Terahertz transmission liquid crystal metasurface rapid imaging method and system

Abstract

This invention discloses a rapid imaging method and system for terahertz transmission liquid crystal metasurfaces, relating to the field of terahertz metasurface imaging technology. The method incorporates an adaptive stopping criterion, enabling dynamic evaluation of acquired measurement data during (or after) acquisition, and determining whether further acquisition is necessary based on data consistency and / or reconstruction stability. When the stopping condition is met, subsequent coded measurements can be stopped early, thus avoiding redundant overhead caused by continuing acquisition after the imaging quality has plateaued, and shortening the total acquisition time. While ensuring the image reconstruction quality in the terahertz band meets requirements, it significantly improves the equivalent imaging speed and can meet the rapid imaging needs in dynamic scenes.

Description

Technical Field

[0001] This invention relates to the field of terahertz metasurface imaging technology, and more specifically to a method and system for rapid imaging of terahertz transmission liquid crystal metasurfaces. Background Technology

[0002] Terahertz waves (THz) generally refer to electromagnetic waves with frequencies in the range of 0.1-10 THz. They possess unique physical properties: on the one hand, terahertz waves can penetrate most non-polar materials such as ceramics and plastics to a certain extent, exhibiting excellent non-destructive detection capabilities; on the other hand, their photon energy is low, and they will not cause ionizing damage to biological tissues. Therefore, they have significant application prospects in fields such as security inspection, medical imaging, and industrial flaw detection.

[0003] Traditional terahertz imaging systems mostly employ mechanical scanning or phased array antenna technology. Mechanical scanning typically relies on point-by-point movement of the target or detector to complete imaging, resulting in low acquisition efficiency and slow imaging speed. While phased array technology can achieve fast beam scanning, it depends on a large number of transmit / receive (T / R) channels and complex RF front-ends, often facing significant link losses, high system integration difficulty, and high hardware costs at high frequencies such as 140 GHz. To overcome these limitations, computational imaging (such as single-pixel imaging) offers a feasible path: this technology uses a spatial modulator to encode and modulate the incident field, uses a single-pixel detector to acquire projection measurements, and then combines compressed sensing algorithms to reconstruct the target image. This approach significantly reduces the system's dependence on a high-performance RF front-end while maintaining imaging capabilities. Liquid crystal metasurfaces possess advantages such as continuous tunability, low power consumption, and easy integration. They can generate various spatial coding modes (spatial coding patterns) by electrically controlling the equivalent electromagnetic response, thus becoming an important candidate for terahertz single-pixel imaging.

[0004] Taking single-pixel imaging as an example, it does not rely on a two-dimensional detector array. Instead, it uses a single-pixel detector to repeatedly acquire data. Before each acquisition, a different spatial code is applied to the incident field through a spatial modulator. In each encoded state, the single-pixel detector measures the "overall weighted sum" (a scalar value) of the scene's transmission information under that spatial code, essentially projecting the two-dimensional scene information onto a numerical value. After repeating this process multiple times with different codes, a set of measurements is obtained. As long as the encoding method is properly designed and the measurements contain enough independent information, algorithms can then be used to deduce the two-dimensional image from these "projections." A commonly used mathematical model is:

[0005] in This represents the reflection / transmission / scattering intensity of the target to be imaged on a two-dimensional grid (expanded into a vector). These are measurement sequences collected according to different spatial codes. For the measurement matrix, Each row corresponds to a weighting coefficient of the spatial coding mode for each pixel. This represents noise and system error. The significance of this model is that the encoding state determines "how to weight," the measured value is the "weighted result," and multiple measurements can be used to inversely solve the image.

[0006] If we understand it according to the traditional sampling theorem, reconstruction Images typically require a number of independent measurements on par with the number of pixels (e.g., ...). (times). However, many target images possess structural characteristics: for example, the target region occupies a small area, the image is mainly composed of edges / contours, or the image can be represented by a small number of coefficients in certain transform domains. This "information sparsity / compressibility" makes it possible to measure the matrix... Under conditions of some randomness or low correlation, images can be recovered with fewer measurements than the number of pixels by using compressed sensing.

[0007] In engineering, common reconstruction methods introduce prior constraints to stabilize the solution, such as Total Variation (TV) regularization. The intuitive meaning of TV is to encourage smoothness in most areas of the image while allowing for significant variations (edges) in a few locations. Therefore, it is particularly effective for "blocky / sharp outlines" and can also suppress grain artifacts caused by noise. In practical applications, a higher sampling rate (number of measurements / number of pixels) and lower noise generally result in better reconstruction quality. When the sampling rate is low, choosing an appropriate regularization strength (such as TV weights) significantly affects image quality.

[0008] For arrayed metasurfaces, when a certain coded state is loaded, it can be understood that each unit in the array is set to "state A or state B" or takes a certain value within a continuously adjustable range, ultimately forming a two-dimensional coded pattern (spatial coded pattern). The system sequentially loads multiple sets of coded patterns and acquires measurement values ​​frame by frame to obtain the corresponding measurement matrix. Common coding methods include random 0 / 1 coding and Hadamard coding. Their common goal is to make measurement patterns as "independent" as possible, thereby improving reconstruction stability and reducing the number of measurements.

[0009] When programmable metasurfaces are used for imaging, they can be classified into reflective and transmissive types according to the direction of electromagnetic wave propagation: Reflective architecture: The metasurface acts like a "programmable mirror," where the incident wave is modulated and reflected back to the incident side. The system's feed, receiver, and imaging target are often arranged on the same side of the metasurface (or through folded optical paths to achieve co-directional transmission and reception). This structure is prone to problems in engineering implementation: the feed / horn antenna occupies space in front of the metasurface, reducing the effective aperture utilization; it may also introduce additional scattering, coupling, and obstruction, which are detrimental to beam quality and imaging consistency.

[0010] Transmissive architecture: The metasurface acts like a "programmable filter," modulating the incident wave before transmitting it to the other side. A typical arrangement is as follows: the feed is located on the incident side of the metasurface, the target / imaging region is on the exit side, and the receiver acquires the transmitted signal carrying the target information. The advantages of the transmissive structure are a more direct optical path, a more compact system spatial layout, and reduced same-side obstruction and unnecessary near-field coupling, making it easier to construct an integrated transmissive imaging system.

