A time-domain coded high-speed atomic terahertz imaging method and system

Abstract

This application provides a time-domain coded high-speed atomic terahertz imaging method and system, relating to the field of terahertz imaging technology. Based on the time-domain coded sequence and the single visible light coded integral image, a coded projection relationship is established between the candidate terahertz full-field time-series image sequence (weighted superposition of each chip's coded value) and the single visible light coded integral image. Under the constraints of non-negative candidate images, spatial continuity, and sparse changes between adjacent time slices, the candidate terahertz full-field time-series image sequence is iteratively adjusted until the deviation meets the stopping condition. The resulting terahertz full-field time-series image sequence is then output according to the chip time order. This application can record rapid changes within a single coding cycle while retaining two-dimensional full-field parallel acquisition.

Description

Technical Field

[0001] This application relates to the field of terahertz imaging technology, and in particular to a time-domain encoded high-speed atomic terahertz imaging method and system. Background Technology

[0002] Terahertz full-field imaging can simultaneously acquire terahertz responses at different spatial locations within a test area using a two-dimensional detector surface. Atom-based terahertz imaging typically applies a terahertz field to an atomic interaction region formed by laser excitation, and then acquires a visible light intensity image corresponding to the spatial distribution of the terahertz field. This method can convert the two-dimensional terahertz field distribution into the visible light band for array readout, making it suitable for observing spatial differences in a sample within a field of view. When the state of the sample changes rapidly over time, it is also necessary to continuously acquire two-dimensional images corresponding to different time slices to distinguish the sequence of changes and their spatial propagation process.

[0003] However, the atomic level population and the establishment, decay, and integration of visible light signals have finite time responses. Visible light cameras also need to accumulate light signals within the exposure window. Existing terahertz full-field imaging based on atomic two-dimensional conversion screens typically uses a single exposure result as one frame image. The resolvable time interval is limited by both the atomic conversion response and the camera's frame-by-frame readout speed. When the terahertz field or the sample under test has changed multiple times within a single camera exposure cycle, the two-dimensional information from different time slices accumulates in the same frame. Existing direct framing methods struggle to distinguish the terahertz full-field distribution corresponding to each time slice from this accumulated result. Therefore, how to obtain terahertz full-field time-series images faster than the atomic conversion response and the camera's frame-by-frame readout speed while retaining the atomic two-dimensional conversion full-field readout capability is a major technical problem that needs to be solved in this field. Summary of the Invention

[0004] This application provides a nanosecond time-resolved terahertz visible light atom conversion full-field imaging method and system, which can record rapid changes within a coding cycle while retaining two-dimensional full-field parallel acquisition.

[0005] To solve the above-mentioned technical problems, the technical solution provided in this application is as follows.

[0006] This application provides a time-domain encoded high-speed atomic terahertz imaging method, including: Obtain a time-domain encoded sequence containing multiple chips and their corresponding encoded values. Control the terahertz high-speed modulation illumination source to output terahertz waves of corresponding intensities according to the chips based on the time-domain encoded sequence, and obtain an encoded terahertz pulse sequence. After collimating the coded terahertz pulse sequence, the sample to be tested is illuminated, so that the terahertz waves transmitted or reflected by the sample carry the two-dimensional spatial information of the sample changing over time, thus obtaining a terahertz image sequence. The terahertz image sequence is sequentially imaged onto a two-dimensional atomic conversion screen formed by the interaction between laser and atoms using a terahertz imaging lens. Based on the population change of atomic energy levels caused by the terahertz two-dimensional field distribution, the terahertz image sequence is converted into a two-dimensional visible light intensity sequence that maintains spatial correspondence. The exposure window covering the entire coding period is determined based on the temporal coding sequence. Within the exposure window, the two-dimensional intensity sequence of visible light is integrated once to obtain a single visible light coded integrated image. Based on the time-domain coding sequence and a single visible light coded integral image, a coding projection relationship is established between the candidate terahertz full-field time-series image sequence and the single visible light coded integral image after the weighted superposition of the coding values ​​of each chip. Under the constraints of non-negative candidate images, spatial continuity, and sparse changes between adjacent time slices, the candidate terahertz full-field time-series image sequence is iteratively adjusted until the deviation meets the stopping condition, and the terahertz full-field time-series image sequence arranged in chip time order is output.

[0007] On the other hand, a time-domain encoded high-speed atomic terahertz imaging system includes a terahertz high-speed modulation illumination source, a sample-carrying optical path, an atomic terahertz-visible light conversion module, a visible light high-speed image acquisition module, a synchronization control module, and an image reconstruction unit 8. The synchronization control module is used to acquire a time-domain encoded sequence containing multiple chips and corresponding encoded values, and to control the terahertz high-speed modulation illumination source to output encoded terahertz pulse sequences according to the chips; the sample-carrying optical path is used to enable the encoded terahertz pulse sequence to be transmitted or reflected by the sample under test to form a terahertz image sequence. The atomic terahertz-visible light conversion module is used to image terahertz image sequences onto an atomic two-dimensional conversion screen and convert them into visible light two-dimensional intensity sequences that maintain spatial correspondence. The synchronization control module is also used to control the visible light high-speed image acquisition module to perform one integration acquisition within the exposure window covering the complete coding cycle, so as to obtain a single visible light coded integration image. The image reconstruction unit is used to establish a coding projection relationship based on the temporal coding sequence and a single visible light coding integral image, and iteratively outputs a terahertz full-field temporal image sequence under the constraints of non-negativity, spatial continuity and sparse changes between adjacent time slices.

[0008] Compared with the prior art, the beneficial effects of this application are as follows.

[0009] This application controls a terahertz high-speed modulated illumination source to output coded terahertz pulse sequences according to a time-domain coding sequence, so that the two-dimensional spatial information of the sample under test at different times corresponding to different code chips carries a known time code. The spatial correspondence between the terahertz two-dimensional field distribution and the visible light two-dimensional intensity distribution is maintained through a terahertz imaging lens and an atomic two-dimensional conversion screen. An integral acquisition is performed within an exposure window covering the complete coding cycle, allowing time changes that would otherwise be faster than the atomic conversion response and the camera's frame-by-frame readout speed to be recorded in a decodeable manner in a single visible light coded integral image. Therefore, the division of imaging time slices is provided by the code chip order of the time-domain coding sequence, eliminating the need for the visible light camera to perform an exposure and readout for each time slice separately. This enables the recording of rapid changes within a coding cycle while preserving parallel acquisition of the entire two-dimensional field.

[0010] This application further establishes a coding projection relationship between the candidate terahertz full-field time-series image sequence and a single visible light coded integral image based on the time-domain coding sequence, and iteratively reduces the coding projection deviation under the constraints of non-negativity, spatial continuity, and sparse changes between adjacent time slices of the candidate image. Since the output sequence simultaneously satisfies the spatiotemporal constraints of the actual acquisition results and the scene under test, the temporal information mixed in the single integral image can be separated according to the chip time order, resulting in multiple quasi-continuous terahertz full-field images. Thus, this application addresses the problem described in the background art where information from different time slices accumulates and is difficult to distinguish in a single exposure, forming a complete processing chain from known time coding, two-dimensional atomic conversion, synchronous integral acquisition to constraint reconstruction. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the structure of a nanosecond time-resolved terahertz visible light atom conversion full-field imaging device in one embodiment of this application.

[0012] Figure 2 This is a flowchart of a nanosecond time-resolved terahertz visible light atom conversion full-field imaging method in one embodiment of this application.

[0013] Label Explanation: Figure 1 In the middle: 1-Terahertz high-speed modulation illumination source, 2-Sample to be tested, 3-Terahertz imaging lens, 4-Atomic gas cell, 5-Optical path unit, 6-High-speed visible light camera, 7-Signal acquisition unit, 8-Image reconstruction unit. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0015] See Figure 1 and Figure 2 As shown, to make the technical solution and working mechanism of the present invention clearer, the time-domain coded high-speed atomic terahertz imaging method and system are described in detail below in conjunction with system composition, signal flow direction and processing steps; the embodiments described are used to illustrate the coordination relationship between various technical features. The specific length of the time-domain coded sequence, chip duration, terahertz frequency band, atom type, laser wavelength, imaging magnification, number of camera pixels and reconstruction parameters can be determined according to the object under test, optical path conditions and target time range. In the absence of actual calibration or experimental basis, a certain value is not written as a fixed performance index.

[0016] The temporal coding sequence in this invention is a set of coded values ​​arranged chronologically. Each coded value remains valid for the duration of its corresponding chip. The coded value is used to control the output state of the terahertz high-speed modulation illumination source 1 during the chip period. The coded value can represent two states: on and off. It can also represent multiple discrete intensity levels or normalized continuous intensity levels, as long as the image reconstruction unit 8 can obtain the coded value corresponding to the actual output state and establish a forward projection relationship accordingly. Since the temporal coding sequence limits both the order of each chip and the contribution of each chip to the integral image, it simultaneously serves as both a time index and a measurement coding function.

[0017] The atomic two-dimensional conversion screen in this invention is an atomic interaction region formed by laser action within the atomic gas chamber 4, capable of responding to the terahertz field and outputting a visible light intensity distribution within a two-dimensional region, without requiring a physical fluorescent plate. This atomic interaction region is set at or within an allowable range coinciding with the image plane of the terahertz imaging lens 3, so that different spatial positions on the sample 2 under test correspond to different positions on the atomic two-dimensional conversion screen after terahertz imaging. The visible light signal output by the atomic two-dimensional conversion screen can be a change in fluorescence intensity or a change in visible light intensity caused by changes in atomic state and obtained by optical readout, as long as the change in visible light intensity can characterize the distribution of the incident terahertz two-dimensional field.

[0018] The single visible light coded integral image in this invention is a frame of digital image formed by the visible light high-speed image acquisition module within a continuous exposure window. This exposure window covers a complete coding cycle or a set of continuous chips that can form a complete coding relationship. "Single frame" means that the camera outputs a two-dimensional image data object after the exposure window ends, and does not limit the camera to using a global shutter, electronic shutter, intensified gating, or other exposure control structures. Each effective pixel of the visible light coded integral image records the cumulative visible light response of the corresponding spatial position of the pixel within the exposure window. Therefore, the same pixel value contains the weighted contribution of multiple chip time slices.

