FPGA-based time encoding compressed sensing imaging method and device and electronic equipment

Abstract

The application provides a time coding compressed sensing imaging method and device based on FPGA and electronic equipment, the method is applied to the technical field of photoelectric detection and computational imaging, and the method comprises the steps of: calibrating an FPGA delay unit based on a reference clock, constructing a delay lookup table and writing a time coding sequence of random coding and a push-scan mode. In response to a synchronization signal, coarse and fine delay tap values are analyzed, a reference pulse is generated and is synchronized to the reference clock, a delay pulse is generated through fine delay, three alignment timing control signals are generated, and the excitation gate and exposure signals are cooperatively controlled. Discrete integral observation values are collected in an exposure window, a continuous observation operator is constructed in combination with the delay lookup table, a target function is constructed based on physical priori, and the target function is iteratively solved to obtain a high signal-to-noise ratio, a transient response curve and physical parameters. The method can solve the problems of slow sampling speed and lack of software flexibility due to dependence on modulation devices in traditional hyperspectral imaging technology.

Description

Technical Field

[0001] This application relates to the field of photoelectric detection and computational imaging technology, and more specifically, to a time-coded compressed sensing imaging method, apparatus and electronic device based on FPGA. Background Technology

[0002] For the precise capture of transient optical phenomena at the nanosecond or even picosecond level (such as nanosecond-level fluorescence lifetime decay, transient photochemical reactions, or time-of-flight depth information), traditional detection methods mainly rely on time-correlated single-photon counting (TCSPC) technology or streak cameras. However, while TCSPC technology can provide extremely high temporal resolution, it is limited by the photon accumulation effect, resulting in extremely slow scanning speeds, making it difficult to meet real-time imaging requirements. Streak cameras offer high precision and single-shot capability, but the equipment is expensive and has a limited field of view, hindering large-scale adoption. In recent years, compressed sensing (CS) technology has provided a new approach to overcoming sampling rate limitations; by introducing modulation in different dimensions such as the spatial spectrum, signals can be recovered using sampling rates far below the Nyquist frequency. Currently, some CS-based acquisition and imaging systems can achieve high-resolution, high-fidelity, and high-reliability detection, such as achieving efficient and short-time hyperspectral information acquisition through encoding with dispersive elements and spatial light modulators followed by computer demodulation.

[0003] In related technologies, CN118190162A discloses a method for efficient spectral image restoration using pseudo-random spectral coding and residual-guided reconstruction algorithms; CN119545198A discloses a video-level spectral imaging system and method based on compressed sensing; and CN115372991B discloses a spectral super-resolution single-pixel imaging method based on compressed sensing, which compensates for spatial resolution with spectral information and combines wavelet fusion algorithms to achieve super-resolution reconstruction within a single-pixel imaging framework. Currently, CS schemes mostly focus on spatial and spectral dimension coding, while temporal compressed sensing research focuses on low frame rate sampling of video streams to remove motion blur, aiming to recover the motion trajectory of macroscopic objects. CN118828194A discloses a real-time imaging and image reconstruction system and method based on compressed sensing, which modulates light intensity signals temporally through temporal dimension-coded exposure and Fourier sparse domain reconstruction, enabling low frame rate cameras to capture high-speed periodic motion, while also addressing microscopic ultrafast transient physical processes. In other words, existing time-domain compressed sensing methods mostly focus on modulating the intensity or phase of the excitation source, but lack coding design based on picosecond-level electrical time gating, and have not yet formed a systematic implementation scheme that directly uses programmable logic devices as time compression dimension generation units. Summary of the Invention

[0004] This application provides a time-coded compressed sensing imaging method, device, and electronic device based on FPGA. This method can solve the problems of slow sampling speed, low signal-to-noise ratio, lack of software flexibility due to reliance on modulation devices, and insufficient resolution of existing time-coding schemes in traditional hyperspectral imaging technology.

[0005] Firstly, a time-coded compressed sensing imaging method based on FPGA is provided, the method comprising:

[0006] The delay cell in the FPGA is calibrated according to the reference clock, the tap step value of the delay cell is obtained and a delay lookup table is constructed.

[0007] A time-coded sequence is generated based on a scanning strategy, and the time-coded sequence is written into the delayed lookup table. The scanning strategy includes a random coding mode and a push-broom mode.

[0008] In response to the synchronization signal, the delay configuration word under the current index is read based on the delay lookup table and parsed into coarse delay tap value and fine delay tap value;

[0009] A reference pulse is generated based on the coarse delay tap value, and the reference pulse and the fine delay tap value are synchronized to the reference clock. Within the reference clock, the reference pulse is finely delayed using the fine delay tap value to generate a delayed pulse.

[0010] Based on the delayed pulse, three aligned timing control signals are generated. These timing control signals include a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time.

[0011] During the validity period of the exposure window signal, the non-uniform comb gating signal is used to control the gated image sensor to acquire discrete integral observations.