[0011] Liquid crystal materials exhibit dielectric anisotropy; their molecular orientations rotate and rearrange under the influence of an electric field, thereby altering the material's equivalent dielectric constant. By introducing liquid crystals into metasurface unit structures, the resonant conditions and electromagnetic responses of the units can be changed by applying different voltages, achieving controllable modulation of transmission amplitude or phase. At higher frequencies, switching schemes based on semiconductor devices such as PIN diodes and varactor diodes are affected by parasitic capacitance / inductance, conduction losses, and package interconnections, leading to increased additional losses, decreased efficiency, and complex large-scale array implementation. Liquid crystal solutions, on the other hand, can achieve large-scale driving through electrode arrays and possess continuously adjustable modulation capabilities, thus becoming an important route for realizing terahertz programmable metasurfaces, suitable for combining with computational imaging to build low-complexity imaging systems.

[0012] In the baseline imaging process, a liquid crystal transmissive programmable metasurface array serves as a spatial encoding device. The control module sequentially selects encoding patterns from a preset encoding codebook and loads them onto the array, causing each pixel in the array to exhibit high and low transmission under two bias states, thus forming a binary amplitude encoding pattern. The spatial modulator is the core hardware that "generates different spatial encoding modes." The richer the spatial encoding modes and the lower their correlation, the better for reconstruction. The programmable metasurface is composed of a large number of subwavelength units, each of which can exhibit different electromagnetic responses (e.g., different transmission amplitudes or phases) under external control, thereby forming a controllable modulation distribution in space. The baseline process pre-sets the sampling budget. (For example, by sampling rate) And load the previous ones in sequence. The measurement is completed using a coded pattern.

[0013] During specific measurements, after the k-th encoded pattern is loaded, the system waits for a fixed dwell time. Then trigger the single-pixel detector to sample and obtain the scalar measurement value corresponding to the coded pattern. Fixed length of stay A conservative setting method is adopted: the maximum time constant is estimated based on the response time constant of the liquid crystal during the rising and falling processes. And set according to a certain proportion (e.g., 99%) to reach steady state. Repeat the above process until you get... Each measurement value forms a measurement vector. The measurement matrix is ​​composed of the transmission weights corresponding to the spatially encoded patterns. Based on measurement model For reconstruction, the preferred method is the Total Variational Constraint Iterative Algorithm (TV-ADMM) to solve for the two-dimensional distribution x of the target and output the imaging results.

[0014] The characteristic of the baseline imaging process is that the number of samples is fixed. Furthermore, each frame measurement uses a fixed dwell time. When the liquid crystal has actually reached the effective modulation level required for imaging, continuing to wait until it approaches a steady state will incur redundant time overhead, thus increasing the total imaging time. and Linear growth.

[0015] In summary, while liquid crystal programmable metasurfaces can achieve considerable modulation capabilities in steady state, their response is not instantaneous: when the voltage switches from one state to another, liquid crystal molecules need a certain amount of time to complete reorientation; when the voltage is removed or switched back to a low-field state, the molecular orientation undergoes a fallback relaxation process. This dynamic process can usually be described by a time constant, and fallback relaxation is often relatively slow. In single-pixel / compressed imaging, a large number of coded states need to be continuously loaded and acquired frame by frame. If each frame adopts a fixed dwell time and waits for near-steady-state resampling, the "waiting time" of each frame will be repeatedly accumulated throughout the measurement sequence, resulting in a significant increase in the total acquisition time, becoming the main limiting factor for the system to achieve fast imaging. On the other hand, if the dwell time is blindly shortened in pursuit of speed, phenomena such as insufficient modulation establishment, insufficient coded contrast, and poor measurement consistency may occur, leading to a decrease in image reconstruction quality. Therefore, how to reduce the time overhead caused by the dynamic response and relaxation of liquid crystals while ensuring imaging effectiveness and quality threshold is the key background technical problem that this invention needs to solve. Summary of the Invention

[0016] To address the aforementioned shortcomings in the prior art, the present invention provides a terahertz transmission liquid crystal metasurface fast imaging method and system that solves the problem of reducing the time overhead caused by liquid crystal dynamic response and relaxation while ensuring the effectiveness and quality threshold of single-pixel imaging.

[0017] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for rapid imaging of terahertz transmission liquid crystal metasurfaces is provided, comprising the following steps: The effective contrast threshold is calculated based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive cell of the liquid crystal metasurface. Between the preset minimum waiting time and the maximum fixed dwell time, based on the switching situation of the current frame relative to the previous frame, the shortest waiting time that meets the effective contrast threshold is searched to obtain the single-frame waiting time corresponding to the current frame. The transmission response corresponding to each spatially encoded pattern is detected by a single-pixel detector. After each spatially encoded pattern is loaded, the dynamic response is updated according to the corresponding single-frame waiting time and a scalar value is collected to obtain the measurement sequence. The transmission response is generated by the transmission unit of the liquid crystal metasurface under the action of bias voltage. A stop criterion check is triggered once at a preset check interval. The nearest Q frame among the currently accumulated P-frame scalar values ​​is used as the verification window, and the remaining PQ frames are used as the training part. Fast single-pixel reconstruction is performed using the training part to obtain temporary imaging results. Perform data consistency verification and / or reconstruction stability verification. If the verification passes, trigger early stop, stop loading of subsequent spatial coding patterns, and perform single-pixel imaging based on the measurement sequence acquired up to the current time and output the results to complete the rapid imaging of the terahertz transmission liquid crystal metasurface. Otherwise, perform single-pixel imaging based on the measurement sequence corresponding to the maximum allowable sampling frame number and output the results to complete the rapid imaging of the terahertz transmission liquid crystal metasurface. The data consistency verification includes: predicting the verification window based on the temporary imaging results, and calculating the difference between the predicted value and the true value of the verification window; Reconstruction stability verification includes calculating the difference between the current provisional imaging result and the previous provisional imaging result.

[0018] Furthermore, the expression for calculating the effective contrast threshold is:

[0019]

[0020] in This is the effective contrast threshold; A constant greater than 0; For contrast; and These represent the equivalent transmission amplitudes of steady-state high transmission and steady-state low transmission for the transmissive unit of the liquid crystal metasurface, respectively.