[0019] The terahertz full-field time-series image sequence in this invention includes multiple two-dimensional images arranged in chip time order. Each two-dimensional image corresponds to a chip, a time slice formed by merging adjacent chips, or a time interval determined based on reconstruction stability. "Full-field" means that each image covers the effective imaging area of ​​the atomic two-dimensional conversion screen, rather than being obtained by point-by-point mechanical scanning and stitching. "Quasi-continuous" means that adjacent images correspond to adjacent or partially overlapping time intervals, and does not mean a mathematically uninterrupted continuous time function. Each image can represent terahertz intensity, relative transmission response, relative reflection response, or other terahertz imaging quantities obtained after calibration.

[0020] In one overall architecture, the time-domain coded high-speed atomic terahertz imaging system includes a terahertz high-speed modulation illumination source 1, a sample-carrying optical path, an atomic terahertz-visible light conversion module, a visible light high-speed image acquisition module, a synchronization control module, and an image reconstruction unit 8. The terahertz high-speed modulation illumination source 1 is located on the input side of the terahertz propagation link, the sample-carrying optical path is located between the illumination source and the atomic terahertz-visible light conversion module, the field of view of the visible light high-speed image acquisition module covers the atomic two-dimensional conversion screen, the synchronization control module provides time references to the illumination source and the image acquisition module respectively, and the image reconstruction unit 8 receives the digital image output by the image acquisition module and reads the time-domain coded sequence corresponding to the digital image.

[0021] Along the direction of the terahertz signal flow, the encoding sequence generation unit first converts the time-domain encoding sequence into a digital control state that the driving unit can recognize. The driving unit then converts the digital control state into the bias, current, switching, or amplitude control quantities of the solid-state terahertz source. The solid-state terahertz source outputs a terahertz wave that varies with the code chip. The collimation unit expands, limits, or collimates the terahertz wave. The sample-carrying optical path allows the terahertz wave to pass through the test sample 2 or be reflected from the surface of the test sample 2. Because the test sample 2 has different effects on the terahertz wave at different positions and at different times, the terahertz field propagating to the terahertz imaging lens 3 carries both spatial and temporal information.

[0022] Along the optical conversion and acquisition link, the terahertz imaging lens 3 images the terahertz two-dimensional field carrying sample information onto the atomic two-dimensional conversion screen inside the atomic gas chamber 4. The optical path unit 5 establishes an atomic state that can respond to the terahertz field inside the atomic gas chamber 4. After the terahertz field changes the atomic state, the visible light output intensity changes accordingly. The high-speed visible light camera 6 images the atomic two-dimensional conversion screen onto the photosensitive surface and integrates the continuously appearing two-dimensional intensity distribution of visible light within the exposure window given by the synchronous control module. The signal acquisition unit 7 then completes analog-to-digital conversion, frame reading, and data encapsulation to form a single visible light coded integrated image.

[0023] Along the data processing link, the synchronization control module records the encoding sequence identifier, number of chips, chip order, chip duration, encoding cycle start time, and exposure start and end times corresponding to this exposure, and associates this information with a single visible light encoded integral image. The image reconstruction unit 8 first preprocesses the integral image based on dark field data, flat field data, and spatial registration data, and then establishes the encoded projection of the candidate temporal image based on the actual encoded value. Subsequently, the difference between the measured integral image and the candidate encoded projection is used as the basis for data consistency. The candidate temporal image is adjusted in combination with nonnegativity, spatial continuity, and temporal sparsity, and finally outputs a terahertz full-field temporal image sequence with time index and spatial coordinate information.

[0024] In atomic two-dimensional conversion full-field imaging, the atomic two-dimensional conversion screen simultaneously receives the incident terahertz field in the entire two-dimensional region, thus preserving spatial parallelism. However, the establishment and decay of atomic states will superimpose the terahertz responses arriving sequentially within a short period of time onto the visible light output. Meanwhile, if the camera directly resolves each rapidly changing moment frame by frame, it needs to complete exposure, readout, and preparation for the next frame within each time slice. When the target time slice is shorter than the atomic response time or the camera frame period, adjacent time slices will be mixed in the physical response or exposure integration, resulting in a single frame that can only reflect the cumulative intensity within a time range and cannot directly indicate the order of spatial changes within that time range.

[0025] This invention pre-assigns known coding values ​​to different time slices at the terahertz illumination end, maintains the correspondence between the sample's spatial position and the atomic two-dimensional conversion screen position at the conversion end, ensures that an exposure window completely covers the required coding slice at the acquisition end, and reconstructs the time dimension at the processing end using the same coding values. Since the contribution of each candidate time slice to the integral image is pre-determined by the temporal coding sequence, the integral image is no longer regarded as merely an indistinguishable cumulative result, but as a measurement result formed by projecting the candidate time series image through known coding. Based on this, spatiotemporal constraints consistent with fast scenes are added, allowing the selection of sequences that conform to the spatial structure and temporal variation patterns from multiple candidate sequences that satisfy the same integral measurement, thereby forming an executable single-frame coded integral image reconstruction path.

[0026] In this embodiment, the time-domain encoded high-speed atomic terahertz imaging method is executed collaboratively by a synchronization control module, a terahertz high-speed modulation illumination source 1, a sample-carrying optical path, an atomic terahertz-visible light conversion module, a visible light high-speed image acquisition module, and an image reconstruction unit 8. The overall process includes steps S101 to S105. Step S101 generates an encoded terahertz pulse sequence, step S102 enables the encoded terahertz pulse sequence to carry the two-dimensional spatial information of the sample 2 under test, step S103 completes the atomic terahertz-visible light two-dimensional conversion, step S104 completes synchronous single-integration acquisition, and step S105 completes encoded reconstruction with spatiotemporal constraints. The output of the previous step is used as the input of the next step until a terahertz full-field time-series image sequence arranged in chip time order is obtained.

[0027] In step S101, the synchronization control module acquires the time-domain encoded sequence to be executed. The time-domain encoded sequence includes the number of chips, the chip order, the encoded value of each chip, and the duration of a single chip. The synchronization control module first checks whether the number of chips is positive, whether each encoded value is within the control range allowed by the illumination source, and whether the total duration of the sequence can be covered by the camera exposure window. After the checks are passed, a loading command is sent to the encoding sequence generation unit. When the period start signal arrives, the encoding sequence generation unit outputs the encoded values ​​according to the chip order. The driving unit converts the encoded values ​​into the output control quantity of the solid-state terahertz source, thereby obtaining an encoded terahertz pulse sequence with intensity varying with the chips. Since the illumination intensity of each pulse time slice corresponds one-to-one with the stored encoded value, the contribution of each time slice in the subsequent integration measurement has a traceable encoded basis.

[0028] In step S102, the collimation unit receives the coded terahertz pulse sequence output in step S101 and adjusts the propagation direction and lateral dimension of the terahertz beam to ensure that the effective beam covers the target area of ​​the sample 2 under test. In transmission mode, the sample carrier places the sample 2 under test between the collimation unit and the terahertz imaging lens 3. The terahertz field transmitted through the sample 2 under test forms a spatial distribution variation due to differences in the sample's internal structure, thickness, density, or material response. In reflection mode, the path switching unit guides the illumination beam to the surface of the sample 2 under test and guides the returned reflected terahertz field to the terahertz imaging lens 3. Since the terahertz field during each chip passes through the same spatial imaging link, the resulting terahertz image sequence inherits the chip order in time and carries two-dimensional information of the target area of ​​the sample 2 under test in space.

[0029] In step S103, the terahertz imaging lens 3 receives the terahertz image sequence obtained in step S102 and images the object plane of the sample 2 to be tested onto the image plane of the atomic two-dimensional conversion screen in the atomic gas cell 4; the optical path unit 5 performs spatial filtering, beam expansion and optical sheet shaping on the probe light, coupling light and Rydberg excitation light respectively, so that multiple laser beams converge in the target area of ​​the atomic gas cell 4 and establish an atomic energy level population that can be affected by the terahertz field. The incident terahertz two-dimensional field changes the corresponding atomic energy level coupling or population state at each spatial position. The change of atomic state is then manifested as a change in visible light intensity, thereby converting the terahertz image sequence that arrives in sequence into a visible light two-dimensional intensity sequence that maintains spatial correspondence; since this step does not scan the spatial position point by point, multiple spatial positions in the area to be tested can be converted in parallel at the same time.

[0030] In step S104, the synchronization control module determines the exposure window based on the periodic start signal of the time-domain coding sequence used in step S101, ensuring that the starting edge of the exposure window is located before or corresponding to the start time of the first valid chip, and that the ending edge of the exposure window is located after or corresponding to the end time of the last valid chip. The high-speed visible light camera 6 performs only one integration exposure within this exposure window, accumulating the two-dimensional intensity distribution of visible light corresponding to each chip output in step S103 onto the same photosensitive array. After the exposure is completed, a frame of analog or digital image is output, and the signal acquisition unit 7 digitizes it and extracts the effective area to obtain a single visible light coded integration image. Since the camera does not need to complete a frame readout at the end of each chip, the chip time interval can be determined by the illumination modulation and synchronization capability, rather than being directly equal to the camera's frame-by-frame period.

[0031] In step S105, the image reconstruction unit 8 reads the single visible light coded integral image obtained in step S104 and the time-domain coded sequence used in step S101, establishes the same number of candidate terahertz full-field images according to the number of chips, and weights the candidate images pixel by pixel according to the coded values ​​of each chip and accumulates them to form a candidate coded projection image; the image reconstruction unit 8 compares the candidate coded projection image with the measured visible light coded integral image, and adjusts each candidate image round by round in the direction of reducing the deviation between the two under the conditions that the candidate image is non-negative, the changes of adjacent pixels in the same time slice remain spatially continuous, and the changes between adjacent time slices remain sparse; when the projection deviation reaches the preset tolerance, no longer decreases significantly in multiple rounds, or the number of iterations reaches the upper limit, the update stops, and then the time index is configured for each candidate image according to the chip order and the terahertz full-field time sequence image is output.