[0012] A continuous observation operator is constructed based on discrete integral observations collected from multiple sets of different time-coded sequences and the delay lookup table. An objective function is generated based on the observation operator and a regularization term, which is a physical prior based on an exponential decay model.

[0013] The objective function is solved iteratively using the plug-and-play alternating direction multiplier method, and the transient response curve with high signal-to-noise ratio and no grid constraint and the related physical parameters are obtained by inversion. These physical parameters include the fluorescence lifetime of the luminescent material and the carrier recombination lifetime.

[0014] The above method enables the calibration of FPGA delay units and the construction of delay lookup tables based on a reference clock. This allows for precise locking of tap step values, compensation for process, voltage, and temperature drift, and quantification and reproducibility of delay quantities. This provides a stable hardware foundation for picosecond-level timing control. Because this solution employs both random and pushbroom dual-mode time encoding, it can simultaneously handle compressed sensing incoherent sampling and high-precision timing scanning, adapting to mixed-signal acquisition and fine waveform reconstruction, thus improving system versatility and sampling efficiency. By using coarse and fine delay tap values, it achieves picosecond-level fine-tuning of timing while maintaining a large nanosecond-level delay range, breaking through the accuracy limit of traditional clock counting and significantly improving the system's time resolution. In this scenario, the signal is synchronized to the reference clock domain, generating three aligned timing signals to eliminate cross-clock domain metastability and path deviation. This enables precise timing coordination between the excitation source, gated image intensifier, and camera, avoiding signal loss and sampling distortion. Utilizing non-uniform comb gating and exposure window integration, weak transient fluorescence signals can be accumulated multiple times, significantly improving the signal-to-noise ratio of observations and solving the problem of effectively detecting weak microbial light signals. A continuous observation operator is constructed based on physical timestamps to eliminate mismatch errors between the ideal sampling model and hardware behavior, providing a high-fidelity forward mapping relationship for inversion. Finally, by using an exponential decay physical prior and employing PnP-ADMM iterative solution, the solution space is constrained, noise is suppressed, and non-physical solutions are avoided. This overcomes the limitations of grid quantization, yielding a high signal-to-noise ratio continuous transient curve, accurately inverting microbial fluorescence lifetime, and improving parameter reliability and measurement accuracy.

[0015] In conjunction with the first aspect, in some possible implementations, the generation of time-encoded sequences based on the barcode scanning strategy includes:

[0016] For the random coding mode, a random sequence that meets the noncoherence requirement of compressed sensing is generated. This random sequence is used to achieve mixed sampling of strong and weak signals.

[0017] For push-broom mode, a linearly increasing delay sequence is generated, which is used to uniformly step scan on the time axis to reconstruct a high-precision waveform profile.

[0018] In conjunction with the first aspect, in some possible implementations, in response to a synchronization signal, the delay configuration word under the current index is read from the delay lookup table and parsed into a coarse delay tap value and a fine delay tap value, including:

[0019] When a rising edge of the synchronization signal is detected, the read address pointer of the dual ports is driven.

[0020] Read the coarse delay tap value and the fine delay tap value under the current index;

[0021] Store the coarse delay tap value and the fine delay tap value into the register.

[0022] In conjunction with the first aspect, in some possible implementations, the fine delay tap value is used to finely delay the reference pulse to generate a delayed pulse, including:

[0023] Load the fine delay tap value;

[0024] The reference pulse is superimposed with a phase shift to generate a delayed pulse.

[0025] In conjunction with the first aspect, in some possible implementations, the generation of three aligned timing control signals based on the delayed pulse includes:

[0026] A non-uniform comb gating signal is generated using pulse width shaping logic;

[0027] The coarse delay value is loaded based on the system clock-driven counter, and a target synchronization signal and an exposure window signal are generated when the coarse delay value returns to zero.

[0028] The exposure window signal, the non-uniform comb gating signal, and the target synchronization signal are synchronously calibrated and aligned.

[0029] In conjunction with the first aspect, in some possible implementations, the construction of a continuous observation operator based on discrete integral observations acquired from multiple sets of different time-coded sequences and the delay lookup table includes:

[0030] Retrieve the physical timestamp and time offset information from the delay lookup table;

[0031] Based on this physical timestamp, a precise integral mapping relationship is established between continuous physical time signals and discrete integral observations;

[0032] A continuous observation operator is constructed based on this mapping relationship.

[0033] In conjunction with the first aspect, in some possible implementations, the generation of the objective function based on the observation operator and the regularization term includes:

[0034] The signal is reconstructed based on prior constraints constructed using a parameterized exponential function;

[0035] Based on the reconstructed signal according to the prior constraints, as well as the observation operator, observation time, observed pixel value, and regularization term, an objective function is generated. The formula for the objective function is: ,in, For the observation operator, The transient signal sequence to be reconstructed This is the vector of measured values ​​after compressed sampling. For regularization parameters, To introduce a physical prior regularization term.