[0021] Furthermore, the specific methods for searching the shortest waiting time that satisfies the effective contrast threshold between the preset minimum waiting time and the maximum fixed dwell time, based on the switching behavior of the current frame relative to the previous frame, include: Based on the switching situation of the current frame relative to the previous frame, the high-transmission units and low-transmission units of the liquid crystal metasurface are determined, and dynamic response models of the average transmission state of the high-transmission units and low-transmission units are established:

[0022]

[0023] in The dynamic response value of the average transmission state of the high-transmission metasurface transmissivity unit group of liquid crystal; The dynamic response value is the average transmission state of the low-transmission metasurface transmission unit group of the liquid crystal. This represents the target transmission state of the current frame rate liquid crystal metasurface transmission unit group; The current frame represents the average transmission state of the high-resolution liquid crystal metasurface transmission unit group at the previous moment; e is the natural constant; t represents time. The rise time constant of the liquid crystal; This represents the target transmission state of the current frame's low-resolution liquid crystal metasurface transmission unit group. This represents the average transmission state of the low-transmission metasurface cell group of the current frame at the previous moment. The liquid crystal descent response time constant; by To meet the criteria for effective contrast threshold, the shortest time that satisfies the criterion is searched between the preset minimum waiting time and the maximum fixed dwell time, and this time is determined as the single-frame waiting time corresponding to the current frame.

[0024] Furthermore, the difference between the predicted and true values ​​in the validation window is represented by the normalized residual, which is expressed as:

[0025] in To verify the normalized residuals between the predicted and actual values ​​of the window; To verify the actual value of the window; To verify the predicted values ​​of the window, To verify the measurement matrix corresponding to the window; This is a temporary imaging result; when If the data is less than the first set threshold, the data consistency verification is considered successful; otherwise, the data consistency verification is considered unsuccessful.

[0026] Furthermore, when performing reconstruction stability verification, if the difference between the current temporary imaging result and the previous temporary imaging result is less than the second set threshold, the reconstruction stability verification is deemed to have passed; otherwise, the reconstruction stability verification is deemed to have failed.

[0027] Furthermore, it also includes overdrive operations: Apply an overdrive voltage higher than the operating voltage during the initial stage of encoding pattern switching. And continue Time allows the liquid crystal to approach the target state at a faster equivalent speed in the initial stage, thereby reducing the single-frame waiting time.

[0028] A system is provided for realizing a fast imaging method for terahertz transmission liquid crystal metasurfaces, comprising: The transmitting module is used to generate and transmit terahertz electromagnetic waves; the terahertz electromagnetic waves are collimated and incident on the optical path between the imaging object and the liquid crystal metasurface. The liquid crystal metasurface modulation module is used to spatially modulate the transmitted terahertz wave under the action of bias voltage, generate a preset spatial coding pattern, so that the imaging object produces different transmission responses under different spatial coding patterns of the transmitted terahertz wave, thereby forming the measurement sequence required for single pixel measurement. The bias voltage control module is used to provide bias voltage to the liquid crystal metasurface modulation module and control its timing switching in order to realize the loading of spatial coding patterns. The receiving and acquisition module is used to perform single-pixel detection on the transmitted terahertz signal and acquire measurement sequences corresponding to different spatial coding patterns. The data processing module is used to calculate the effective contrast threshold based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive unit of the liquid crystal metasurface; between the preset minimum waiting time and the maximum fixed dwell time, it searches for the shortest waiting time that satisfies the effective contrast threshold based on the switching situation of the current frame relative to the previous frame, and obtains the single-frame waiting time corresponding to the current frame; it receives the measurement sequence and the corresponding spatial coding pattern information, triggers a stop criterion check at a preset check interval, constructs a measurement model and performs image reconstruction, inverts to obtain the two-dimensional distribution information of the imaging object in the imaging area, and realizes the single-pixel imaging result output.

[0029] Furthermore, the liquid crystal metasurface modulation module is a periodic unit array structure, including several liquid crystal metasurface transmission units; each liquid crystal metasurface transmission unit includes, along the incident direction, an upper glass substrate, an upper metal layer resonant structure, a liquid crystal layer, a lower metal layer resonant structure, and a lower glass substrate; wherein the upper metal layer resonant structure and the lower metal layer resonant structure are arranged opposite to each other, and together with the liquid crystal layer, they constitute a transmission resonant unit; by adjusting the voltage to change the equivalent dielectric constant of the liquid crystal, the equivalent electromagnetic response of the transmission resonant unit is changed, thereby modulating the transmission coefficient.

[0030] Furthermore, when applying a bias voltage to the liquid crystal metasurface modulation module, a bias voltage is applied to each liquid crystal metasurface transmission unit according to a preset spatial coding pattern, so that the array presents a corresponding spatial distribution of transmission coefficients; wherein different spatial coding patterns correspond to different transmission distributions, which are used for subsequent single-pixel measurement and image reconstruction.

[0031] Furthermore, the array electrodes are divided into row electrodes and column electrodes. The row electrodes and column electrodes cross each other to form the bias loading point of each liquid crystal metasurface transmission unit. The control module establishes a potential difference between the selected row and column electrodes by row / column gating and level superposition, thereby realizing the bias voltage loading and spatial encoding state switching of the corresponding liquid crystal metasurface transmission unit.

[0032] The beneficial effects of this invention are as follows: 1. Reduced redundant measurements and shortened total acquisition time: This invention is equipped with an adaptive stop criterion, which can dynamically evaluate the acquired measurement data during (or after) the acquisition process and determine whether to continue acquisition based on data consistency and / or reconstruction stability. When the stop condition is met, subsequent coded measurements can be stopped in advance, thereby avoiding redundant overhead caused by continuing acquisition after the imaging quality has reached a plateau, and shortening the total acquisition time.

[0033] 2. Significantly reduce imaging time under the same sampling rate: This invention introduces a threshold dwell time (overdrive) mechanism, which adaptively determines the dwell time frame by frame under the premise of meeting the preset modulation effectiveness threshold (such as the ER threshold), so that the waiting time for each frame is no longer fixed at the maximum value, thereby reducing the cumulative dwell time and thus reducing the total imaging time.