[0032] In one possible implementation, the time-domain coding sequence employs binary coding, with each chip's coded value being either a first state or a second state. The first state causes the solid-state terahertz source to output a terahertz wave for imaging, while the second state causes the solid-state terahertz source to stop outputting or output a reference intensity lower than that of the first state. To avoid the corresponding time slice completely losing its observation contribution when the coded value is zero, the second state can also be set to a known non-zero reference intensity. The image reconstruction unit 8 normalizes the actual output intensities of the first and second states respectively and uses them as coding weights. Using two distinguishable states simplifies digital driving and timing alignment, while retaining the weak contribution of each chip to the integral measurement through the non-zero reference state.

[0033] In another possible implementation, the time-domain coding sequence employs multi-level intensity coding. The coding sequence generation unit outputs multiple discrete levels, and the driving unit searches for a pre-calibrated driving control quantity based on the level. The solid-state terahertz source outputs the terahertz intensity corresponding to that level during each chip period. The calibration table records at least the coding level, driving control quantity, measured relative terahertz intensity, applicable frequency state, and valid flag. If multiple calibration records exist for the same coding level, the latest record that is consistent with the current frequency state and environmental state and is valid is selected. Multi-level intensity coding can increase the difference in the contribution of different time chips to the integral image, but the actual relative intensity after calibration is used during reconstruction, rather than directly treating the digital level as the ideal output intensity. Therefore, it can reduce the influence of illumination source nonlinearity on the coding projection relationship.

[0034] The time-domain coded sequence can be generated by pseudo-random rules, orthogonal coding rules, or non-uniform rules determined according to the time characteristics of the process under test. When using pseudo-random rules, the coding sequence generation unit generates coded values ​​chip by chip according to a preset shift feedback relationship after a given initial state and sequence length, and saves the initial state and generation results together. When using orthogonal coding rules, a row can be selected from the Hadamard coding matrix as the time-domain coded sequence for a single exposure, or multiple rows can be selected sequentially in multiple measurement periods to form multiple single-exposure data. When using non-uniform rules, shorter chips can be set in the expected fast-changing time interval, and longer chips can be set in the expected slow-changing time interval, and the start and end times of each chip can be saved separately, so that the output time index is consistent with the actual chip boundary.

[0035] The encoding period is determined by the number of chips and the duration of a single chip. When the durations of all chips are the same, the number of chips is multiplied by the duration of a single chip to obtain the sequence duration from the first chip to the last chip. When the durations of the chips are different, the durations of each chip are accumulated sequentially according to the chip order to obtain the sequence duration. The synchronization control module can also add a preamble interval before the valid chips and a tail interval after the valid chips. The preamble interval is used to wait for the drive and camera to enter a stable state, and the tail interval is used to receive the tail response of atomic transformation and camera exposure. When the exposure window covers the preamble interval, the valid chip interval, and the tail interval that needs to be retained, the image reconstruction unit 8 records the baseline contribution of these intervals in the forward projection.

[0036] Solid-state terahertz sources can be terahertz radiation sources capable of accepting high-speed digital control or analog amplitude control. The output interface of the drive unit is matched with the control interface of the selected terahertz source. When there is a settling time in the response of the terahertz source to the control quantity, the synchronous control module updates the drive control quantity in advance before the chip boundary, so that the measured terahertz output reaches the target state within the effective range of the chip. The advance is obtained by repeatedly measuring the control signal and the terahertz monitoring signal. Specifically, the edge time of the control signal is recorded first, and then the time when the terahertz monitoring signal reaches the target change ratio is recorded. The time difference between the two is used as the initial advance. Subsequently, the measurement is repeated at different output levels to form a level-related advance table.

[0037] To ensure that the coding weights used in reconstruction reflect the actual illumination, the terahertz high-speed modulated illumination source 1 can be equipped with a reference monitoring branch. The reference monitoring branch separates a small portion of the energy from the terahertz beam before or after collimation and outputs chip-by-chip monitoring values. The signal acquisition unit 7 saves the monitoring values ​​of each chip according to the exposure correspondence. The image reconstruction unit 8 prioritizes using the monitoring values ​​to form the actual coding weights. If a chip monitoring value is missing, it is first determined whether adjacent chips have the same target coding level. In the same case, the average result of adjacent valid monitoring values ​​is used as a substitute. In different cases, the relative intensity of the corresponding level in the calibration table is used as a substitute, and a low reliability mark is set for the chip so as to reduce the amount of related updates for the chip during reconstruction.

[0038] In the transmission imaging implementation, the collimated terahertz beam passes sequentially through the sample under test 2 and the terahertz imaging lens 3 along the first propagation direction. The sample carrier ensures that the target area of ​​the sample under test 2 is located within the object plane range corresponding to the terahertz imaging lens 3. The terahertz imaging lens 3 images the two-dimensional field distribution after transmission through the sample onto the atomic two-dimensional conversion screen. The sample under test 2 can remain stationary within the encoding period, or it can change due to external motion, internal state changes, or movement of the transport mechanism, as long as the target area still falls within the effective field of view within the required encoding period. The image obtained by transmission imaging mainly reflects the transmission effect of terahertz waves at different spatial positions. The output image can be retained as a relative intensity, or it can be compared with a reference image without a sample to form a relative transmission response image.

[0039] In the reflection imaging implementation, the path switching unit includes a terahertz beam splitting structure, a deflection structure, or a separate transmit and receive structure. The illumination beam reaches the surface of the sample 2 under test through the path switching unit. The reflected terahertz wave returned by the sample 2 under test propagates to the terahertz imaging lens 3 along a receiving path that partially coincides with or separates from the incident path. The synchronization control and encoding method is the same as that of transmission imaging. The image reconstruction unit 8 recovers the two-dimensional distribution of reflected terahertz waves in the target area for each time slice. Since the reflection mode does not require the terahertz wave to pass through the sample 2 under test, it can be used for objects under test that can only be approached from one side. Moreover, the path difference of this mode only changes the formation method of the terahertz image sequence and does not change the subsequent atomic two-dimensional conversion, integration acquisition, and encoding reconstruction processing chain.

[0040] The terahertz imaging lens 3 can be composed of at least one transmissive terahertz lens, at least one reflective terahertz mirror, or a combination of transmissive and reflective components. When setting the imaging magnification, first determine the lateral dimension of the target area of ​​the sample 2 on the object plane, then determine the lateral dimension of the effective area of ​​the atomic two-dimensional conversion screen on the image plane, and select the imaging magnification that makes the target area fall into the effective conversion area based on the ratio of the two. If the target area exceeds the boundary of the atomic two-dimensional conversion screen after imaging, increase the object field of view or decrease the magnification. If the target area only occupies a small part of the effective conversion area, increase the magnification or truncate the corresponding area during reconstruction, so that the spatial sampling matches the range of the atomic two-dimensional conversion screen.

[0041] Spatial registration is accomplished using reference markers or terahertz calibration elements of known shape. Specifically, the calibration element is placed on the sample support, and a visible light image corresponding to the calibration element is acquired under stable illumination. Multiple feature positions of the calibration element are identified and their positions in the camera pixel coordinates are recorded. A spatial mapping table is established based on the correspondence between the object plane coordinates of the calibration element and the camera pixel coordinates. During the formal measurement, the signal acquisition unit 7 crops or resamples the camera image into terahertz imaging coordinates according to the spatial mapping table. If an output position falls between four adjacent camera pixels, the contribution of each pixel is determined and weighted based on the relative distance from that position to the center of the four adjacent pixels, thereby obtaining the estimated intensity of the corresponding output position.

[0042] To reduce the impact of invalid data at the field of view edges on reconstruction, the spatial mapping table also sets a valid flag and reliability for each output position. The valid flag is determined based on whether the position falls within the common range of the illumination coverage area, the atomic 2D conversion screen area, and the camera field of view. The reliability is determined based on the response intensity, repeatability, and distance from the upper limit of saturation in the flat-field image. The image reconstruction unit 8 only establishes candidate temporal pixels for valid positions, reduces the contribution of data consistency terms to low-reliability positions, and adds a valid region mask at output to avoid interpreting the values ​​outside the field of view, occluded areas, or long-term saturation areas as the actual terahertz response of the sample 2 under test.

[0043] The atomic terahertz-visible light conversion module includes a terahertz imaging lens 3, an atomic gas chamber 4, and an optical path unit 5. The atomic gas chamber 4 is a sealed container that can contain alkali metal atomic vapor and allow terahertz waves and lasers to pass through. The alkali metal atoms can be rubidium atoms or cesium atoms. The incident and exit regions of the atomic gas chamber 4 in the terahertz propagation direction adopt a structure that allows transmission of the terahertz frequency band used, and the corresponding region in the laser propagation direction adopts a structure that allows transmission of the laser used. The specific materials and thickness are selected according to the working frequency band, sealing requirements, and mechanical strength, and are not limited to a single material.

[0044] The optical path unit 5 includes a probe light branch, a coupling light branch, and a Rydberg excitation light branch. Each branch includes at least one of the following: a light source, a frequency control unit, a spatial filter unit, a beam expander unit, a polarization control unit, a reflection or beam combiner unit, and a light sheet shaping unit. The frequency control unit ensures that the corresponding laser meets the excitation conditions for the selected atomic energy level transition. The spatial filter unit removes high spatial frequency components that affect the uniformity of the light sheet. The beam expander expands the beam to cover the size of the target conversion region. The polarization control unit adjusts the polarization state according to the atomic transition selection rule. The light sheet shaping unit compresses the beam in one direction and expands it in another direction, so that multiple laser beams form a two-dimensional intersection region in the atomic gas chamber 4.