[0036] In conjunction with the first aspect, in some possible implementations, the objective function is iteratively solved using the plug-and-play alternating direction multiplier method to invert and obtain a high signal-to-noise ratio, mesh-free transient response curve and related physical parameters, including:

[0037] The solution is obtained by alternating iterative steps using the plug-and-play alternating direction multiplier method.

[0038] The least squares method or gradient descent method is used to ensure that the reconstruction results are consistent with the actual observation data;

[0039] Physical model-based proximal operators or exponential fitting operators project intermediate variables onto the physical constraint manifold, update and output physical parameters.

[0040] Secondly, a time-coded compressed sensing imaging device based on FPGA is provided, the device comprising:

[0041] The module is used to calibrate the delay cell in the FPGA according to the reference clock, obtain the tap step value of the delay cell, and build a delay lookup table.

[0042] The generation and writing module is used to generate a time-encoded sequence based on the scanning strategy and write the time-encoded sequence into the delay lookup table. The scanning strategy includes a random encoding mode and a push-broom mode.

[0043] The read and parse module is used to respond to the synchronization signal, read the delay configuration word under the current index based on the delay lookup table, and parse it into coarse delay tap value and fine delay tap value;

[0044] The generation and synchronization module is used to generate a reference pulse based on the coarse delay tap value, and synchronize the reference pulse and the fine delay tap value to the reference clock. Within the reference clock, the fine delay tap value is used to finely delay the reference pulse to generate a delayed pulse.

[0045] The generation module is used to generate three aligned timing control signals based on the delayed pulse. The timing control signals include a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time.

[0046] The control and acquisition module is used to control the gated image sensor and acquire discrete integral observations within the validity period of the exposure window signal using the non-uniform comb gating signal.

[0047] The acquisition and construction module is used to construct a continuous observation operator based on discrete integral observations acquired from multiple sets of different time-coded sequences and the delay lookup table. Based on the observation operator and the regularization term, an objective function is generated. The regularization term is a physical prior based on the exponential decay model.

[0048] The iterative solution module is used to iteratively solve the objective function based on the plug-and-play alternating direction multiplier method, and invert to obtain a high signal-to-noise ratio, grid-free transient response curve and related physical parameters, including the fluorescence lifetime of the luminescent material and the carrier recombination lifetime.

[0049] Thirdly, an electronic device is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, causing the electronic device to perform the method described above in the FPGA-based time-coded compressed sensing imaging method.

[0050] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to execute the method described above in the FPGA-based time-coded compressed sensing imaging method.

[0051] Fifthly, a computer-readable storage medium is provided that stores computer program code, which, when run on a computer, causes the computer to execute the method described above in the FPGA-based time-coded compressed sensing imaging method. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the implementation environment of a time-coded compressed sensing imaging method based on FPGA provided in an embodiment of this application;

[0053] Figure 2 This is a schematic flowchart of a time-coded compressed sensing imaging method based on FPGA provided in an embodiment of this application;

[0054] Figure 3 This is a timing diagram illustrating three timing signals provided in an embodiment of this application;

[0055] Figure 4 This application provides an embodiment of a time-coded compressed sensing imaging system based on FPGA.

[0056] Figure 5 This is a schematic diagram of the structure of dynamic reconfiguration of an encoding strategy provided in an embodiment of this application;

[0057] Figure 6This is a schematic diagram of the structure of a time-coded compressed sensing imaging device based on FPGA provided in an embodiment of this application;

[0058] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0060] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0061] In the following description of the embodiments of this application, it is used as... Figure 1 Taking an example, the implementation environment of the embodiments of this application will be introduced.

[0062] For example, such as Figure 1 As shown, the implementation environment includes a camera 110 and a field-programmable gate array (FPGA) 120.

[0063] Camera 110 is a time-coded integrator. Camera 110 is used to record the mixed light intensity modulated by a high-speed time gate during the exposure time. Camera 110 can generate an integral result after weighted sampling of transient signals. That is, the single-frame output of camera 110 is a compressed numerical value containing time-coded information, rather than a direct image.

[0064] FPGA120 is a product of further development based on devices such as Programmable Array Logic (PAL), General Purpose Array Logic (GAL), and Erasable Programmable Logic Device (EPLD). An FPGA is a programmable logic chip capable of performing general-purpose functions; that is, it can be programmed to implement specific logic processing functions. In some embodiments, the FPGA is used to control the camera 110 to perform an image detection method.

[0065] To address at least one of the aforementioned technical problems, embodiments of this application provide a time-coded compressed sensing imaging method based on FPGA. This method can solve the problems of slow sampling speed, low signal-to-noise ratio, reliance on modulation devices and lack of software flexibility, and insufficient resolution of existing time-coding schemes in traditional hyperspectral imaging techniques.

[0066] Figure 2 This is a schematic flowchart of a time-coded compressed sensing imaging method based on FPGA provided in an embodiment of this application.