[0034] 3. Array control is more easily expandable, and system integration complexity is lower: This invention adopts a row-column multiplexing cross-addressing structure, reducing the number of external control lines from... Reducing the number of paths to 2N paths can lower the complexity of wiring and driving, providing favorable conditions for array scaling and engineering implementation.

[0035] 4. Constructing a low-cost, low-loss 140GHz hardware platform: This invention proposes a liquid crystal metasurface modulation module, which utilizes the low-loss and continuously tunable characteristics of liquid crystal materials in the terahertz band to achieve spatial modulation, reducing the dependence on high-frequency semiconductor switches and large-scale T / R components. This ensures modulation capability while significantly reducing system cost, loss and implementation complexity, and has engineering feasibility for scaling up to large-scale arrays.

[0036] 5. The present invention adopts a transmission structure, which places the feed source and the imaging target on opposite sides of the metasurface, thus avoiding the obstruction and interference of the imaging area by the feed source from a physical structure perspective, and greatly improving the modulation efficiency and imaging accuracy of electromagnetic waves. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the method. Figure 2 This is a schematic diagram of the system structure; Figure 3 This is a schematic diagram of the structure of a liquid crystal metasurface modulation module; Figure 4 This is a schematic diagram of a periodic cell array structure; Figure 5 This is a transmission amplitude diagram of the transmissive unit of the liquid crystal metasurface; Figure 6 This is a schematic diagram of the original image and the reconstructed image; Figure 7 This is a schematic diagram of the threshold dwell time and overdrive mechanism; Figure 8 A comparison chart of total imaging time; Figure 9 This is a schematic diagram of the cumulative imaging time-PSNR curve (target PSNR=16 dB); Figure 10 A time comparison chart to achieve the target PSNR; Figure 11 This is the PSNR-sampling rate relationship curve. Detailed Implementation

[0038] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0039] like Figure 1 As shown, the terahertz transmission liquid crystal metasurface fast imaging method includes the following steps: S1. Calculate the effective contrast threshold based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive unit of the liquid crystal metasurface. S2. Between the preset minimum waiting time and the maximum fixed dwell time, based on the switching situation of the current frame relative to the previous frame, search for the shortest waiting time that meets the effective contrast threshold, and obtain the single-frame waiting time corresponding to the current frame. S3. The transmission response corresponding to each spatial coding pattern is detected by a single pixel detector. After each spatial coding pattern is loaded, the dynamic response is updated according to the corresponding single frame waiting time and a scalar value is collected to obtain the measurement sequence. The transmission response is generated by the transmission unit of the liquid crystal metasurface under the action of bias voltage. S4. Trigger a stop criterion check at a preset check interval, take the most recent Q frame among the currently accumulated P-frame scalar values ​​as the verification window, and the remaining PQ frames as the training part; use the training part to perform fast single-pixel reconstruction to obtain temporary imaging results. S5. Perform data consistency verification and / or reconstruction stability verification. If the verification is successful, trigger early stop, stop loading of subsequent spatial coding patterns, and perform single-pixel imaging based on the measurement sequence collected up to the current time and output the results to complete the terahertz transmission liquid crystal metasurface fast imaging; otherwise, perform single-pixel imaging based on the measurement sequence corresponding to the maximum allowable sampling frame number and output the results to complete the terahertz transmission liquid crystal metasurface fast imaging. The data consistency verification includes: predicting the verification window based on the temporary imaging results, and calculating the difference between the predicted value and the true value of the verification window; Reconstruction stability verification includes calculating the difference between the current provisional imaging result and the previous provisional imaging result.

[0040] It's important to clarify that when only data consistency verification or reconstruction stability verification is used, passing either verification method means that early stop needs to be triggered. When both data consistency verification and reconstruction stability verification are used simultaneously, both verification methods must pass before early stop is triggered.

[0041] like Figure 2 As shown, the system for implementing this terahertz transmission liquid crystal metasurface fast imaging method includes: The transmitting module (feed source) is used to generate and transmit terahertz electromagnetic waves; the terahertz electromagnetic waves are incident on the optical path of the imaging object and the liquid crystal metasurface after passing through quasi-optical components (such as lenses); The liquid crystal metasurface modulation module is used to spatially modulate the transmitted terahertz wave under the action of bias voltage, generate a preset spatial coding pattern, so that the imaging object produces different transmission responses under different spatial coding patterns of the transmitted terahertz wave, thereby forming the measurement sequence required for single pixel measurement. The bias voltage control module is used to provide bias voltage to the liquid crystal metasurface modulation module and control its timing switching to realize the loading of spatial coding patterns. This module can be implemented by the FPGA control unit, outputs control signals according to the preset codebook, and is synchronously triggered with the receiving and acquisition module to ensure that each coding mode matches the corresponding measurement value one by one. The receiving and acquisition module is used to perform single-pixel detection on the transmitted terahertz signal and acquire measurement sequences corresponding to different spatial coding patterns. The data processing module is used to calculate the effective contrast threshold based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive unit of the liquid crystal metasurface; between the preset minimum waiting time and the maximum fixed dwell time, it searches for the shortest waiting time that satisfies the effective contrast threshold based on the switching situation of the current frame relative to the previous frame, and obtains the single-frame waiting time corresponding to the current frame; it receives the measurement sequence and the corresponding spatial coding pattern information, triggers a stop criterion check at a preset check interval, constructs a measurement model and performs image reconstruction, inverts to obtain the two-dimensional distribution information of the imaging object in the imaging area, and realizes the single-pixel imaging result output.

[0042] The data processing module can also send control parameters back to the bias voltage control module to optimize the sampling number and single-frame dwell time.

[0043] In some embodiments, the receiving and acquisition module includes a single-pixel detector and an acquisition circuit / data acquisition device connected thereto, used to convert the detector output into digital measurement values ​​to obtain the measurement sequence corresponding to each encoding mode.

[0044] like Figure 3 and Figure 4 As shown, the liquid crystal metasurface modulation module is a liquid crystal programmable transmissive metasurface (e.g., a 32×32 array). This programmable transmissive metasurface has a periodic unit array structure, enabling controllable modulation of incident terahertz waves in the target frequency band (e.g., 140 GHz), used to implement / load preset spatial coding patterns (masks). The entire array constitutes a transmissive spatial modulator. The bias voltage control module applies voltage to each unit of the array according to the preset coding pattern, causing the array to exhibit a corresponding spatial distribution of transmission coefficients; different coding patterns correspond to different transmission distributions, used for subsequent single-pixel measurements and image reconstruction.