[0045] When establishing the two-dimensional atomic conversion screen, each laser branch is first activated without coded terahertz wave incident. The propagation direction of each laser is gradually adjusted so that the effective spots of the probe light, coupling light, and Rydberg excitation light overlap in the target image area. Then, the frequency control of each branch is adjusted so that the atoms form an energy level population state that can be changed by the terahertz field in the overlapping area. The beam expander and light sheet shaping part are then adjusted so that the overlapping area covers the image area formed by the terahertz imaging lens 3. The high-speed visible light camera 6 acquires a reference image without terahertz incident. If there are obvious dark areas, stripes, or saturation areas in the reference image, the beam position, beam size, or light power is adjusted until the effective area meets the preset uniformity and dynamic range requirements.

[0046] After the encoded terahertz image reaches the atomic two-dimensional conversion screen, the terahertz field couples with the corresponding atomic energy levels at different spatial locations, causing the atomic energy level population or coherent state to change with the local terahertz field. This change is manifested as a two-dimensional intensity distribution of visible light through the fluorescence intensity generated by atomic transitions or the readout intensity of the probe light. The terahertz imaging lens 3 maintains the correspondence between the object plane position and the conversion screen position, while the high-speed visible light camera 6 maintains the correspondence between the conversion screen position and the camera pixel position. Therefore, the same output pixel continuously observes the same corresponding position of the sample 2 under test during different code chips. Since the spatial relationship remains unchanged throughout the entire encoding cycle, the image reconstruction unit 8 can establish an integral relationship for each spatial location using the same set of temporal coding weights.

[0047] The atomic conversion response may cause the visible light contribution of a certain chip to extend to adjacent chips. When this effect needs to be considered, the time response of the atomic two-dimensional conversion screen is first measured using a terahertz test sequence with known time edges. Specifically, under stable spatial field conditions, the terahertz illumination is switched from a first intensity state to a second intensity state. The delay of the camera gating relative to the switching time is continuously changed and the response is collected. The normalized visible light changes are recorded in chronological order to obtain a discrete time response sequence. When establishing the coded projection, the image reconstruction unit 8 does not directly use the original coded values, but instead allocates the response contribution of each chip's coded value to the current chip and subsequent chips according to the discrete time response sequence to obtain the effective coded weights corrected by the atomic response.

[0048] When performing textual calculations on the time response, first obtain the encoded values ​​of each input chip before the current output chip, then obtain the time intervals between these input chips and the current output chip, and query the discrete sequence of the time response based on the time intervals to obtain the response ratio of each input chip. Multiply the encoded value of each input chip by the corresponding response ratio, and finally sum the multiplications to obtain the effective encoded value of the current output chip. If the query time falls between two recorded time points, then perform linear interpolation on the two response ratios based on the relative distance between the query time and the two recorded time points. If the query time exceeds the calibrated response range, then set its response ratio to the tail boundary value or zero, and indicate the boundary processing method used in the reconstructed record.

[0049] To correct for the difference in response between the atomic 2D conversion screen and the camera to different spatial locations, flat-field images can be acquired under conditions where there is no sample 2 to be tested and the terahertz illumination remains stable, while dark-field images can be acquired under conditions where the terahertz illumination is off. For each effective pixel in the formal measurement image, the corresponding pixel value in the dark-field image is subtracted from the measured value of that pixel to obtain the first pixel value after removing the background. Then, the corresponding pixel value in the flat-field image is subtracted from the corresponding pixel value in the dark-field image to obtain the effective response range at that location. Finally, the ratio of the first pixel value to the effective response range is used to form the normalized pixel value. When the effective response range is zero or lower than the stability threshold, the location is set as an invalid location, no division is performed, and it is excluded by the effective area mask.

[0050] When the temperature, laser frequency, or optical power in the atomic gas chamber 4 slowly drifts, reference state data can be collected before and after the formal coding cycle, and the two reference state data can be interpolated according to the collection time. For any time slice within the coding cycle, the relative position of the time slice from the previous reference collection time and the next reference collection time is first determined, and then the contribution ratio of the previous and next references to the time slice is determined according to the relative position. The two reference data are weighted using the corresponding ratio to obtain the estimated reference for the time slice. During reconstruction, the estimated reference is subtracted from the candidate time slice or the forward projection so that the slow drift is not misjudged as a rapid time change of the sample 2 under test.

[0051] The synchronization control module sets the master clock source and generates a chip clock, a cycle start signal, and a camera trigger signal from the master clock source. The encoding sequence generation unit updates the encoding value at the edge of the chip clock. The high-speed visible light camera 6 starts exposure after the camera trigger signal arrives. To determine the actual time relationship, the synchronization control module simultaneously records the encoding output monitoring signal and the camera exposure valid signal, and compares their valid edges. If the exposure start is later than the first valid chip interval or the exposure end is earlier than the last valid chip interval, the measurement is marked as incomplete time coverage. Instead of directly entering formal reconstruction, the acquisition is retried or a shortened encoding relationship is established according to the actual covered chip range.

[0052] The exposure window does not necessarily have to be exactly equal to the theoretical coding period, but it must be able to enable the image reconstruction unit 8 to determine the contribution of each effective time interval to the integral image. When the exposure window is longer than the coding period, the excess part can be set as a dark reference interval with terahertz off, or it can be set as a known reference intensity interval. The image reconstruction unit 8 estimates the additional integral contribution based on the duration of the excess part and the known intensity and subtracts it from the measured integral image. When the exposure window only covers a part of the complete coding period and that part is insufficient to form the coding relationship used, the signal acquisition unit 7 discards the frame and records the reason for the discard, thereby preventing incomplete data from entering the timing reconstruction.

[0053] The imaging lens of the high-speed visible light camera 6 is focused on the area where the atomic two-dimensional conversion screen is located. The spatial positions of the photosensitive pixels and the conversion screen are associated through the aforementioned spatial mapping table. Before the camera is exposed, the signal acquisition unit 7 checks whether the camera gain, exposure mode, effective bit depth and photosensitive area are consistent with the configuration of this measurement. If the configuration is inconsistent, the target configuration is written first and the feedback status is read. After the exposure is completed, the signal acquisition unit 7 reads the image data and camera status data, and writes the coded sequence identifier, trigger time, exposure duration, camera configuration, spatial mapping version, dark field version and flat field version into the same data record to ensure that the reconstruction can trace all the inputs used in this measurement.

[0054] Dark field correction is performed pixel by pixel. If the actual measured value is less than the dark field value, the subtraction result is set to zero and the under-range flag is retained. If the actual measured value reaches the camera's saturation limit, the saturation flag is set and the reliability of that position is reduced. Flat field correction is only performed on pixels whose effective response range is greater than the stable threshold. When the correction result exceeds the preset normalization limit, it is truncated to the limit and the out-of-bounds flag is retained. For isolated bad pixels, the median value of its upper, lower, left, and right effective neighboring pixels can be used to replace it. If there are not enough adjacent effective pixels, the bad pixel is kept invalid and is not filled across a large area.

[0055] After spatial registration, the image reconstruction unit 8 obtains a visible light encoded integral image on a regular two-dimensional grid. The horizontal and vertical sampling intervals of the grid are determined based on the camera pixel size, optical magnification, and the effective resolution of the atomic two-dimensional conversion screen. If the original sampling density of the camera is higher than the effective spatial resolution, multiple camera pixels within the same output grid range can be summed or averaged to reduce the amount of data. Summing is suitable for maintaining the integrated light quantity, while averaging is suitable for maintaining the normalized intensity. If the original sampling density is lower than the target grid requirement, new spatial details are not claimed to be obtained through interpolation. Instead, the original effective sampling interval is retained, or the interpolated pixels are explicitly used only for display in the output.

[0056] Before proceeding with temporal reconstruction, background region identification can be performed on the visible light coded integral image. The background region is determined by a no-sample-reference area, a fixed occlusion area, or a user-defined area outside the field of view. When calculating the representative value of the effective pixels within the background region, saturated and invalid pixels are first removed. Then, the pixel values ​​are sorted and the corresponding value at the middle position is taken, or the average value of the remaining pixels is taken. Subsequently, the representative value is subtracted from each pixel in the effective imaging area. When the subtraction result is negative, it is set to zero. This process can reduce the influence of the background light that changes across the entire field of view on the coded projection deviation, while not changing the relative spatial position of the structure under test.

[0057] Image reconstruction unit 8 represents the terahertz full-field time-series image sequence to be recovered within one coding cycle as multiple two-dimensional candidate images arranged in chip order. The horizontal and vertical dimensions of each candidate image are consistent with the preprocessed visible light coded integral image. For any spatial location, the pixel values ​​of multiple candidate images at that location constitute the time series to be recovered, and the pixel values ​​of the measured integral image at that location constitute a coded integral measurement. Since an integral measurement at a single spatial location cannot uniquely determine multiple arbitrary time values ​​without other conditions, this embodiment does not simplify the reconstruction process to direct inversion. Instead, it simultaneously uses the spatial continuity of the full field of view, the sparsity of changes between adjacent time slices, and the non-negativity of pixels to determine the output sequence that meets the set constraints from the candidate sequences that satisfy the coded integral relationship.

[0058] When establishing the coded projection, the actual coded weights corresponding to each chip are first obtained. The actual coded weights can be obtained from the calibration table, the reference monitoring branch, or the effective coded values ​​corrected by the atomic time response. For the first candidate image, each pixel value is multiplied by the coded weight of the first chip to obtain the first weighted image. For the second candidate image, each pixel value is multiplied by the coded weight of the second chip to obtain the second weighted image. The remaining candidate images are processed in the same way. Then, all weighted images are accumulated pixel by pixel at the same spatial position to obtain the candidate coded projection image. This process explicitly gives the forward path from the candidate temporal image to the single integral image. The data consistency deviation is calculated based on this forward path in each subsequent update.

[0059] Data consistency deviation is calculated per pixel. First, the corresponding pixel value in the measured visible light coded integral image is subtracted from the effective pixel value in the candidate coded projection image to obtain the projection residual at that position. Then, the absolute value of the projection residual or the product of the projection residual and itself is obtained as the deviation at that position. Finally, the deviations of all effective positions are summed or averaged to obtain the overall projection deviation for this round. If the effective area mask sets a reliability for a certain position, the deviation at that position is multiplied by the reliability before being accumulated. The reliability of saturated positions, invalid positions, and out-of-field positions is zero, and the reliability of low-response positions is less than that of normal positions, thereby avoiding a small number of abnormal pixels dominating the update direction.

