represents the modulator plane that the terahertz field reaches after propagating from the object plane to be imaged through a distance d, where d represents the propagation distance. Indicated in Complex amplitude distribution of the terahertz transmission field formed on the plane, x and y represent the complex amplitude distribution The spatial coordinates of Represents the generated terahertz diffraction image.

[0075] Step 103: Based on the imaging area, encode the terahertz diffraction image to form one-dimensional under-sampling data of the object to be imaged.

[0076] Further, in an embodiment of the present invention, based on the imaging area, the terahertz diffraction image is encoded to form one-dimensional under-sampling data of the object to be imaged, specifically:

[0077] Acquiring a mask spatial response of the imaging area;

[0078] decomposing the mask space response into a positive mask space response and a negative mask space response;

[0079] Based on the positive mask spatial response and the negative mask spatial response, the terahertz diffraction image is encoded to obtain a single pixel intensity value of an image of the object to be imaged;

[0080] The single pixel intensity value of the image of the object to be imaged is determined as one-dimensional under-sampling data of the object to be imaged.

[0081] In an embodiment of the present invention, the imaging area on the modulator can control the terahertz wave by inducing the conductivity change of high-resistance silicon by pump light. The imaging area irradiated by the pump light is conductive. In order to achieve single-pixel measurement of the terahertz diffraction pattern, the Walsh-Hadamard matrix encoding scheme can be used to encode the terahertz wave of the imaging area as "0", while other areas remain transparent and encode the terahertz wave as "1".

[0082] In the embodiment of the present invention, considering the influence of strong background noise level in the actual measurement of the detector, the mask on the modulator can be Decompose into positive mask and negative mask, encode the terahertz diffraction image on the modulator plane, and form the single pixel intensity value of the image of the object to be imaged, specifically:

[0083]

[0084] In the formula, is the single pixel intensity value of the image; is the terahertz diffraction image; is the positive mask spatial response; is the negative mask spatial response.

[0085] in, is the spatial response of the ith mask. Calculate the positive mask spatial response, using the formula Compute the negative mask spatial response. Therefore, for a series of M masks , the encoding process of the terahertz diffraction image of the object to be imaged can be written as a matrix convolution operation form , and get a one-dimensional vector of length M As the direct input of the untrained neural network. For a differential mask with M number to reconstruct an image with N pixels, the compression ratio is defined as M / N.

[0086] Step 104: cyclically utilize the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculate the loss function value of the neural network according to the reconstructed image, and then update the parameters of the neural network according to the loss function value until the neural network converges.

[0087] Further, in an embodiment of the present invention, a neural network is used to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, specifically:

[0088] Using a single pixel imaging method, a rough intensity estimate of the terahertz diffraction image is calculated according to the mask spatial response and the single pixel intensity value of the image;

[0089] A rough intensity estimate of the terahertz diffraction image is input into a neural network framework to obtain an object light field intensity estimate of the terahertz diffraction image, and a reconstructed image is formed based on the object light field intensity estimate of the terahertz diffraction image.

[0090] In the embodiment of the present invention, after obtaining the one-dimensional under-sampled data of the object to be imaged, the one-dimensional under-sampled data is used as the direct input of the untrained neural network. In order to solve the reconstruction uncertainty caused by the under-sampling strategy, an intermediate physical coding layer can be constructed as the indirect input of the neural network, so that the neural network has better noise robustness. Specifically, a single pixel imaging method can be used to calculate the spatial response of the mask. And the image single pixel intensity value , a rough intensity estimate of the THz diffraction image is calculated using the second-order correlation method, specifically:

[0091]

[0092] In the formula, For a rough strength estimate; is the mask space response; is the single pixel intensity value of the normalized difference image; is the average value of the set, that is .