[0045] Specifically, the liquid crystal metasurface modulation module includes several liquid crystal metasurface transmission units; each liquid crystal metasurface transmission unit includes, along the incident direction, an upper glass substrate, an upper metal layer resonant structure, a liquid crystal layer, a lower metal layer resonant structure, and a lower glass substrate; wherein the upper metal layer resonant structure and the lower metal layer resonant structure are arranged opposite to each other, and together with the liquid crystal layer, they form a transmission resonant unit; the equivalent electromagnetic response of the transmission resonant unit is changed by voltage-controlled liquid crystal equivalent dielectric constant, thereby modulating the transmission coefficient. The liquid crystal layer is located between the upper and lower metal resonant structures, and its thickness is defined by spacers (e.g., microspheres) disposed between the two substrates; the liquid crystal cavity is encapsulated and sealed using encapsulant to define the liquid crystal filling area and prevent liquid crystal leakage.

[0046] The liquid crystal metasurface transmissive unit alters the orientation of liquid crystal molecules by applying an external bias voltage, thereby changing the equivalent dielectric constant of the liquid crystal layer. This, in turn, changes the unit's equivalent electromagnetic response in the target frequency band, resulting in a change in the unit's transmission coefficient. The liquid crystal metasurface transmissive unit operates in an "amplitude-encoded" mode, meaning that under different equivalent dielectric constants, the unit exhibits significantly different transmission amplitudes near the target frequency, achieving a "high transmission state / low transmission state," providing usable encoded contrast for subsequent single-pixel imaging.

[0047] To reduce the wiring and driving complexity of large-scale arrays, this embodiment preferably adopts a row-column multiplexed cross-addressing bias structure to achieve pixel addressing: the array electrodes are divided into row electrodes and column electrodes, and the row electrodes and column electrodes cross to form the bias loading point of each pixel unit. The bias voltage control module establishes a potential difference between the selected row and column electrodes through row / column gating and level superposition, thereby realizing the bias loading and encoding state switching of the corresponding pixel unit. After adopting the above cross-addressing structure, the number of array external control lines is reduced from N×N paths with independent pins per pixel to 2N paths (for example, a 32×32 array only requires 64 row and column lines), thereby significantly reducing the number of wiring, driving channels, and system integration difficulty, and improving the array scalability and engineering feasibility.

[0048] like Figure 5 As shown, in two states of equivalent dielectric constant of liquid crystal ( and Under these conditions, the transmissive units of the liquid crystal metasurface exhibit significantly different transmission amplitude responses in the 130–150 GHz range, forming a clear difference in transmission amplitude near the target frequency. This enables binary amplitude encoding: when the liquid crystal is in the first dielectric constant state, it corresponds to a higher transmission amplitude (i.e., the equivalent transmission amplitude of steady-state high transmission, which can be defined as the "ON state"), and when the liquid crystal is in the second dielectric constant state, it corresponds to a lower transmission amplitude (i.e., the equivalent transmission amplitude of steady-state low transmission, which can be defined as the "OFF state"). This "ON / OFF" transmission contrast provides the physical basis for subsequent spatial encoding, enabling the liquid crystal metasurface array to switch its spatial transmission state according to a preset encoding sequence. The single-pixel detector collects the measurement values ​​corresponding to each encoded state, and finally, the two-dimensional distribution of the imaging target is recovered by a computational reconstruction algorithm.

[0049] This embodiment employs a single-pixel computational imaging method to achieve terahertz imaging: During the imaging process, the liquid crystal metasurface modulation module, under the action of the bias voltage control module, sequentially loads a set of preset spatial coding patterns, causing the liquid crystal metasurface modulation module to exhibit different spatial distributions of transmission coefficients under different patterns; the single-pixel detector detects the transmission response corresponding to each coding pattern and outputs a scalar measurement value, acquiring a measurement sequence frame by frame, thereby mapping the two-dimensional target information into a one-dimensional measurement vector, providing input for subsequent calculation and reconstruction. The above imaging process can be described using existing measurement models.

[0050] The spatial coding pattern is preferably a Hadamard codebook or its scrambled codebook. Specifically, the bias voltage control module selects (or scrambles) M coding vectors from the Hadamard codebook and maps them to the binary bias state of the array, so that the array cells are in a high transmission state (ON) or a low transmission state (OFF), thereby forming a binary amplitude spatial coding pattern; correspondingly, the measurement matrix... Each row consists of the ON / OFF transmission weights of the spatial coding pattern at each pixel position, used to establish the above measurement equation and perform image reconstruction.

[0051] The fast imaging method of this invention reduces the total imaging time in two ways without changing the coding codebook family (still using the same out-of-order Hadamard pool): first, by reducing the single-frame dwell time through threshold dwell time and overdrive strategies; and second, by reducing the sampling rate required to achieve the target imaging quality, enabling the system to obtain usable quality reconstruction results at a lower sampling rate. Correspondingly, the optimized acquisition process no longer uses a fixed dwell time for each frame measurement. Instead, it adaptively determines the dwell time (single frame waiting time) based on threshold criteria, and introduces an adaptive early stop strategy during the acquisition process to terminate subsequent measurements in advance.

[0052] For example, in reducing the single-frame dwell time by using a threshold dwell time, this invention replaces "steady-state waiting" with a "coding validity threshold" as the sampling trigger criterion. Specifically, after the spatial coding pattern of the array pixels switches, their transmission response gradually approaches the target state from the previous state. For imaging, as long as the difference between the high-transmission state and the low-transmission state reaches a preset threshold, the frame coding can be considered sufficient for measurement, and there is no need to wait until it approaches a steady state. To this end, this invention defines a threshold criterion and calculates the minimum dwell time of the frame accordingly. The preferred criterion is the contrast threshold (ER criterion), based on the equivalent transmission amplitude of the final steady-state high transmittance. Equivalent transmission amplitude with low transmission Calculate the final contrast and take (For example This serves as the effective contrast threshold. Given the known rise and fall response time constants of liquid crystals... , In the case of switching in the current frame, calculate the shortest time required to satisfy the effective contrast threshold, and limit it to [ , [Obtained within the interval] (in (Assuming a minimum wait limit, e.g., 30 ms). Then simply wait... This involves sampling to obtain the measurement value of the current frame and then moving on to the next frame, thereby significantly reducing the average single-frame waiting time.