[0060] The initial values ​​of the candidate sequence can be generated using a uniform distribution method. Specifically, the absolute values ​​of the actual coding weights of each chip are taken and accumulated to obtain the total coding weight. When the total coding weight is greater than zero, the pixel values ​​of each pixel in the measured integral image are divided by the total coding weight to obtain the baseline candidate image. This baseline candidate image is then copied to each chip position as the initial candidate sequence. If the total coding weight is zero, it is determined that the coding sequence used cannot form an effective illumination measurement and the reconstruction is terminated. Using a uniform distribution of initial values ​​does not pre-assume which time slice the change occurred in. Subsequent updates redistribute pixel intensity among time slices based on data consistency and spatiotemporal constraints.

[0061] When the output results of the previous coding cycle exist in the same scene, the historical initialization method can also be used. First, check whether the spatial mapping version, chip number and time order of the previous output sequence are consistent with the current measurement. If they are consistent, copy the previous output sequence as the current initial candidate sequence. If they are inconsistent, resample according to the current spatial grid and chip boundary. If the previous output sequence contains invalid values, negative values ​​or values ​​that exceed the current dynamic range, replace invalid values ​​with adjacent valid values, set negative values ​​to zero, and truncate out-of-range values ​​to the current upper limit. Historical initialization can reduce the number of iterations in stable scenes, but when the difference between the measured integral image and the forward projection of the historical sequence exceeds the switching threshold, uniformly distribute the initial value to avoid the historical state from masking the newly emerging rapid changes.

[0062] Each round of data consistency update includes three stages: forward projection, residual calculation, and residual re-projection. Forward projection accumulates candidate images according to the aforementioned coding weights. Residual calculation obtains the pixel-by-pixel difference between the candidate projection and the measured integral image. Residual re-projection multiplies the residual image with the coding weight of each chip to obtain the adjustment direction of the candidate image for the current residual. The image reconstruction unit 8 multiplies the adjustment direction with the current step size and then subtracts the product from the corresponding candidate image to form the candidate image after data consistency update. Chips with larger absolute values ​​of coding weights have a greater impact on the projection results, and their re-projection adjustment is correspondingly larger, thus ensuring that the update relationship is consistent with the forward coding relationship.

[0063] The current step size can be a preset initial step size or it can be adaptively adjusted according to changes in the projection deviation. When using the adaptive method, a trial update is first completed with the current step size, and the overall projection deviation after the trial update is calculated. If the deviation after the trial update is less than the deviation before the update, the trial update result is accepted, and the step size is maintained or appropriately increased as long as it does not exceed the upper limit of the step size. If the deviation after the trial update is not less than the deviation before the update, the trial update is canceled, the current step size is reduced, and the update is re-executed. When the deviation still does not decrease after the step size is reduced to the allowable lower limit, the candidate sequence before the update is retained, and the stop condition or anomaly flag is triggered to avoid repeated oscillations of the update under unstable step sizes.

[0064] Non-negativity constraints are executed after each round of data consistency updates. All candidate images are checked pixel by pixel. Pixel values ​​greater than or equal to zero remain unchanged, while pixel values ​​less than zero are set to zero. When the output imaging quantity adopts a differential form relative to the reference and allows positive and negative variations, non-negativity constraints are not directly applied to the differential values. Instead, the candidate sequence is represented as a non-negative absolute response sequence. Non-negativity constraints are first applied to the absolute response, and then the reference is subtracted from the reconstructed absolute response to obtain the differential output. Thus, the reconstructed internal structure still meets the physical condition that the light intensity or terahertz intensity is not negative, while allowing for relative increases and decreases in the display.

[0065] Spatial continuity constraints are executed independently for each candidate time slice. For a candidate pixel, its upper, lower, left, and right neighboring effective pixels are obtained, and the difference between the current pixel and its neighboring pixels is calculated. If the difference is less than the edge preservation threshold, the current pixel is updated with the weighted average of the current pixel and its neighboring pixels, where neighboring pixels that are closer or have higher reliability have a larger weight. If the difference reaches the edge preservation threshold, the smoothing weight across that direction is reduced or cross-edge averaging is not performed, thereby preserving the structural boundary of the sample 2 under test while suppressing isolated noise. Pixels located at the boundary of the effective region only use the effective neighboring pixels in the region, and retain the original value when there are no effective neighboring pixels.

[0066] The sparse constraint of adjacent time slice changes is applied along the time direction. For the same spatial location, the difference between the pixel value of the next candidate time slice and the pixel value of the previous candidate time slice is calculated sequentially to obtain a time difference sequence. For differences whose absolute value is less than the time change threshold, the difference is compressed to zero or reduced by a preset ratio. For differences whose absolute value is not less than the time change threshold, the direction of change is retained and the corresponding amount of the threshold is subtracted from the absolute value. Then, starting from the pixel value of the first time slice, the pixel values ​​of subsequent time slices are reconstructed by accumulating the processed time differences. This process keeps most adjacent time slices with no actual change consistent and retains a limited number of significant changes at the corresponding time positions, thereby providing a time structure constraint for single underdetermined measurements.

[0067] The time variation threshold can be determined based on dark field noise and repeated benchmark measurements. First, under the condition that the state of the sample 2 remains unchanged, multiple visible light coded integral images are repeatedly acquired. After completing the same preprocessing, the variation range between corresponding pixels is statistically analyzed. Then, the initial threshold is determined according to the representative value of the variation range in the effective area. If the expected variation range of the scene under test is small, the threshold is reduced but kept higher than the representative value of the benchmark noise. If the reconstruction result shows a large number of alternating positive and negative changes or flickering in adjacent time slices, the threshold is increased. The threshold adjustment record is associated with the coding sequence, camera configuration, and calibration version. The threshold obtained under one measurement condition cannot be unconditionally used under another measurement condition.

[0068] A complete iteration includes data consistency update, non-negativity constraint, spatial continuity constraint, and temporal sparsity constraint. After the constraints are executed, the candidate coded projection image and the overall projection deviation are recalculated. To compare the deviations of two consecutive rounds, the deviation reduction is obtained by subtracting the overall projection deviation of the current round from the overall projection deviation of the previous round. The relative reduction is obtained by the ratio of the deviation reduction to the overall projection deviation of the previous round. When the overall projection deviation of the current round is zero, the deviation stopping condition is met directly without performing division. The image reconstruction unit 8 writes the current round number, overall projection deviation, relative reduction, current step size, and constraint processing amount into the iteration record to determine whether the reconstruction has converged normally.

[0069] Stopping conditions include at least one of the following: the overall projection deviation is not greater than the preset tolerance, the relative decrease of the consecutive preset number of rounds is not greater than the stability threshold, the number of iteration rounds reaches the upper limit, and the deviation still cannot be reduced after the current step size reaches the lower limit. When the first two conditions are met, the current candidate sequence is output as normal and the corresponding stopping reason is recorded. When stopping due to the number of iteration rounds reaching the upper limit, the current sequence can still be output, but an incomplete convergence flag is added. When stopping due to step size failure or numerical abnormality, the candidate sequence with the smallest deviation in the most recent iteration is retained and a reconstruction abnormality flag is added. After clarifying the stopping conditions and abnormality flags, the program can determine the end time of the iteration and distinguish between normal output, incomplete convergence output, and abnormal output.

[0070] After reconstruction, the image reconstruction unit 8 numbers each time-slice image according to the chip sequence. The start time of the first time slice is determined by the period start signal, and the start and end times of subsequent time slices are obtained by accumulating the corresponding chip boundaries. Each output image is associated with at least the time slice number, start and end time, spatial mapping version, coding sequence identifier, overall projection deviation, stopping reason, and effective area mask. When displayed, it can be played continuously in chronological order, or adjacent time slices can be selected to calculate the difference image. When saving, multiple images can be written to a sequence file with time index or written to a single frame file respectively, while maintaining the association between pixel values ​​and metadata.

[0071] When the process under test does not meet the condition of sparse variation between adjacent time slices, an extended measurement method with multiple coding cycles can be adopted. Each coding cycle is still completed by the high-speed visible light camera 6 for one integration exposure, but different temporal coding sequences are used for different coding cycles, thereby obtaining multiple single-image coded integration images. The image reconstruction unit 8 obtains the coding weight of each candidate time slice in each coding cycle, forms multiple candidate coded projection images for the candidate sequence, and simultaneously reduces the deviation between each candidate projection and the corresponding measured integration image. If the multiple sets of codes used come from a complete orthogonal coding matrix and the number of measurements meets the solution conditions, the time-series image can be recovered according to the inverse transformation relationship of the coding matrix. If the number of measurements is still less than the number of time slices, joint reconstruction is continued using spatial continuity, temporal sparse variation and non-negative constraints.

[0072] In a data stream example of transmission imaging, the operator first fixes the sample 2 to be tested on the sample carrier and selects the spatial mapping configuration covering the target area. The synchronization control module reads a set of time-domain encoded sequences that have been calibrated by illumination and checks the number of chips, chip duration, and encoding weight version of the sequence. The terahertz high-speed modulation illumination source 1 outputs the encoded terahertz pulse sequence according to the sequence. The collimated beam passes through the sample 2 to be tested. The terahertz imaging lens 3 images the transmission field onto the atomic two-dimensional conversion screen. The high-speed visible light camera 6 completes one integrated exposure within a complete encoding cycle. The signal acquisition unit 7 sends the obtained image, along with the current sequence identifier and synchronization status, to the image reconstruction unit 8.

[0073] After receiving the data, the image reconstruction unit 8 first checks whether the encoded sequence identifier in the image record is consistent with the local encoded sequence version. If they are inconsistent, it retrieves the corresponding sequence from the synchronization control module. If it still cannot retrieve the sequence, the frame is marked as missing encoded information and reconstruction is stopped. After the versions are consistent, the background is subtracted from the current dark field image, the spatial response is corrected from the flat field image, and the effective area of ​​the atomic two-dimensional conversion screen is extracted according to the spatial mapping table. Reliability and invalid flags are set for saturation, under-range, and bad pixel positions, respectively, to obtain the visible light encoded integral image, reliability map, and effective area mask used for reconstruction.