[0093] In the embodiment of the present invention, four convolution blocks with depths of 32, 16, 8 and 1 respectively may be used to form a neural network, and a normalization layer and a nonlinear activation layer are included between the convolution layers. As the input of the neural network, the output of the neural network can be obtained as the object light field intensity estimation of the terahertz diffraction image The object light field intensity estimate of the terahertz diffraction image is obtained to form the reconstructed image , the reconstructed image can be expressed as: .in, is a parameterized neural network, are the weights and biases of the network.

[0094] In the embodiment of the present invention, by constructing a propagation model, the process of transmitting the reconstructed image to the modulator can be simulated, thereby analyzing the terahertz field distribution of the modulator. The terahertz field distribution includes a uniform plane wave field component and an evanescent wave field component; based on the reconstructed image and the terahertz field distribution, the intensity distribution of the reconstructed image can be calculated.

[0095] Specifically, the simulation will reconstruct the image Transfer to The modulator plane, and the terahertz field distribution of the modulator Including uniform plane wave field components and evanescent wave field component , calculated according to the scalar diffraction integral formula of angular spectrum propagation theory:

[0096]

[0097]

[0098] In the formula, λ is the wavelength of the terahertz wave; k is the wave number of the terahertz wave; d is the diffraction propagation distance; Object Light Field The spatial spectrum of the 2D Fourier transform; and is the spatial frequency along the x and y directions; and is the normalized spatial frequency, where is the normalized spatial frequency along the x-direction, is the normalized spatial frequency along the y direction, and λ is the wavelength of the terahertz wave. Reconstructed image After propagating a distance d to reach the silicon modulator plane, it can be calculated according to the formula Calculate the intensity distribution of the reconstructed image .

[0099] Furthermore, in an embodiment of the present invention, the loss function value of the neural network is calculated according to the reconstructed image, and then the parameters of the neural network are updated according to the loss function value, specifically:

[0100] calculating the intensity distribution of the reconstructed image in combination with the terahertz field distribution of the modulator;

[0101] performing a convolution operation on the intensity distribution of the reconstructed image and the mask spatial response to obtain a one-dimensional intensity estimation of the reconstructed image;

[0102] Calculating a loss function value of the neural network based on the single pixel intensity value of the image and the one-dimensional intensity estimate;

[0103] The parameters of the neural network are updated according to the loss function value.

[0104] In the embodiment of the present invention, after calculating the intensity distribution of the reconstructed image Then, compare it with the modulator's mask spatial response Perform a convolution operation to obtain a one-dimensional intensity estimate of the reconstructed image, expressed as By calculating the one-dimensional intensity estimate And the image single pixel intensity value The mean square error between them is obtained, so the loss function of the neural network is expressed as Based on the calculated loss function value, the weights and biases of the neural network are optimized by gradient descent using the Adma optimizer to minimize the loss function.

[0105] In the embodiment of the present invention, the terahertz diffraction image is reconstructed by cyclically utilizing the neural network reconstruction, thereby updating the weights and biases of the neural network. As the neural network optimization iteration proceeds, the one-dimensional intensity estimate output by the neural network is Converge to the actual measured value I. When the neural network converges, it can indirectly capture the experimental measurement value I and the object light field intensity estimate The mapping relationship between them can be optimized to restore the amplitude distribution of the object light field by the updated neural network. Therefore, under the constraint of the physical propagation model, the neural network acts as a back-propagation operator to refocus the terahertz diffraction pattern of the defocused plane. This embodiment is not only applicable to sub-diffraction resolution image reconstruction when the sample is in close contact with the modulator plane, but also to the case where the object plane moves beyond the terahertz near-field region.

[0106] In an embodiment of the present invention, during the optimization process of the neural network, there is no need to pre-train a large number of labeled data sets. It directly acts on the single-pixel measurement intensity value I of the terahertz diffraction image, which can effectively save the model pre-training time and achieve efficient image reconstruction, thereby improving the efficiency of terahertz imaging.