[0067] For example, such as Figure 2 As shown, taking an FPGA as the execution subject as an example, this application describes a time-coded compressed sensing imaging method based on an FPGA. The method 200 includes the following steps.

[0068] Step 201: Calibrate the delay cell in the FPGA according to the reference clock, obtain the tap step value of the delay cell, and construct a delay lookup table.

[0069] It should be understood that the accuracy of time-coded signals generated by conventional digital circuits is limited by the system clock cycle. In this case, in order to optimize the accuracy of the time-coded signal, a delay calibration module is instantiated in the Programmable Logic (PL) unit, and a high-frequency reference clock is introduced as a reference.

[0070] The reference clock is 200 MHz, and the delay calibration module is used to monitor and compensate for the chip's process, voltage, and temperature (PVT) drift in real time.

[0071] In some embodiments, IDELAYCTRL is instantiated within PL, the step value of the input delay unit (IDELAY) is locked, and a highly linear time base is established.

[0072] IDELAYCTRL is used to monitor and compensate for this drift in real time using a stable reference clock, ensuring that the time represented by each delay step is accurate, stable, and linear.

[0073] In some embodiments, a dual-port memory is instantiated within the PL, with port A connected to the PS bus and port B connected to the PL time encoding logic, to construct a shared hardware and software delay lookup table.

[0074] Step 202: Generate a time-coded sequence based on the scanning strategy and write the time-coded sequence into the delay lookup table. The scanning strategy includes a random coding mode and a push-scan mode.

[0075] It should be understood that in practical applications, corresponding time-coded instruction sets are generated according to different imaging requirements and loaded into the FPGA's delay lookup table, thereby controlling the hardware to generate the corresponding precise timing. For example, the scanning strategy corresponding to high signal-to-noise ratio imaging requirements is different from the scanning strategy corresponding to high-precision waveforms.

[0076] Among them, the pushbroom model is used to generate random sequences that satisfy the incoherence of compressed sensing, and the pushbroom model is used to generate linearly increasing delay sequences.

[0077] In one possible implementation, for the random coding mode, a random sequence that satisfies the incoherence requirement of compressed sensing is generated, which is used to achieve mixed sampling of strong and weak signals; for the pushbroom mode, a linearly increasing delay sequence is generated, which is used to uniformly scan the time axis to reconstruct a high-precision waveform profile.

[0078] In some embodiments, random sampling points are randomly distributed along the time axis, potentially falling in regions of strong signal (such as the early stages of fluorescence decay) or regions of weak signal (the tail end of decay). During a single long exposure integration, the strong signal portion contributes a large number of photons, increasing the overall signal-to-noise ratio, while the weak signal portion is also captured.

[0079] In some embodiments, pushbroom scanning is used when directly observing the precise decay curve of fluorescence lifetime, in calibration systems, or in scenarios where high fidelity is required for signal morphology. Outside of high-fidelity scenarios, it provides a true reference for reconstructing results from random patterns.

[0080] In this implementation, the random coding mode can obtain sufficient reconstructed signal information with fewer samplings, greatly improving acquisition efficiency; the push-broom mode can provide the most intuitive and distortion-free signal profile.

[0081] In some embodiments, the random sequence and the delayed sequence are written into a delay lookup table inside the FPGA.

[0082] Step 203: In response to the synchronization signal, read the delay configuration word under the current index based on the delay lookup table, and parse it into coarse delay tap value and fine delay tap value.

[0083] The synchronization signal serves as the unified time starting point for the system. Using this signal as a trigger condition to read the configuration word ensures that each sampling starts from a preset time node, avoiding timing chaos caused by random reading.

[0084] In one possible implementation, when a rising edge of the synchronization signal is detected, the read address pointer of the dual port is driven; the coarse delay tap value and the fine delay tap value under the current index are read; and the coarse delay tap value and the fine delay tap value are stored in the register.

[0085] The current index is the address identifier of the delay configuration word in the dual-port memory. The coarse delay tap value corresponds to the quantization value of a large-range delay at the nanosecond level, while the fine delay tap value corresponds to the quantization value of high-precision fine-tuning at the picosecond level.

[0086] It should be understood that the accuracy of using only coarse delay tap values ​​is limited by the system clock cycle and cannot achieve picosecond-level control, while using only fine delay tap values ​​cannot cover a wide range of delays at the nanosecond level.

[0087] In this implementation, if nanosecond-level delay is achieved using pure fine delay, thousands of taps are required, which would consume a large amount of FPGA logic resources; after splitting, only a small number of taps are needed for fine-tuning, which saves resources significantly.

[0088] Step 204: Generate a reference pulse based on the coarse delay tap value, and synchronize the reference pulse and the fine delay tap value to the reference clock. Within the reference clock, use the fine delay tap value to finely delay the reference pulse to generate a delayed pulse.