[0074] Subsequently, the image reconstruction unit 8 establishes candidate time-series images based on the number of chips, and uses the result of dividing the measured integral image by the total coding weight as the initial image for each candidate time slice. In the first round, the initial images are weighted and accumulated according to the coding weight of each chip to obtain the first candidate coded projection image. The residual between the first candidate image and the measured integral image is calculated, and the residual is then back-projected to each candidate time slice according to the corresponding coding weight. After the back-projection update is completed, the negative values ​​are set to zero, the spatial differences smaller than the edge preservation threshold are smoothed in the neighborhood, and the differences between adjacent time slices smaller than the time change threshold are compressed to obtain the first round of constrained candidate sequences and the first round of overall projection deviation.

[0075] In subsequent rounds, forward projection, residual back projection, and constraint processing are repeated, and the update step size is adjusted according to whether the deviation has decreased. When the overall projection deviation reaches the preset tolerance, the image reconstruction unit 8 stops iterating, configures the time index of the current candidate sequence according to the chip order, and keeps the pixels outside the effective area as invalid. If the iteration reaches the upper limit but the deviation still does not reach the tolerance, the candidate sequence with the smallest deviation in each iteration is selected as the output with the incomplete convergence flag. This output is not described as error-free recovery, so that the user can decide whether to re-acquire or change the coding sequence based on the reconstruction status.

[0076] The output terahertz full-field time-series image sequence can be used to observe the two-dimensional response changes of the target region of the test sample 2 within the coding period. Each image corresponds to a chip time interval, and the pixel coordinates correspond to the position of the test sample 2 through a spatial mapping table. If it is necessary to generate a change map of adjacent time chips, the corresponding pixels of the previous time chip are subtracted from each effective pixel in the subsequent time chip image to obtain the change amount in both positive and negative directions. Then, the regions with significant changes are obtained by sorting according to the absolute value of the change amount or by filtering according to a threshold. This post-processing only uses the already output time-series images and does not change the main processing chain from coded illumination to time-series image output.

[0077] The terahertz high-speed modulation illumination source 1 includes a clock unit, an encoding sequence generation unit, a driving unit, a solid-state terahertz source, and a collimation unit. The clock unit receives the master clock from the synchronization control module or is locked to the master clock by its own clock. The encoding sequence generation unit includes a sequence storage area for storing encoded values, counting logic for reading encoded values ​​by address, and interface logic for outputting the current encoding state. The driving unit converts the digital encoding state into the control signal required by the solid-state terahertz source and returns the drive ready and output monitoring status to the synchronization control module. The collimation unit is located on the output side of the solid-state terahertz source to keep the beam center and coverage range within the allowable variation range during different chip periods.

[0078] The sample-carrying optical path includes a sample-carrying section and a path switching section. The sample-carrying section provides a carrying position for fixing, transporting, or positioning the sample 2 under test, and sets a mechanical or optical reference that can repeatedly determine the target area. The path switching section connects the collimation unit, the sample 2 under test, and the terahertz imaging lens 3 in the transmission state, and connects the collimation unit, the reflecting surface of the sample 2 under test, and the terahertz imaging lens 3 in the reflection state. Path switching can be achieved by moving the reflecting component, switching the beam splitting component, or enabling different transmit and receive paths. After switching is completed, the synchronization control module reads the path status, and the image reconstruction unit 8 selects the transmission calibration data or the reflection calibration data accordingly.

[0079] In the atomic terahertz-visible light conversion module, the terahertz imaging lens 3 is installed on the terahertz receiving path between the sample under test 2 and the atomic gas cell 4. The atomic gas cell 4 is installed in the image plane area of ​​the imaging lens. Multiple laser branches of the optical path unit 5 are arranged around the atomic gas cell 4 and the beams converge inside the gas cell. To maintain long-term spatial relationship, each component mounting base has a position adjustment and locking structure. During the adjustment stage, the position of the imaging lens, atomic gas cell 4 or light sheet shaping part is changed. After locking, it remains unchanged within a measurement batch. When any component affecting spatial mapping is readjusted, the spatial mapping version is updated and flat field data is reacquired to prevent pixel position mismatch caused by using the old mapping.

[0080] The high-speed visible light image acquisition module includes a high-speed visible light camera 6, a camera imaging lens, and a signal acquisition unit 7. The camera imaging lens images the atomic two-dimensional conversion screen onto the camera's photosensitive surface. The high-speed visible light camera 6 receives the trigger from the synchronization control module and returns an exposure valid signal. The signal acquisition unit 7 connects to the camera data interface and encapsulates a frame image into a single visible light coded integral image. The signal acquisition unit 7 may also include a buffer area, which temporarily stores the data of the next measurement cycle when the image reconstruction unit 8 processes the previous frame. The buffer entries are sorted by cycle identifier. If the buffer capacity reaches the upper limit, data that is synchronized and has no saturation anomalies is retained first, and the cycle identifier and reason of the discarded entries are recorded.

[0081] The synchronization control module can be implemented by a programmable logic device, microcontroller, timing control card or a combination thereof. It internally stores a uniform time count value and generates control signals with a defined relative delay for the illumination source and camera. The configuration interface receives the number of chips, chip duration, preamble interval, end interval and camera trigger delay, and writes these parameters into the timing register after checking whether they exceed the hardware allowable range. During operation, the synchronization control module writes the actual trigger time, illumination ready state, camera exposure state and abnormal events into the measurement log. The image reconstruction unit 8 only performs formal reconstruction on data with complete cycle start records and exposure coverage records.

[0082] The image reconstruction unit 8 can be implemented by a general-purpose processor, a graphics processor, a programmable logic device, or a combination thereof, and includes an image input interface, an encoded data interface, a calibration data storage area, a candidate sequence storage area, a projection calculation unit, a constraint processing unit, and a result output interface. The projection calculation unit performs candidate image weighting and residual back-projection according to the encoded weights, the constraint processing unit performs non-negative truncation, spatial neighborhood processing, and temporal difference compression, and the result output interface outputs multiple two-dimensional images and their metadata in chip time order. Different computing units can process different spatial blocks in parallel, but adjacent blocks retain overlapping boundaries when performing spatial continuity constraints to avoid artificial discontinuities at block boundaries.

[0083] The system may also include a display and storage unit. The display unit is used to present a single visible light coded integral image, a reconstructed terahertz full-field time-series image, a change diagram of adjacent time slices, and the reconstruction status. The storage unit is used to save the original image, the coded sequence, calibration data, iteration records, and the output sequence. When the display and storage unit does not participate in the core imaging relationship, it can be set as an external computing device. The image reconstruction unit 8 sends the results through a wired or wireless data interface. Regardless of whether the display and storage unit is located inside or outside the system, the original image and the coded sequence are associated through a common period identifier to prevent the incorrect combination of data from different periods during batch processing.

[0084] After the system starts up, it first performs a self-test. The synchronization control module checks the clock lock status, the reading status of the encoding sequence generation unit, the ready status of the drive unit, the camera trigger and data interface status, and the storage capacity of the image reconstruction unit 8. After the optical path unit 5 establishes the atomic two-dimensional conversion screen, it acquires a reference image. If the effective area, average response, or saturated pixel ratio does not meet the current configuration requirements, the system remains in calibration state and prompts for adjustment, without entering the formal encoding acquisition. After the self-test and reference check pass, the spatial mapping, dark field, flat field, illumination calibration, and atomic time response data corresponding to the current transmission or reflection path are loaded.

[0085] Pre-measurement calibration includes temporal calibration and spatial calibration. Temporal calibration determines the drive advance, camera trigger delay, and atomic response correction sequence by comparing the illumination output monitoring signal, atomic visible light response, and camera exposure valid signal with a known control edge. Spatial calibration determines the mapping between the position of the sample under test 2, the position of the atomic two-dimensional conversion screen, and the position of the camera pixels through calibration components. All calibration data are marked with configuration identifiers and valid status. When the illumination frequency, optical path mode, camera area, imaging magnification, or position of atomic gas cell 4 changes, the system marks the no longer matching calibration data as invalid to avoid using inconsistent calibration relationships.

[0086] During formal measurement, if the reference monitoring branch detects that the terahertz output of a certain chip is below the allowable range, the signal acquisition unit 7 still saves the current integrated image and monitoring data, but sets the chip to low reliability; if multiple consecutive chips fail to form a valid output, the current coding cycle lacks sufficient measurement contribution, the image reconstruction unit 8 stops reconstruction and requests re-acquisition; if only a single chip is abnormal and its actual monitoring value is available, the target coding weight is replaced with the actual monitoring value and reconstruction continues, while the abnormal chip position is recorded in the output metadata so that subsequent analysis can identify the coding integrity of the output.

[0087] If local saturation occurs in the camera image, first determine whether the saturated region overlaps with the target region based on the saturation mask. If the saturated region is outside the target region, it is excluded only from the effective region mask. If the saturated region covers the target structure, prioritize reducing the camera gain, shortening the allowable reference interval, or reducing the terahertz and laser output before re-acquiring. If the overall camera image is close to a dark field, check the atomic two-dimensional conversion screen establishment status, terahertz output monitoring value, and camera exposure coverage status. It is not possible to claim recovery of effective terahertz information simply by enlarging the dark image in the software. This processing ensures that the handling of anomalies corresponds to the actual data quality.

[0088] If numerical overflow, non-numerical results, or a continuous increase in the candidate sequence occur during the reconstruction process, image reconstruction unit 8 reverts to the candidate sequence with the most recent valid value and the smallest overall projection deviation, reduces the update step size, and re-executes the current round; if an abnormality still occurs after repeating the preset number of times, the iteration stops, and the abnormal status is output instead of the unlabeled image sequence; if the overall projection deviation is very small but high-frequency alternations that do not conform to the scene occur between time slices, the time change threshold and the actual coding weight are checked, and if necessary, the time change constraint is increased or a multi-coding period extension measurement is adopted to reduce the instability of single-image underdetermined reconstruction.