[0107] Step 105: When the neural network converges, a reconstructed image output by the neural network is obtained, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

[0108] In an embodiment of the present invention, when the neural network converges, the reconstructed image output by the neural network is determined as the terahertz single-pixel image of the object to be imaged. The present invention combines the propagation model formed by the untrained neural network and diffraction, sets the reverse propagation distance to the thickness of the high-resistance silicon modulator, and can achieve high-quality super-resolution terahertz images under undersampling.

[0109] In summary, the embodiment of the present invention provides a super-resolution terahertz single-pixel imaging method, which emits pump light to a modulator according to a sampling scheme and forms an imaging area on the first side of the modulator; emits terahertz waves to an object to be imaged according to a sampling scheme and forms a terahertz diffraction image on the second side of the modulator; encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampling data of the object to be imaged; recycles the terahertz diffraction image based on the one-dimensional under-sampling data using a neural network to reconstruct the terahertz diffraction image, calculates the loss function value of the neural network, updates the parameters of the neural network until the neural network converges, obtains the reconstructed image when the neural network converges, and determines it as the terahertz single-pixel image of the object to be imaged. The present invention can save data costs, realize efficient image reconstruction, and thus improve the efficiency of Hertz imaging.

[0110] Example 3

[0111] See also Figure 2-Figure 3 The present invention provides a super-resolution terahertz single-pixel imaging system, which includes a terahertz source module 10, a pump light module 20, a spatial light modulation module 30, an object module to be imaged 40, a terahertz modulation module 50, a terahertz detection module 60, a signal acquisition module 70 and a signal processing module 80;

[0112] Controlling the pump light module 20 to emit pump light to the spatial light modulation module 30;

[0113] The spatial light modulation module 30 is used to project the pump light onto the terahertz modulation module 50 through the terahertz detection module 60, and form an imaging area on a first side of the terahertz modulation module 50;

[0114] Controlling the terahertz source module 10 to emit terahertz waves to the object module 40 to be imaged, so that the terahertz waves pass through the object module 40 to form a terahertz diffraction image on the second side of the terahertz modulation module 50;

[0115] The terahertz modulation module 50 encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampled data of the object to be imaged;

[0116] The signal acquisition module 70 acquires the one-dimensional under-sampling data through the terahertz detection module 60;

[0117] The signal processing module 80 is used to cyclically use the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data, obtain a reconstructed image, and calculate the loss function value of the neural network according to the reconstructed image, and then update the parameters of the neural network according to the loss function value until the neural network converges; when the neural network converges, obtain the reconstructed image output by the neural network, and determine the reconstructed image as the terahertz single-pixel image of the object to be imaged.

[0118] In the embodiment of the present invention, along the optical path, the terahertz source module 10 includes a terahertz source 11 and a terahertz beam expander 12. The terahertz source 11 is used to emit continuous terahertz waves. The operating frequency of the terahertz source is selected according to the actual imaging system parameters, and the operating frequency of the terahertz source 11 is 336 GHz. The terahertz beam expander 12 is used to collimate and expand the terahertz waves emitted by the terahertz source 11, and the imaging object module 40 is used to receive the expanded terahertz waves and reach the surface of the terahertz modulation module 50.

[0119] In the embodiment of the present invention, the terahertz modulation module 50 includes but is not limited to at least one of a nonlinear crystal and a semiconductor crystal modulator. The terahertz modulation module 50 is a high-resistance silicon modulator with a thickness within the terahertz wavelength, which is used to obtain near-field terahertz evanescent waves. The high-resistance silicon modulator is configured to spatially modulate the transmitted terahertz wave to achieve spatial encoding of the terahertz wave, and to achieve terahertz imaging in the imaging area of ​​the terahertz modulation module 50.

[0120] In the embodiment of the present invention, the imaging area is generated by the pump light module 20 and the spatial light modulation module 30. Along the optical path, the pump light module 20 includes a 532 nm continuous laser 21 and a visible light beam expander 22. The spatial light modulation module 30 includes a digital micromirror array device 31 and a projection lens 32. The projection lens 32 projects the pump light encoded by the digital micromirror array device 31 to the other side of the terahertz modulation module 50, thereby realizing the spatial encoding of the terahertz wave.