[0089] It should be understood that relying solely on coarse delay tap values, while offering a sufficiently large delay range, lacks sufficient precision and cannot meet the requirements of picosecond-level transient sampling; conversely, relying solely on fine delay tap values, while offering sufficient precision, suffers from an extremely small delay range and cannot cover nanosecond-level transient processes. In practical applications, the input signal must operate within the reference clock domain. Therefore, the input signal needs to be a signal that originates from and is in phase with the reference clock. After IDELAY, the delay tap value and the actual delay time become linear, predictable, and calibrable. In other words, synchronization with the reference clock is a prerequisite for the fine delay unit to operate legally, normally, and accurately.

[0090] The reference pulse is a standard time pulse signal generated in the system clock domain based on the coarse delay tap value.

[0091] In one possible implementation, the fine delay tap value is loaded; the reference pulse is superimposed with a phase shift to generate a delayed pulse.

[0092] It should be understood that the resolved fine delay tap values ​​are configured to the programmable fine delay units inside the FPGA. Superimposed phase shift means that the waveform shape and pulse width of the entire pulse remain unchanged, but the whole pulse is shifted backward on the time axis.

[0093] In some embodiments, a signal is generated using a cascaded architecture of coarse and fine delay tap values. A multi-stage beat synchronizer is used to synchronously transmit this pulse and the fine delay tap value to the reference clock domain to eliminate cross-clock domain metastability. Within the reference clock domain, a programmable input delay unit primitive is used to dynamically load the fine delay tap value, applying a corresponding picosecond-level fine delay to the reference pulse.

[0094] In this implementation, a large-range delay basis for the reference pulse is determined by coarse delay tap values, and then the reference pulse is phase-shifted by picosecond-level superposition using fine delay tap values. This breaks through the upper limit of delay accuracy generated by traditional digital circuits that rely solely on clock counting, that is, it improves the timing control accuracy to the picosecond level, meeting the sampling requirements of ultrafast physical processes such as transient optical signals and fluorescence lifetime.

[0095] Step 205: Based on the delayed pulse, generate three aligned timing control signals. The timing control signals include a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time.

[0096] It should be understood that microbial fluorescence is a transient light signal, and the duration of a transient light signal is only on the order of nanoseconds to picoseconds. Moreover, the time window for the appearance of a transient light signal is extremely narrow, the intensity is extremely weak, and it only exists for a very short instant after the microbial fluorescence is excited. Therefore, in order to capture transient light signals completely, accurately, and without distortion, it is necessary to ensure that the excitation action, sampling action, and recording action are aligned on the time axis, and that the timing control signals of the three aligned actions are without misalignment or offset.

[0097] Among them, the target synchronization signal is used to control the time for the sample to generate transient light, the non-uniform comb gating signal is used to control the time for opening the extremely narrow time window, and the exposure window signal controls the time period for accumulating the weak photocharge from multiple samplings.

[0098] In one possible implementation, a non-uniform comb gating signal is generated by pulse width shaping logic; a coarse delay value is loaded based on a system clock-driven counter; a target synchronization signal and an exposure window signal are generated when the coarse delay value returns to zero; and the exposure window signal, the non-uniform comb gating signal, and the target synchronization signal are synchronously calibrated and aligned.

[0099] The coarse delay value is a delay count in units of a reference clock cycle. The coarse delay value represents a delay of an integer number of clock cycles.

[0100] In some embodiments, the timing control signal is output through the FPGA's I / O pins.

[0101] It should be understood that gated image intensifiers have strict constraints on the pulse width, edge rate, and effective level of the control signal. The original delayed pulse pulse width and amplitude may not conform to the device driving specifications. Therefore, it is necessary to convert it into a standard driving pulse through pulse width shaping. By shaping, a non-uniform comb gating signal is generated, and the gating window can be arranged randomly or according to the coding rules on the time axis to meet the requirements of compressed sensing for the incoherence of the measurement matrix and realize the simultaneous sampling of strong and weak transient signals. Only by generating an ultra-narrow gating pulse through pulse width shaping can the transient component of the target time be accurately intercepted in the massive background light and the noise and stray light in non-target time periods be suppressed.

[0102] In this implementation, a non-uniform comb gating signal is generated through pulse width shaping logic, which can be adapted to the gated image intensifier driving specifications, realize non-uniform time-coded sampling, meet the non-coherence requirements of compressed sensing, and improve the sampling fidelity and signal-to-noise ratio of weak transient signals. The target synchronization signal and exposure window signal are generated based on the system clock and coarse delay values, which can realize nanosecond-level delay control, accurately locate the excitation time, and provide a continuous integration interval for the camera to realize the accumulation of multiple excitation signals. The unified timing calibration and alignment of the three signals can eliminate the fixed delay deviation and random jitter caused by the internal logic of the FPGA, and ensure that the sampling window accurately covers the effective time period of the transient light signal.