[0089] For the sample 2 to be tested, which moves with the conveying mechanism, the sample carrier can be a continuously running conveyor belt. The synchronous control module obtains the position trigger signal from the conveying controller and starts the encoding cycle when the target area enters the terahertz field of view. If the sample displacement within one encoding cycle is less than the allowable spatial registration error, it is directly reconstructed according to the fixed spatial mapping. If the displacement is not negligible and the conveying speed is known, the image reconstruction unit 8 determines the relative displacement according to the midpoint time of each chip and the conveying speed, moves the candidate time slice image in the opposite direction along the conveying direction by the corresponding distance, and then performs encoding projection, thereby compensating for the spatial misalignment caused by the sample movement in the forward model.

[0090] When calculating motion compensation using textual methods, the time difference between the midpoint of a given chip and the midpoint of the first chip is first obtained by subtracting the midpoint of the first chip from the midpoint of a given chip. Then, the transport speed is multiplied by the time difference to obtain the positional change of the sample relative to the first chip during the given chip period. Subsequently, based on the object surface sampling interval in the spatial mapping table, the positional change is converted into integer pixel movement and remaining distance. Pixel translation is performed on the integer part, and the remaining distance is interpolated based on the relative distance between adjacent pixels and the target position. If the positional change causes some pixels to move out of the effective field of view, the moved-out area is marked as invalid and is not cyclically filled from the other side of the field of view.

[0091] For periodically changing objects under test, the coding period can maintain a definite phase relationship with the periodic triggering of the object under test. In multiple repetitive periods, different coding sequences are used to acquire single coded integral images, and candidate time slices with the same phase in different periods are used as the same sequence to be recovered. If there are slight fluctuations in the period of the object under test, the actual periodic triggering time is recorded each time, and the chip time is mapped to a unified period coordinate according to the phase ratio. This implementation method uses a repeatable process to obtain multiple sets of coded measurements, which is suitable for scenarios that require improved reconstruction determinism and where the testing process can be repeated. Each measurement period still maintains atomic two-dimensional full-field transformation and single-integration acquisition by the camera.

[0092] For non-repeatable transient processes, a single integral acquisition within a single coding cycle is used, and spatiotemporal constraints are employed for reconstruction. The coding sequence, number of chips, and chip duration are loaded into the synchronization control module before the event occurs. Upon the arrival of the external event, coding output and camera exposure are initiated simultaneously. If there is an uncertain delay between the external event trigger and the first chip, a known preamble code can be set before the time-domain coding sequence, and the position of the preamble response is compared during reconstruction to estimate the actual time offset. Subsequently, the output time index is corrected as a whole. When the time offset cannot be estimated, the time coordinate relative to the start signal of the cycle is retained, and the absolute event time is not fabricated.

[0093] The specific collaborative execution method of the imaging system is as follows: In one system collaborative execution mode, the synchronization control module first sends the time-domain coding sequence, chip clock, and period start signal to the terahertz high-speed modulation illumination source 1, and sends the camera trigger signal corresponding to the period start signal to the visible light high-speed image acquisition module; the terahertz high-speed modulation illumination source 1 returns the actual output status of each chip, and the visible light high-speed image acquisition module returns the exposure valid signal and a single visible light coded integral image; the synchronization control module associates the returned data with the same period identifier and sends it to the image reconstruction unit 8; the image reconstruction unit 8 checks the exposure coverage and actual coding weight accordingly, and completes image preprocessing, coding projection, constraint iteration, and time-series image output, so that each module in the system can complete collaborative work through explicit data and control signals.

[0094] The image reconstruction unit 8 can be deployed in a computing device connected to the visible light high-speed image acquisition module. The computing device includes a processor, a memory, and a data interface. The memory stores the image preprocessing program, the encoding projection program, the constraint iteration program, the calibration data, and the output management program. The data interface receives a single visible light encoded integral image, a time-domain encoded sequence, a reliability map, an effective region mask, and synchronous recording. When the processor executes the program, it sequentially completes version verification, dark field and flat field correction, spatial registration, candidate sequence initialization, encoding projection, residual back projection, non-negativity constraint, spatial continuity constraint, temporal variation sparse constraint, stop judgment, and time-series image output. The computing device, as the image reconstruction unit 8 of the imaging system, maintains consistent periodic identification and data format with other modules.

[0095] In the absence of technical logic conflicts, the aforementioned transmission imaging method, reflection imaging method, binary encoding method, multi-level encoding method, reference monitoring method, atomic time response correction method, single encoding period constraint reconstruction method, multi-encoding period joint reconstruction method, fixed sample method, and motion compensation method can be combined with each other. When combined, the time-domain encoding sequence, spatial mapping, exposure window, atomic response correction, dark field and flat field data used for the same measurement data are all associated with a common period identifier and configuration version. If any preceding data is missing or the version is inconsistent, it will be handled according to the corresponding abnormal rules, so that each output time slice can be traced back along the data link of encoded illumination, sample action, atomic two-dimensional transformation, integral acquisition and constraint reconstruction.

[0096] When determining the chip duration, the inputs include the effective switching time of the terahertz high-speed modulation illumination source 1, the clock resolution capability of the synchronization control module, the calibrated time response of the atomic two-dimensional conversion screen, and the target time scale of the process to be observed. First, the lower limit of the chip duration is determined from the minimum stable control interval allowed by the hardware. Then, the expected chip duration is determined according to the adjacent state intervals that the target process needs to distinguish. If the expected value is less than the hardware lower limit, the hardware lower limit is used and the corresponding time resolution mark is used in the output. If the expected value is not less than the hardware lower limit, the expected value is used. This selection does not use the camera's frame-by-frame readout period as a direct limitation for each chip, but the total length of the exposure window should still be within the camera's allowable range.

[0097] When determining the number of chips, the input includes the observation time range to be covered and the already determined chip duration. If the duration of each chip is the same, the observation time range is divided by the chip duration. If there is a remaining time that is not divisible, one chip is added to ensure that all chips cover the required observation range. If adding chips results in the coding period exceeding the camera's allowed single exposure range, the chip duration can be increased, the observation range shortened, or the observation range can be divided into multiple coding periods. The specific choice should be made without disrupting the temporal relationship of the observed process. After the number of chips is determined, the candidate sequence layer number and the number of output time indices are updated synchronously.

[0098] When determining the exposure window, first obtain the start time of the encoding cycle, the total duration of valid chips, the drive setup advance, the camera trigger delay, and the atomic response tail time that needs to be retained. Move the start time of the encoding cycle forward by the time required for drive setup and camera trigger correction to obtain the exposure start reference. Extend the end time of the last valid chip backward by the response tail time to obtain the exposure end reference. The difference between the two is the target exposure duration. If the target exposure duration exceeds the camera's allowable upper limit, prioritize shortening the leading and trailing intervals that do not carry target information. If it still exceeds the upper limit, redetermine the number of chips or the chip duration. Do not forcibly meet the exposure upper limit by truncating the last valid chip.

[0099] The validity check of the encoded sequence includes energy check, discrimination check, and correlation check. The energy check sums the absolute values ​​of the actual coding weights of each chip. If the sum is zero or lower than the effective illumination threshold, the sequence is deemed invalid. The discrimination check counts the number of different coding levels and the frequency of each level. If only a single level appears and there is no other time constraint information, it is deemed unsuitable for separating time changes. When multiple coding periods are measured jointly, the correlation check compares the coding relationship of different sequences at the same chip position. If two sequences are the same or too close, one of the sequences is replaced to increase the ability of multiple measurements to distinguish different time slices.

[0100] The image size boundary is jointly determined by the effective area of ​​the camera, the effective area of ​​the spatial mapping, and the coverage area of ​​the atomic two-dimensional conversion screen. When the intersection of the three is empty, the measurement is stopped and the optical path is requested to be readjusted. When the intersection is smaller than the target area, the uncovered area is marked in the output and the imaging magnification is readjusted according to the task requirements. Each image of the candidate sequence required for reconstruction uses the same size and effective area mask. If a local area moves out of the field of view due to motion compensation in a certain time slice, only the corresponding position of that time slice is set to invalid, without changing the image size of other time slices, thereby ensuring the consistency of the time series data structure.

[0101] In the equivalent coding method where the coding weights contain negative values, when the actual illumination cannot output negative intensity, a negative weight can be split into two non-negative illumination contributions or a complementary sequence can be used in multiple coding cycles. The image reconstruction unit 8 forms an equivalent signed measurement by subtracting the two measured images. The main implementation method with a single coding cycle and a single integral image preferentially adopts non-negative coding weights so that the control state of each chip can be directly realized by the actual illumination intensity. When complementary measurements are used, the two images must be corrected for dark field and flat field respectively and spatially registered to avoid treating the overall drift between the two acquisitions as a signed coding response.

[0102] Processing rules are set for empty images, size mismatches, and missing sequence items. When the image data is empty or the number of effective pixels is zero, no candidate sequence is built and the acquisition invalid status is returned directly. When the image size does not match the current spatial mapping size, it is first checked whether the size can be restored through the cropping area of ​​the current record. If it cannot be restored, the reconstruction is stopped and the image is not automatically stretched. When the encoding sequence is missing a certain chip encoding value, if the reference monitoring data can provide the actual output of the chip, the monitoring value is used to fill in the missing value. Otherwise, the single-image reconstruction is stopped and the complete encoding data is requested to prevent the forward projection relationship from being changed by arbitrary default values.

[0103] When managing repetitive periodic data, the same period identifier can only be associated with one formal integral image. If a repetitive image with the same period identifier is received, the data integrity flag, synchronization status, and saturation pixel ratio are compared. The image that is complete and has fewer anomalies is retained, and the rest are considered repetitive data. Different period identifiers must not share unconfirmed coded sequence records. Even if the coded sequence names are the same, the version, actual monitoring weight, and period start time must be checked. This data consistency processing ensures that multiple measurements, re-triggering, and buffer retransmission will not cause mismatch between the coded sequence and the image.