[0121] In the embodiment of the present invention, the terahertz detection module 60 is arranged at the output end of the terahertz modulation module 50. The terahertz detection module 60 includes an off-axis parabolic mirror 61 with a light hole and a terahertz single-pixel detector 62. The light hole of the off-axis parabolic mirror 61 projects the pump light from the projection lens 32 to the terahertz modulation module 50 to realize the spatial encoding of the terahertz wave. The parabola of the off-axis parabolic mirror 61 focuses the terahertz wave passing through the terahertz modulation module 50 to the terahertz single-pixel detector 62. The signal acquisition module 70 is connected to the terahertz single-pixel detector 62, and the collected terahertz signal is transmitted to the signal processing module 80 to realize the super-resolution terahertz single-pixel imaging of the object 40 to be imaged.

[0122] In summary, an embodiment of the present invention provides a super-resolution terahertz single-pixel imaging system, which is based on the organic combination of modules, controls the pump light module to emit pump light to the spatial light modulation module; the spatial light modulation module is used to project the pump light onto the terahertz modulation module through the terahertz detection module, and forms an imaging area on the first side of the terahertz modulation module; controls the terahertz source module to emit terahertz waves to the object module to be imaged, so that the terahertz waves pass through the object module to be imaged to form a terahertz diffraction image on the second side of the terahertz modulation module; the terahertz modulation module encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampling data of the object to be imaged; the signal acquisition module obtains the one-dimensional under-sampling data through the terahertz detection module; the signal processing module is used to cyclically use the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculates the loss function value of the neural network according to the reconstructed image, and then updates the parameters of the neural network according to the loss function value until the neural network converges; when the neural network converges, obtains the reconstructed image output by the neural network, and determines the reconstructed image as the terahertz single-pixel image of the object to be imaged. The present invention can save data costs, realize efficient image reconstruction, and further improve the efficiency of Hertz imaging.

[0123] Example 4

[0124] See also Figure 4 , is a structural schematic diagram of an embodiment of a super-resolution terahertz single-pixel imaging device provided by the present invention. The device can be a server, including a processor, a memory and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0125] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0126] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any super-resolution terahertz single-pixel imaging method.

[0127] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any super-resolution terahertz single-pixel imaging method.

[0128] The network interface is used for network communication, such as sending assigned tasks.

[0129] The computer program can be specifically called by the processor to enable the super-resolution terahertz single-pixel imaging device to perform the following operations:

[0130] emitting pump light to the modulator and forming an imaging region on a first side of the modulator;

[0131] emitting a terahertz wave toward an object to be imaged and forming a terahertz diffraction image on a second side of the modulator;

[0132] Based on the imaging area, encoding the terahertz diffraction image to form one-dimensional under-sampled data of the object to be imaged;

[0133] Recyclingly using a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculating a loss function value of the neural network according to the reconstructed image, and then updating the parameters of the neural network according to the loss function value until the neural network converges;

[0134] When the neural network converges, a reconstructed image output by the neural network is acquired, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

[0135] This embodiment emits pump light and terahertz waves to both sides of the modulator according to the sampling scheme, encodes the terahertz diffraction image using the imaging area formed by the pump light on the modulator, and obtains one-dimensional under-sampling data of the object to be imaged. The present invention can complete image analysis and image reconstruction using one-dimensional under-sampling data, saving data processing costs. According to the one-dimensional under-sampling data, the neural network is cyclically used to reconstruct the terahertz diffraction image, and the loss function value of the neural network is calculated after the reconstructed image is obtained, so as to update the parameters of the neural network according to the loss function value until the neural network converges, and the cyclic operation of the neural network is stopped, and the reconstructed image output by the neural network at this time is determined as the terahertz single-pixel image of the object to be imaged. The present invention uses a neural network that has not been trained with samples to reconstruct the terahertz diffraction image, which can effectively save model pre-training time, achieve efficient image reconstruction, and thus improve the efficiency of terahertz imaging.