[0103] It should be understood that the generated timing control signal is sent to an external photoelectric detection system, where the target synchronization signal can trigger a pulsed power supply or pulsed laser to excite the sample under test, enabling repeatable transient observation. For example, the sample under test may be microbial fluorescence. A non-uniform comb gating signal is used to control the detector's integral gating window. The system employs a mixed sampling strategy for strong and weak signals: under a constant gain configuration, the incoherence of random coded sequences is used to perform mixed sampling of signal strength and weakness regions; without gain doubling, a sparse reconstruction algorithm can maintain a high signal-to-noise ratio over a wide dynamic range, achieving synchronous high-fidelity recovery of strong and weak signals.

[0104] Step 206: During the validity period of the exposure window signal, the gated image sensor is controlled by the non-uniform comb gating signal to acquire discrete integral observations.

[0105] It should be understood that, according to the preset gain strategy configuration, data from random encoding or time-series scanning performed on the hardware side is collected to obtain discrete integral observations containing time-encoded information.

[0106] like Figure 3 As shown, Figure 3 This illustration shows a timing diagram representing three timing signals according to an embodiment of this application. Figure 3For example, T0 represents the time reference point aligned with the synchronization signal. T0 represents the start time of each observation cycle. For both the synchronization signal and the exposure signal, they strictly begin at time T0 each time. dt=1 and dt=2 represent fixed step delays relative to T0. This is a characteristic of pushbroom mode, where the delay increases linearly each time (e.g., 1 unit time, 2 units time). dt=random represents a random delay relative to T0, which is a characteristic of random coding mode, where the delay is random and non-uniform each time. In this case, in pushbroom mode, a high-precision and intuitive waveform can be obtained through fixed step delays, while in random mode, high signal-to-noise ratio and high compression ratio coded data can be obtained through random delays. In some embodiments, during the effective duration of the exposure window signal, a non-uniform comb gating signal is used as a gating trigger command to control the gated image sensor to open the photosensitive channel within a preset discrete time window, accumulate and sense the transient light signals generated by multiple excitations, and output a single pixel integral value, i.e., a discrete integral observation value.

[0107] Step 207: Construct a continuous observation operator based on discrete integral observations collected from multiple sets of different time-coded sequences and the delay lookup table; generate an objective function based on the observation operator and a regularization term, where the regularization term is a physical prior based on an exponential decay model.

[0108] It should be understood that, based on discrete integral observations corresponding to multiple sets of different time-coded sequences, and combined with the physical timestamps and time offset information stored in the hardware delay lookup table, a mathematical mapping relationship between continuous physical time-domain transient signals and discrete integral observations is established, thereby constructing a continuous observation operator; and a physical prior based on an exponential decay model is introduced as a regularization term to constrain the solution space of the objective function, so that the objective function matches the observation data acquired by the hardware.

[0109] In one possible implementation, the physical timestamp and time offset information are obtained from the delay lookup table; an accurate integral mapping relationship between the continuous physical time signal and the discrete integral observation is established based on the physical timestamp; and a continuous observation operator is constructed based on the mapping relationship.

[0110] In some embodiments, a non-convex optimization objective function is defined; for transient physical processes such as fluorescence lifetime decay and time of flight, an exponential prior is introduced as a regularization term to replace the general total variation or L1 sparse regularization term.

[0111] In this implementation, a continuous observation operator is constructed based on the physical timestamps of the delay lookup table, which can accurately map the physical process from continuous transient signals to discrete integral observations. The introduction of an exponentially decaying physical prior regularization term to generate the objective function can constrain the solution space, suppress noise, ensure that the reconstruction results fit the hardware data and conform to physical laws, and improve the accuracy of parameter inversion and the robustness of the algorithm.

[0112] Step 208: The objective function is iteratively solved based on the plug-and-play alternating direction multiplier method to obtain a high signal-to-noise ratio, grid-free transient response curve and related physical parameters, including the fluorescence lifetime of the microorganism.

[0113] It should be understood that the Plug and Play Alternating Directional Multiplier Method (PnP-ADMM) is used to iteratively optimize and solve the objective function that integrates hardware timing information and physical priors. This breaks through the quantization limitations of traditional grid sampling and inverts the transient response curve with high signal-to-noise ratio and continuous time domain. Based on this curve, core physical parameters such as microbial fluorescence lifetime are extracted. In particular, by constructing signal prior constraints through parameterized exponential functions, the physical law of exponential decay of fluorescence signals is incorporated into the objective function, ensuring that the iterative solution process always adheres to the physical essence and improving the accuracy of parameter inversion.

[0114] In one possible implementation, a priori constraint reconstruction signal is constructed based on a parameterized exponential function; an objective function is generated based on this prior constraint reconstruction signal, the observation operator, the observation time, the observed pixel value, and the regularization term. The formula for the objective function is: .

[0115] It should be understood that the dimension of discrete integral observations is much lower than that of continuous transient signals, and the traditional least squares method cannot obtain a unique solution. ADMM decomposes the complex optimization problem into subproblems of data consistency and physical denoising by splitting the objective function. It can derive a high signal-to-noise ratio, grid-free, high-time-resolution transient response curve from the undersampled integral measurements.