[0053] For example, in this embodiment, high-transmission units and low-transmission units of liquid crystal metasurface are determined based on the switching situation of the current frame relative to the previous frame, and a dynamic response model of the average transmission state of the high-transmission unit group and the low-transmission unit group of liquid crystal metasurface is established:

[0054]

[0055] in The dynamic response value of the average transmission state of the high-transmission metasurface transmissivity unit group of liquid crystal; The dynamic response value is the average transmission state of the low-transmission metasurface transmission unit group of the liquid crystal. This represents the target transmission state of the current frame rate liquid crystal metasurface transmission unit group; The current frame represents the average transmission state of the high-resolution liquid crystal metasurface transmission unit group at the previous moment; e is the natural constant; t represents time. The rise time constant of the liquid crystal; This represents the target transmission state of the current frame's low-resolution liquid crystal metasurface transmission unit group. This represents the average transmission state of the low-transmission metasurface cell group of the current frame at the previous moment. The liquid crystal descent response time constant; by To meet the criteria for effective contrast threshold, the shortest time that satisfies the criterion is searched between the preset minimum waiting time and the maximum fixed dwell time, and this time is determined as the single-frame waiting time corresponding to the current frame.

[0056] Furthermore, to shorten the time required to reach the threshold, this invention can introduce overdrive technology: applying an overdrive voltage higher than the operating voltage during the initial stage of spatial coding pattern switching. And continue The time required allows the liquid crystal to approach the target state at a faster equivalent speed in the initial stage, and then the operating voltage is restored to complete the frame measurement. The overdrive technique only changes the process time to reach the threshold, without changing the codebook or the final operating state definition. Therefore, it can be used in conjunction with the threshold dwell time strategy to further reduce the single frame dwell time.

[0057] To reduce the sampling rate while maintaining image quality, this invention introduces an adaptive early stopping strategy. Optimized acquisition still relies on... This serves as the maximum budget (maximum allowed number of sampling frames), but allows for early termination if the stopping condition is met, thus reducing the actual number of samplings. The specific implementation is as follows: after collecting a fixed number of frames, an online check is triggered (e.g., every 8 frames). Most of the currently collected data is used as the training segment for rapid reconstruction to obtain a temporary reconstruction result. The most recent measurement period is used as the validation window to calculate the consistency index. Measurement domain consistency is characterized by normalized residuals, i.e., the consistency index is calculated based on the validation window measurements. Compared with the predicted value The difference and normalization are obtained ,when When the difference is less than a threshold (e.g., 0.06), the newly added measurement is considered to be interpretable by the current reconstruction. To avoid misjudgment, this embodiment further adopts an image domain stability criterion (stability verification): comparing the difference between the current temporary reconstruction and the previous temporary reconstruction, when the difference is less than a threshold (e.g., 0.03), the reconstruction is considered to have stabilized. This difference is preferably calculated within the ROI region, that is, after sampling reaches a certain proportion, a significant region is extracted from the temporary reconstruction image and appropriately expanded to obtain the ROI mask, and the change is calculated within the ROI to improve the criterion's specificity; when the ROI has not yet been established, the global change can be used as a substitute. The early stopping decision can adopt the method of "simultaneous satisfaction of residual criterion and stability criterion", and can require several consecutive satisfactions (e.g., 2 times) before triggering stopping to improve robustness. Once early stopping is triggered, the system stops loading subsequent spatial coding patterns and records the reason for stopping, and then only performs terminal reconstruction on the acquired M-frame data and outputs the results.

[0058] Through the above mechanism, two types of acceleration are achieved by optimizing the acquisition under the same codebook conditions: on the one hand, the average single-frame waiting time is reduced by threshold dwell time (optional overdrive), and on the other hand, the actual number of samplings is reduced by adaptive early stopping. Thus, the total imaging time is reduced while ensuring imaging quality, and the fast imaging effect can be further demonstrated by comparing indicators such as total imaging time and speedup ratio.

[0059] In one embodiment of the present invention, a liquid crystal programmable transmissive metasurface array with N=32 is used as the object. A spatial coding pattern is generated using a scrambled Hadamard codebook, and a single-pixel detector obtains scalar measurement values ​​frame by frame to form a measurement vector. The array pixels employ binary amplitude coding, with the equivalent transmission amplitudes for high and low transmission values ​​respectively taken as... , The noise level was measured. The reconstruction algorithm uses TV-ADMM, with 250 iterations for terminal reconstruction. The main parameters are as follows: , And enable DCT sparse prior ( , In this embodiment, the target sampling rate is set to... Corresponding sampling number (268 times when N=32), and set the full sampling reference group sampling rate to 268 times. (correspond (This is used to give the upper bound of performance under the same device and the same reconstruction algorithm conditions).

[0060] The comparison group is defined as follows: Full is the full-sampled reference group ( ), using fixed residence time And complete full sampling; Baseline reference group ( Using the same fixed residence time And complete the corresponding sampling; Optimized is the optimization group (both are... This optimization group introduces threshold dwell time and overdrive techniques to shorten single-frame dwell time without changing the codebook family. It also configures an adaptive early stopping criterion to terminate acquisition early when certain conditions are met. The threshold criterion in the optimization group adopts the ER criterion. And set a minimum stay time limit. ; through driving parameters , The early stopping parameters were set to a residual threshold of 0.06, a stability threshold of 0.03, a check interval of 8, a verification window of 32, and a consecutive satisfaction count of 2. An Area of ​​Interest (ROI) was constructed for stability determination after a certain sampling ratio was reached. In this simulation, the `stop_reason` output by the optimization group was the upper limit of the sampling rate corresponding to the preset sampling rate, indicating that under these parameters and noise conditions, early stopping did not prematurely truncate the sampling count. Therefore, the main source of the "total time acceleration" in this simulation is the compression of the single-frame dwell time due to the threshold dwell time and overdrive. Meanwhile, the time advantage of "reaching the target quality threshold earlier" still reflects the rapid imaging effect.