[0136] Example 5

[0137] An embodiment of the present invention provides a computer-readable storage medium, which stores at least one executable instruction. When the executable instruction is executed on a super-resolution terahertz single-pixel imaging device, the super-resolution terahertz single-pixel imaging device executes the super-resolution terahertz single-pixel imaging method in any of the above method embodiments.

[0138] The executable instructions can be specifically used to enable the super-resolution terahertz single-pixel imaging device to perform the following operations:

[0139] emitting pump light to the modulator and forming an imaging region on a first side of the modulator;

[0140] emitting a terahertz wave toward an object to be imaged and forming a terahertz diffraction image on a second side of the modulator;

[0141] Based on the imaging area, encoding the terahertz diffraction image to form one-dimensional under-sampled data of the object to be imaged;

[0142] Recyclingly using a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculating a loss function value of the neural network according to the reconstructed image, and then updating the parameters of the neural network according to the loss function value until the neural network converges;

[0143] When the neural network converges, a reconstructed image output by the neural network is acquired, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

[0144] This embodiment emits pump light and terahertz waves to both sides of the modulator according to the sampling scheme, encodes the terahertz diffraction image using the imaging area formed by the pump light on the modulator, and obtains one-dimensional under-sampling data of the object to be imaged. The present invention can complete image analysis and image reconstruction using one-dimensional under-sampling data, saving data processing costs. According to the one-dimensional under-sampling data, the neural network is cyclically used to reconstruct the terahertz diffraction image, and the loss function value of the neural network is calculated after the reconstructed image is obtained, so as to update the parameters of the neural network according to the loss function value until the neural network converges, and the cyclic operation of the neural network is stopped, and the reconstructed image output by the neural network at this time is determined as the terahertz single-pixel image of the object to be imaged. The present invention uses a neural network that has not been trained with samples to reconstruct the terahertz diffraction image, which can effectively save model pre-training time, achieve efficient image reconstruction, and thus improve the efficiency of terahertz imaging.

[0145] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system or other device. In addition, the embodiments of the present invention are not directed to any particular programming language.

[0146] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. Similarly, in order to simplify the present invention and help understand one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. Wherein, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself is a separate embodiment of the present invention.

[0147] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and further may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive.

[0148] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

Claims

1. A super-resolution terahertz single-pixel imaging method, characterized in that: include: emitting pump light to the modulator and forming an imaging region on a first side of the modulator; emitting a terahertz wave toward an object to be imaged and forming a terahertz diffraction image on a second side of the modulator; Based on the imaging area, the terahertz diffraction image is encoded to form one-dimensional under-sampling data of the object to be imaged, specifically: Acquiring a mask spatial response of the imaging area; decomposing the mask space response into a positive mask space response and a negative mask space response; Based on the positive mask spatial response and the negative mask spatial response, the terahertz diffraction image is encoded to obtain a single pixel intensity value of the image of the object to be imaged, specifically: In the formula, is the single pixel intensity value of the image; is the terahertz diffraction image; is the positive mask spatial response; is the negative mask spatial response; in, is the spatial response of the ith mask; through the formula Calculate the positive mask spatial response, using the formula Calculate the negative mask spatial response; Determining a single pixel intensity value of the image of the object to be imaged as one-dimensional under-sampled data of the object to be imaged; Recyclingly utilizing a neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data to obtain a reconstructed image, and calculating a loss function value of the neural network according to the reconstructed image, and then updating parameters of the neural network according to the loss function value until the neural network converges; When the neural network converges, a reconstructed image output by the neural network is acquired, and the reconstructed image is determined as a terahertz single-pixel image of the object to be imaged.

2. The super-resolution terahertz single-pixel imaging method according to claim 1, characterized in that: The step of emitting pump light to the modulator and forming an imaging area on the modulator is specifically: determining a first wavelength of the pump light in a preset sampling scheme; Based on the first wavelength, emitting corresponding pump light to the modulator to form a pump light image on the modulator; The area on the modulator where the pump light image is located is determined as an imaging area.