[0104] The temporal coding sequence is combined with the terahertz high-speed modulation illumination source 1 to give the terahertz illumination during different code slices a known contribution weight. The rapid changes of the sample under test 2 within a coding cycle participate in the visible light coding integral image according to the corresponding code slice weight, avoiding the formation of cumulative results without time markers. This technical relationship provides known positive conditions for subsequent temporal dimension reconstruction. Its direct effect is to treat multiple time slices within an exposure window as objects to be recovered, avoiding treating the entire window as just an indivisible time point.

[0105] The terahertz imaging lens 3, the atomic two-dimensional conversion screen, and the visible light camera maintain the spatial correspondence between the object surface, the atomic interaction surface, and the photosensitive surface, so that an effective pixel of the integral image continuously receives the visible light contribution of the same corresponding position of the sample under test 2 during each chip. This spatial correspondence allows the encoding projection and residual back projection to be performed in parallel on a pixel-by-pixel basis, and enables the spatial continuity constraint to utilize the structural relationship of adjacent positions. The direct effect is to preserve the two-dimensional parallelism of terahertz full-field imaging and avoid scanning the target area point by point in order to obtain spatial information.

[0106] The synchronization control module establishes a recordable time relationship between the encoding period and the camera exposure window, thus enabling the image reconstruction unit 8 to determine which chips enter the current integration measurement and the effective weight of each chip. Without this synchronization relationship, even if the encoding value is known, the start and end positions of the encoding sequence corresponding to the integrated image cannot be reliably determined. In this embodiment, a complete verification relationship is formed through a common time reference, a period start signal, an effective exposure signal, and an abnormal coverage check. The direct effect is to reduce the impact of chip index misalignment and exposure truncation on time slice recovery.

[0107] The encoding projection relation transforms the candidate time series image into a single candidate integral image with the same acquisition form as the actual acquisition. Data consistency updates make the candidate integral image gradually approach the measured integral image. Non-negativity constraints eliminate candidate values ​​that do not conform to the physical meaning of intensity. Spatial continuity constraints utilize the two-dimensional structure of the same time slice. Temporal sparse constraints utilize the condition that fast events usually only change in part of the time and location. These constraints work together on the same candidate sequence, so that the output of a single underdetermined measurement is not determined by arbitrary allocation, but by the measured data and the published spatiotemporal rules, thus obtaining a repeatable terahertz full-field time series image sequence with a stopping state.

[0108] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A time-domain encoded high-speed atomic terahertz imaging method, characterized in that, include: A time-domain coded sequence containing multiple chips and their corresponding coded values ​​is obtained. Based on this time-domain coded sequence, a terahertz high-speed modulated illumination source is controlled to output terahertz waves of corresponding intensities according to the chip values, resulting in a coded terahertz pulse sequence, including: Acquire the number of chips, chip order, chip encoding value, and chip duration; check whether the chip encoding value is within the control range of the terahertz high-speed modulated illumination source. After the inspection is passed, the chip encoding values ​​are output sequentially according to the chip order based on the same clock signal, and the current chip encoding value is converted into the output control quantity of the terahertz high-speed modulation lighting source. The terahertz output intensity during the current chip period is adjusted according to the output control quantity, and the actual output intensity corresponding to each chip is recorded as the encoding weight. After collimating the coded terahertz pulse sequence, the sample to be tested is illuminated, so that the terahertz waves transmitted or reflected by the sample carry the two-dimensional spatial information of the sample changing over time, thus obtaining a terahertz image sequence. The terahertz image sequence is sequentially imaged onto a two-dimensional atomic conversion screen formed by the interaction between laser and atoms using a terahertz imaging lens. Based on the population change of atomic energy levels caused by the terahertz two-dimensional field distribution, the terahertz image sequence is converted into a two-dimensional visible light intensity sequence that maintains spatial correspondence. Based on the temporal coding sequence, an exposure window covering the entire coding period is determined. Within the exposure window, an integral acquisition of the visible light two-dimensional intensity sequence is performed to obtain a single visible light coded integral image, including: The coding period is determined based on the number of chips and the duration of each chip, and a period start signal and a camera trigger signal that are of the same origin as the time-domain coding sequence are generated. The first chip is output based on the cycle start signal, and the high-speed visible light camera is turned on for exposure based on the camera trigger signal. Acquire the valid exposure signal returned by the high-speed visible light camera, compare the valid exposure signal with the valid range of each chip, and retain the single visible light coded integral image when the valid exposure signal covers the valid range of all chips; otherwise, re-acquire or determine a new coding period according to the actual continuous chips covered. Based on the time-domain coded sequence and a single visible light coded integral image, a coding projection relationship is established between the candidate terahertz full-field time-series image sequence and the single visible light coded integral image, calculated by weighting and superimposing the coded values ​​of each chip. Under the constraints of non-negativity, spatial continuity, and sparse changes between adjacent time slices, the candidate terahertz full-field time-series image sequence is iteratively adjusted until the deviation meets the stopping condition. The output is a terahertz full-field time-series image sequence arranged in chip time order, including: Each candidate image is weighted pixel by pixel according to the coding weight of each chip and then accumulated to obtain the candidate coded projection image; The projection residual is obtained by comparing the candidate coded projection image with the single visible light coded integral image. The projection residual is multiplied by the coding weight of each chip to obtain the adjustment amount of each candidate image, and each candidate image is updated in the direction that reduces the projection residual. The negative values ​​in the updated candidate image are set to zero. Spatial continuity processing is performed on adjacent pixels in the same candidate image that do not meet the edge preservation condition. The pixel differences between adjacent candidate images that do not meet the time variation condition are compressed. Recalculate the deviation until the deviation reaches the preset tolerance, no longer decreases after multiple consecutive rounds, or the number of iterations reaches the upper limit; Obtain the discrete time response sequence of an atomic two-dimensional conversion screen measured by a terahertz test sequence with known time edges; For each output chip, the time interval between the current output chip and each of the preceding input chips is determined. Based on the time interval, the time response discrete sequence is queried to obtain the response ratio of each of the preceding input chips to the current output chip. The coding weights of each of the previous input chips are multiplied by the corresponding response ratios, and the resulting multiplications are accumulated to obtain the effective coding weights of the current output chip. The coding projection relationship is established based on the effective coding weights of each output chip to correct the temporal mixing of atomic transformation responses across adjacent chips.

2. The method according to claim 1, characterized in that, The sample to be tested is irradiated after being collimated with an encoded terahertz pulse sequence, including: The transmission imaging path or reflection imaging path is determined according to the detection method of the sample to be tested; when determining the transmission imaging path, the coded terahertz pulse sequence passing through the sample to be tested is guided to the terahertz imaging lens. When determining the reflection imaging path, the coded terahertz pulse sequence reflected from the surface of the sample under test is guided to the terahertz imaging lens; based on the calibration data corresponding to the determined imaging path, the spatial mapping relationship between the target area of ​​the sample under test and the effective imaging area of ​​the atomic two-dimensional conversion screen is determined.

3. The method according to claim 2, characterized in that, Atomic two-dimensional conversion screens are formed in the following way: In an atomic gas chamber containing alkali metal atomic vapor, the probe light, coupling light, and Rydberg excitation light are spatially filtered, expanded, and shaped by a light sheet, so that the probe light, coupling light, and Rydberg excitation light converge in the atomic gas chamber to form a two-dimensional laser action region. Adjusting the frequency and propagation position of each laser allows atoms within the two-dimensional laser action region to form an atomic energy level population state that can be affected by a terahertz field. The two-dimensional laser action area is defined as an atomic two-dimensional conversion screen, and it is made to cover the terahertz imaging area formed by the terahertz imaging lens.

4. The method according to claim 3, characterized in that, The terahertz image sequence is sequentially imaged onto an atomic two-dimensional conversion screen using a terahertz imaging lens, including: The imaging magnification is determined based on the target area size of the sample to be tested and the effective area size of the atomic two-dimensional conversion screen. The imaging magnification is adjusted to ensure that the plane of the sample under test and the image plane of the atomic two-dimensional conversion screen form a conjugate relationship. The correspondence between the object plane coordinates of the sample under test, the position of the atomic two-dimensional conversion screen, and the camera pixel coordinates is determined by using terahertz calibration components, and a spatial mapping table is generated. Based on the spatial mapping table, the same image positions corresponding to different chips are kept pointing to the same spatial position on the sample under test.

5. The method according to claim 4, characterized in that, Before establishing the encoding projection relationship, the following is also included: Acquire dark-field images with terahertz illumination off, flat-field images without samples and with terahertz illumination in the baseline state, and spatial mapping tables; Subtract the corresponding pixel values ​​from the dark field image from each pixel value in a single visible light coded integral image to obtain the pixel values ​​after removing the background; The pixel values ​​after removing the background are normalized based on the corresponding pixel differences between the flat field image and the dark field image. The normalized image is cropped and spatially registered according to the spatial mapping table, and saturated positions, non-response positions and out-of-view positions are set as invalid positions.

6. A time-domain encoded high-speed atomic terahertz imaging system, wherein the system implements the method as described in any one of claims 1 to 5, characterized in that, It includes a terahertz high-speed modulation illumination source, a sample-carrying optical path, an atomic terahertz-visible light conversion module, a visible light high-speed image acquisition module, a synchronization control module, and an image reconstruction unit; The synchronization control module is used to acquire a time-domain encoded sequence containing multiple chips and corresponding encoded values, and to control the terahertz high-speed modulation illumination source to output encoded terahertz pulse sequences according to the chips; the sample-carrying optical path is used to enable the encoded terahertz pulse sequence to be transmitted or reflected by the sample under test to form a terahertz image sequence. The atomic terahertz-visible light conversion module is used to image terahertz image sequences onto an atomic two-dimensional conversion screen and convert them into visible light two-dimensional intensity sequences that maintain spatial correspondence. The synchronization control module is also used to control the visible light high-speed image acquisition module to perform one integration acquisition within the exposure window covering the complete coding cycle, so as to obtain a single visible light coded integration image. The image reconstruction unit is used to establish a coding projection relationship based on the temporal coding sequence and a single visible light coding integral image, and iteratively outputs a terahertz full-field temporal image sequence under the constraints of non-negativity, spatial continuity and sparse changes between adjacent time slices.

Images