[0061] Reconstruction results as follows Figure 6 As shown. Full sampling reference group (Full, As the upper limit of quality, the outline of the target letter "N" is clear and the details are complete; the baseline group (Baseline, At a lower sampling rate, the main structure of "N" can be recovered, but background noise and block artifacts are relatively more obvious; the optimized group (Optimized, It can also recover the main outline and structural information of "N", and the overall recognizability is on the same order of magnitude as the reference group, indicating that the acceleration strategy proposed in this invention does not lead to imaging failure.

[0062] The first acceleration mechanism (reducing single-frame dwell time) is demonstrated by both the mechanism verification diagram and the total imaging time results. The mechanism is illustrated below. Figure 7 As shown, under the same threshold conditions, the time required for the transmission response to reach the threshold line is shorter after overdrive is added, corresponding to a reduction in the threshold dwell time; this means that a fixed conservative waiting time is not required for each frame measurement. Instead, it can adaptively select shorter frames. This reduces single-frame waiting overhead and decreases cumulative dwell time.

[0063] In this embodiment, the optimization group and the baseline group use the same sampling rate (both are...). Furthermore, the actual number of samples was consistent (both reached the preset sampling limit, and stop_reason was the Reached budget), therefore the difference in total imaging time can be mainly attributed to the compression of single-frame dwell time. Figure 8 As shown, the total imaging times for Baseline and Optimized are 4349.59 s and 2622.79 s, respectively, indicating that the threshold dwell time and overdrive strategy can effectively reduce the total acquisition time without changing the sampling rate.

[0064] It should be noted that in this embodiment, the "total imaging time" is calculated based on the acquisition time, i.e., it is obtained by summing the dwell times of each frame (excluding the time spent on online decision-making and reconstruction calculations). Under this statistical caliber, although the optimized group is configured with adaptive early stopping and ROI decision-making modules, the early stopping criterion did not trigger early termination in this simulation, and the actual number of samplings was not reduced. Therefore, the difference in total imaging time between the baseline group and the optimized group under the same sampling rate conditions is mainly reflected in the change in the dwell time of a single frame from a fixed value. Towards Adaptive The change resulted in a decrease in the cumulative length of stay.

[0065] Acceleration Mechanism Two (reaching the target imaging quality threshold faster, and the adaptive early stopping framework) is reflected in the result of "time required to reach the target PSNR (peak signal-to-noise ratio)". In this embodiment, the target PSNR is set to 16 dB. The cumulative imaging time-PSNR curve is shown below. Figure 9As shown, the optimized group's curve crossed the target dashed line of 16 dB earlier; the corresponding time comparison histogram is shown below. Figure 10 As shown, the cumulative time for the optimized group to reach the target PSNR is significantly lower than that of the baseline group. This indicates that even under the same sampling rate, the optimized group can still enter the "usable imaging quality" range more quickly, demonstrating "fast imaging" from an engineering perspective. Since the total acquisition time increases approximately linearly with the number of samples, when the sampling rate required to reach the same quality threshold is reduced from a higher value to 0.30, the number of samples and the total imaging time can be reduced significantly proportionally, thus demonstrating the advantage of this invention in "fast imaging with low sampling volume".

[0066] It should be noted that the present invention can also optionally configure an adaptive stopping strategy to avoid redundant measurements after the quality plateau period during the acquisition process; its function can further reduce the actual number of samplings, but the core advantage of this mechanism is to "achieve the target quality threshold with a lower sampling rate", thereby reducing the required sampling amount and total acquisition time from the source.

[0067] Furthermore, to illustrate the rationality of the sampling rate selection, this embodiment presents the average PSNR variation trend of TV-ADMM under different sampling rates, such as... Figure 11 As shown, PSNR increases with the increase of sampling rate, and the lowest sampling rate that meets the target quality threshold can be selected accordingly to balance imaging quality and acquisition time.

[0068] Comprehensive quantitative results show that the optimized group reduces the total imaging time by approximately 5.53 times compared to the fully sampled reference group, and by approximately 1.66 times compared to the baseline group. While maintaining the identifiability of the target structure, this embodiment verifies that the present invention, under fixed codebook conditions, can significantly reduce the total imaging time through a rapid imaging technique that combines "threshold dwell time / reduced single-frame time" with "online criterion evaluation (with early stopping capability) to enable imaging to reach the target quality threshold earlier."

[0069] In summary, this invention, while maintaining the required image reconstruction quality in the terahertz band, reduces the effective waiting time for each encoding measurement by controlling the acquisition of the liquid crystal dynamic response process, thereby reducing the total imaging time. Under the condition of meeting the preset contrast / modulation threshold or image quality index (such as PSNR threshold), it completes the required measurement and reconstruction earlier, significantly improving the equivalent imaging speed and meeting the rapid imaging needs in dynamic scenes.

Claims

1. A method for rapid imaging of a terahertz transmission liquid crystal metasurface, characterized in that, Includes the following steps: The effective contrast threshold is calculated based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive cell of the liquid crystal metasurface. Between the preset minimum waiting time and the maximum fixed dwell time, based on the switching situation of the current frame relative to the previous frame, the shortest waiting time that meets the effective contrast threshold is searched to obtain the single-frame waiting time corresponding to the current frame. The transmission response corresponding to each spatial coding pattern is detected by a single pixel detector. After each spatial coding pattern is loaded, the dynamic response is updated according to the corresponding single frame waiting time and a scalar value is collected to obtain the measurement sequence. The transmission response is generated by the transmission units of the liquid crystal metasurface under the action of a bias voltage; A stop criterion check is triggered once at a preset check interval. The most recent Q frame among the currently accumulated P frame scalar values ​​is used as the verification window, and the remaining PQ frames are used as the training part. Fast single-pixel reconstruction is performed using the trained part to obtain temporary imaging results; Perform data consistency verification and / or reconstruction stability verification. If the verification passes, trigger early stop, stop loading of subsequent spatial coding patterns, and perform single-pixel imaging based on the measurement sequence acquired up to the current time and output the results to complete the rapid imaging of the terahertz transmission liquid crystal metasurface. Otherwise, perform single-pixel imaging based on the measurement sequence corresponding to the maximum allowable sampling frame number and output the results to complete the rapid imaging of the terahertz transmission liquid crystal metasurface. The data consistency verification includes: predicting the verification window based on the temporary imaging results, and calculating the difference between the predicted value and the true value of the verification window; Reconstruction stability verification includes: calculating the difference between the current provisional imaging result and the previous provisional imaging result; The formula for calculating the effective contrast threshold is: in This is the effective contrast threshold; A constant greater than 0; For contrast; and These represent the equivalent transmission amplitudes of steady-state high transmission and steady-state low transmission for the transmissive unit of the liquid crystal metasurface, respectively.