3. The super-resolution terahertz single-pixel imaging method according to claim 1, characterized in that: The emitting of terahertz waves to the object to be imaged and forming a terahertz diffraction image on the modulator is specifically: determining a second wavelength of the terahertz wave in a preset sampling scheme; Based on the second wavelength, a corresponding terahertz wave is emitted to the object to be imaged, and the terahertz wave is transmitted through the object to be imaged to the modulator to form a terahertz diffraction image.

4. The super-resolution terahertz single-pixel imaging method according to claim 3, characterized in that: The method of reconstructing the terahertz diffraction image based on the one-dimensional under-sampling data using a neural network to obtain a reconstructed image is specifically as follows: Using a single pixel imaging method, a rough intensity estimate of the terahertz diffraction image is calculated according to the mask spatial response and the single pixel intensity value of the image; A rough intensity estimate of the terahertz diffraction image is input into a neural network framework to obtain an object light field intensity estimate of the terahertz diffraction image, and a reconstructed image is formed based on the object light field intensity estimate of the terahertz diffraction image.

5. The super-resolution terahertz single-pixel imaging method according to claim 4, characterized in that: The step of calculating the loss function value of the neural network according to the reconstructed image, and then updating the parameters of the neural network according to the loss function value, is specifically as follows: calculating the intensity distribution of the reconstructed image in combination with the terahertz field distribution of the modulator; performing a convolution operation on the intensity distribution of the reconstructed image and the mask spatial response to obtain a one-dimensional intensity estimation of the reconstructed image; Calculating a loss function value of the neural network based on the single pixel intensity value of the image and the one-dimensional intensity estimate; The parameters of the neural network are updated according to the loss function value.

6. The super-resolution terahertz single-pixel imaging method according to claim 5, characterized in that: The step of calculating the intensity distribution of the reconstructed image in combination with the terahertz field distribution of the modulator is specifically as follows: Simulating the transmission of the reconstructed image to the modulator, and analyzing the terahertz field distribution of the modulator; wherein the terahertz field distribution includes a uniform plane wave field component and an evanescent wave field component; Based on the reconstructed image and the terahertz field distribution, an intensity distribution of the reconstructed image is calculated.

7. A super-resolution terahertz single-pixel imaging system, characterized in that: A method for implementing a super-resolution terahertz single-pixel imaging method as claimed in any one of claims 1 to 6, comprising: a terahertz source module, a pump light module, a spatial light modulation module, an object module to be imaged, a terahertz modulation module, a terahertz detection module, a signal acquisition module and a signal processing module; Controlling the pump light module to emit pump light to the spatial light modulation module; The spatial light modulation module is used to project the pump light onto the terahertz modulation module through the terahertz detection module, and form an imaging area on a first side of the terahertz modulation module; Controlling the terahertz source module to transmit terahertz waves to the object module to be imaged, so that the terahertz waves pass through the object module to be imaged to form a terahertz diffraction image on the second side of the terahertz modulation module; The terahertz modulation module encodes the terahertz diffraction image based on the imaging area to form one-dimensional under-sampled data of the object to be imaged; The signal acquisition module acquires the one-dimensional under-sampling data through the terahertz detection module; The signal processing module is used to cyclically use the neural network to reconstruct the terahertz diffraction image based on the one-dimensional under-sampling data, obtain a reconstructed image, and calculate the loss function value of the neural network according to the reconstructed image, and then update the parameters of the neural network according to the loss function value until the neural network converges; when the neural network converges, obtain the reconstructed image output by the neural network, and determine the reconstructed image as the terahertz single-pixel image of the object to be imaged.

8. A super-resolution terahertz single-pixel imaging device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the super-resolution terahertz single-pixel imaging method as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the super-resolution terahertz single-pixel imaging method according to any one of claims 1 to 6.

Images