[0116] in, For the observation operator, The transient signal sequence to be reconstructed This is the vector of measured values ​​after compressed sampling. For regularization parameters, To introduce a regularization term based on physical priors. That is, This is a high-resolution light intensity variation curve. This represents the low-resolution integral value actually captured by the camera. Used to balance the weights of data fidelity terms and regularization terms, if If the value is relatively large, then the regularization term in the objective function will be larger. The proportion is relatively high. Used to suppress noise and constrain the solution space during iterative solution, for example, It can be a sparsity constraint or a total variation constraint.

[0117] Understandably, this prior constraint forces the reconstructed signal to lie on a low-dimensional manifold composed of a family of parameterized exponential functions, thereby effectively distinguishing physical signals from random noise and suppressing artifacts under low sampling rate conditions.

[0118] The expression for the parameterized exponential function is: This exponential function is used to calculate fluorescence lifetime. Specifically, the change in fluorescence intensity over time can be described using the parameterized exponential function described above. Let be the fluorescence intensity at time t. For the first The amplitude of each component, For the first The fluorescence lifetime of each component, For the first There are three attenuation components, where c is the background constant.

[0119] In this implementation, the objective function of fusing parameterized exponential priors is solved iteratively using PnP-ADMM, which can solve the underdetermined inversion problem and overcome the time grid limitation.

[0120] In one possible implementation, the solution is obtained by iteratively applying the plug-and-play alternating direction multiplier method; the least squares method or gradient descent method is used to ensure that the reconstruction results are consistent with the actual observation data; the intermediate variables are projected onto the physical constraint manifold by the proximal operator or exponential fitting operator based on the physical model, and the physical parameters are updated and output.

[0121] Among them, the data fidelity subproblem is optimized by using the least squares method or gradient descent method to ensure that the reconstruction results are highly matched with the actual observation data acquired by the hardware.

[0122] In some embodiments, the above iterative solution is implemented using a programming language.

[0123] It should be understood that directly solving the objective function, which simultaneously includes data fidelity terms and physical prior terms, is extremely complex. PnP-ADMM breaks it down into two sub-problems that are optimized alternately, transforming the complex joint optimization into a simple step-by-step solution, significantly reducing computational complexity. Through the above iterations, a high signal-to-noise ratio (SNR) and grid-free high-time-resolution transient response curve is finally derived from undersampled integral measurements. In the example, the reconstruction quality is evaluated using the SNR. An SNR exceeding 15 dB can be achieved at a compression ratio of 20:1, indicating good compression and reconstruction results. Regarding data acquisition, ultrafast gated signals at different time points within a time window of several hundred nanoseconds can be integrated under the same exposure time, greatly improving acquisition efficiency.

[0124] In this implementation, the objective function is solved by alternating PnP-ADMM. The least squares method or gradient descent method is used to ensure that the reconstruction results are consistent with the observation data. The intermediate variables are projected onto the exponentially decaying constrained manifold using physical model operators. This approach can balance data authenticity and physical rationality, accurately invert parameters such as microbial fluorescence lifetime, and improve algorithm efficiency and inversion accuracy.

[0125] This application provides a time-coded compressed sensing imaging method based on FPGA. This method calibrates FPGA delay units and constructs a delay lookup table based on a reference clock, accurately locking tap step values, compensating for process, voltage, and temperature drift, and achieving quantifiable and reproducible delay. This provides a stable hardware foundation for picosecond-level timing control. Because this solution employs both random and pushbroom dual-mode time coding, it can simultaneously handle incoherent compressed sensing sampling and high-precision timing scanning, adapting to mixed strong and weak signal acquisition and fine waveform reconstruction, thus improving system versatility and sampling efficiency. By using coarse and fine delay tap values, it achieves picosecond-level fine timing fine-tuning while ensuring a large nanosecond-level delay range, breaking through the accuracy limit of traditional clock counting and significantly improving the system's time resolution. In this scenario, the signal is synchronized to the reference clock domain, generating three aligned timing signals to eliminate cross-clock domain metastability and path deviation. This enables precise timing coordination between the excitation source, gated image intensifier, and camera, avoiding signal loss and sampling distortion. Utilizing non-uniform comb gating and exposure window integration, weak transient fluorescence signals can be accumulated multiple times, significantly improving the signal-to-noise ratio of observations and solving the problem of effectively detecting weak microbial light signals. A continuous observation operator is constructed based on physical timestamps to eliminate mismatch errors between the ideal sampling model and hardware behavior, providing a high-fidelity forward mapping relationship for inversion. Finally, by using an exponential decay physical prior and employing PnP-ADMM iterative solution, the solution space is constrained, noise is suppressed, and non-physical solutions are avoided. This overcomes the limitations of grid quantization, yielding a high signal-to-noise ratio continuous transient curve, accurately inverting microbial fluorescence lifetime, and improving parameter reliability and measurement accuracy.