2. The terahertz transmission liquid crystal metasurface fast imaging method according to claim 1, characterized in that, Between the preset minimum waiting time and the maximum fixed dwell time, specific methods for searching the shortest waiting time that satisfies the effective contrast threshold based on the switching behavior of the current frame relative to the previous frame include: Based on the switching situation of the current frame relative to the previous frame, the high-transmission units and low-transmission units of the liquid crystal metasurface are determined, and dynamic response models of the average transmission state of the high-transmission units and low-transmission units are established: in The dynamic response value of the average transmission state of the high-transmission metasurface transmissivity unit group of liquid crystal; The dynamic response value is the average transmission state of the low-transmission metasurface transmission unit group of the liquid crystal. This represents the target transmission state of the current frame rate liquid crystal metasurface transmission unit group; The current frame represents the average transmission state of the high-resolution liquid crystal metasurface transmission unit group at the previous moment; e is the natural constant; t represents time. The rise time constant of the liquid crystal; This represents the target transmission state of the current frame's low-resolution liquid crystal metasurface transmission unit group. This represents the average transmission state of the low-transmission metasurface cell group of the current frame at the previous moment. The liquid crystal descent response time constant; by To meet the criteria for effective contrast threshold, the shortest time that satisfies the criterion is searched between the preset minimum waiting time and the maximum fixed dwell time, and this time is determined as the single-frame waiting time corresponding to the current frame.

3. The terahertz transmission liquid crystal metasurface fast imaging method according to claim 1, characterized in that, The difference between the predicted and true values ​​in the validation window is represented by the normalized residual, which is expressed as: in To verify the normalized residuals between the predicted and actual values ​​of the window; To verify the actual value of the window; To verify the predicted values ​​of the window, To verify the measurement matrix corresponding to the window; This is a temporary imaging result; when If the data is less than the first set threshold, the data consistency verification is considered successful; otherwise, the data consistency verification is considered unsuccessful.

4. The terahertz transmission liquid crystal metasurface fast imaging method according to claim 1, characterized in that, When performing reconstruction stability verification, if the difference between the current temporary imaging result and the previous temporary imaging result is less than the second set threshold, the reconstruction stability verification is deemed to have passed; otherwise, the reconstruction stability verification is deemed to have failed.

5. The terahertz transmission liquid crystal metasurface fast imaging method according to claim 1, characterized in that, It also includes overdrive operations: Apply an overdrive voltage higher than the operating voltage during the initial stage of encoding pattern switching. And continue Time allows the liquid crystal to approach the target state at a faster equivalent speed in the initial stage, thereby reducing the single-frame waiting time.

6. A system for implementing the terahertz transmission liquid crystal metasurface rapid imaging method according to any one of claims 1 to 5, characterized in that, include: The transmitting module is used to generate and transmit terahertz electromagnetic waves; The terahertz electromagnetic wave is collimated and then incident on the optical path between the imaging object and the liquid crystal metasurface. The liquid crystal metasurface modulation module is used to spatially modulate the transmitted terahertz wave under the action of bias voltage, generate a preset spatial coding pattern, so that the imaging object produces different transmission responses under different spatial coding patterns of the transmitted terahertz wave, thereby forming the measurement sequence required for single pixel measurement. The bias voltage control module is used to provide bias voltage to the liquid crystal metasurface modulation module and control its timing switching in order to realize the loading of spatial coding patterns. The receiving and acquisition module is used to perform single-pixel detection on the transmitted terahertz signal and acquire measurement sequences corresponding to different spatial coding patterns. The data processing module is used to calculate the effective contrast threshold based on the equivalent transmission amplitude of steady-state high transmission and steady-state low transmission of the transmissive unit of the liquid crystal metasurface; between the preset minimum waiting time and the maximum fixed dwell time, it searches for the shortest waiting time that satisfies the effective contrast threshold based on the switching situation of the current frame relative to the previous frame, and obtains the single-frame waiting time corresponding to the current frame; it receives the measurement sequence and the corresponding spatial coding pattern information, triggers a stop criterion check at a preset check interval, constructs a measurement model and performs image reconstruction, inverts to obtain the two-dimensional distribution information of the imaging object in the imaging area, and realizes the single-pixel imaging result output.

7. The terahertz transmission liquid crystal metasurface fast imaging system according to claim 6, characterized in that, The liquid crystal metasurface modulation module is a periodic unit array structure, including several liquid crystal metasurface transmission units; each liquid crystal metasurface transmission unit includes, along the incident direction, an upper glass substrate, an upper metal layer resonant structure, a liquid crystal layer, a lower metal layer resonant structure, and a lower glass substrate. The upper metal layer resonant structure and the lower metal layer resonant structure are arranged opposite to each other, and together with the liquid crystal layer, they form a transmissive resonant unit. The equivalent electromagnetic response of the transmissive resonant unit is changed by voltage regulation of the liquid crystal equivalent dielectric constant, thereby modulating the transmission coefficient.

8. The terahertz transmission liquid crystal metasurface fast imaging system according to claim 7, characterized in that, When a bias voltage is applied to the liquid crystal metasurface modulation module, a bias voltage is applied to each liquid crystal metasurface transmission unit according to a preset spatial coding pattern, so that the array presents a corresponding spatial distribution of transmission coefficients; different spatial coding patterns correspond to different transmission distributions, which are used for subsequent single-pixel measurement and image reconstruction.

9. The terahertz transmission liquid crystal metasurface fast imaging system according to claim 7, characterized in that, The array electrodes are divided into row electrodes and column electrodes. The row electrodes and column electrodes cross each other to form the bias loading point of each liquid crystal metasurface transmission unit. The control module establishes a potential difference between the selected row and column electrodes by row / column gating and level superposition, thereby realizing the bias voltage loading and spatial encoding state switching of the corresponding liquid crystal metasurface transmission unit.

Images