[0126] Figure 4 This is a schematic diagram of the structure of a time-coded compressed sensing imaging system based on FPGA provided in an embodiment of this application.

[0127] For example, such as Figure 4 As shown, a time-coded compressed sensing imaging system based on FPGA of this application is described.

[0128] The sample to be tested is repeatedly excited by a controlled pulsed excitation source. For example, the sample to be tested includes the fluorescence lifetime of microorganisms. In this embodiment, it is based on a System on Chip (SoC), such as a heterogeneous SoC based on an ARM processor and an FPGA logic architecture. The core logic source of the system is the top-level design module at the Programmable Logic (PL) end, which interacts with the Process System (PS) end through an on-chip high-speed bus, a dual-port memory, and a high-speed on-chip bus. This architecture allows the PS end to arbitrarily modify the encoding sequence without reconstructing the logic on the PL side, ensuring no read / write conflicts and supporting high-speed random access. Based on this, the system constructs a complete physical signal driving link: the time encoding engine at the PL end generates three strictly aligned timing signals, which are then output after level conversion and pulse shaping by an external signal shaping and conditioning module; among them, the first synchronization signal triggers a pulsed laser or pulsed power supply to generate transient excitation, the second delayed gating signal drives a gated image intensifier to perform discrete sampling, and the third exposure window signal controls the camera to perform long-exposure integration acquisition. The entire hardware and software co-engineering architecture encompasses two core modules: hardware time-coding acquisition and software algorithm reconstruction.

[0129] Figure 5 This illustration shows a schematic diagram of the structure of a dynamic reconfiguration of an encoding strategy provided in an embodiment of this application.

[0130] For example, such as Figure 5As shown, through the configuration information of the host computer, the host computer determines whether to use random mode or pushbroom mode. The host computer calculates the physical timestamp, precise delay sequence, and generates a delay pattern, which is then written into the FPGA's BRAM (Browser RAM). The BRAM is the FPGA's internal memory, essentially a delay lookup table. The FPGA can receive the synchronization signal and read the current delay tap value from the BRAM. Based on this value, it generates the corresponding delay pulse and uses IDELAY fine-tuning to generate three strictly aligned timing signals. This means that the FPGA's internal high-speed clock domain register and output buffer generate low-jitter, steep-edge, high-quality timing signals. The target synchronization signal triggers the laser or power supply, the non-uniform comb gating controls the gated image intensifier, and the exposure window signal controls the camera to begin integration. In practical applications, the laser excites the sample, generating a transient fluorescence signal, i.e., an exponential decay curve. The gated image intensifier, under the control of the delayed gating signal, opens at a precise, extremely short moment (Gate window), allowing only a small segment of fluorescence signal to pass through at that moment. The camera, under the control of the exposure window signal, performs a long-term integration of all light passing through the intensifier throughout the process, ultimately outputting an observation value y. The host computer uses the previously calculated physical timestamp t and the Delay Pattern to construct the observation operator A in the software. It also initializes the initial value x(0) and regularization parameter λ of the transient signal sequence number x to be reconstructed. Finally, based on the observation operator A and the compressed sampled measurement vector y, the transient signal sequence x to be reconstructed is updated to ensure that Ax is as consistent as possible with the compressed sampled measurement vector y. In practical applications, to reduce interference, prior physical knowledge of the signal (such as the exponential nature of fluorescence decay) is used to constrain and denoise the updated signal, projecting it onto a reasonable physical form. The system checks whether the change between the two iterations is less than a certain threshold; if convergence is not achieved, it returns to continue iterating. If convergence is achieved, the loop exits, and the high-quality result obtained from the inversion, i.e., the fluorescence lifetime distribution map, is output. (Graph) or transient decay curve.

[0131] Figure 6 This is a schematic diagram of the structure of a time-coded compressed sensing imaging device based on FPGA provided in an embodiment of this application.

[0132] For example, the device 600 is specifically used for:

[0133] The module 601 is used to calibrate the delay unit in the FPGA according to the reference clock, obtain the tap step value of the delay unit and construct the delay lookup table.

[0134] The generation and writing module 602 is used to generate a time-encoded sequence based on the scanning strategy and write the time-encoded sequence into the delay lookup table. The scanning strategy is specifically used for random encoding mode and push-broom mode.

[0135] The read and parse module 603 is used to read the delay configuration word under the current index based on the delay lookup table in response to the synchronization signal, and parse it into coarse delay tap value and fine delay tap value;

[0136] The generation and synchronization module 604 is used to generate a reference pulse based on the coarse delay tap value, and synchronize the reference pulse and the fine delay tap value to the reference clock. Within the reference clock, the fine delay tap value is used to finely delay the reference pulse to generate a delayed pulse.

[0137] The generation module 605 is used to generate three aligned timing control signals based on the delayed pulse. The timing control signals are specifically used for a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time.

[0138] The control and acquisition module 606 is used to control the gated image sensor and acquire discrete integral observations using the non-uniform comb gating signal during the validity period of the exposure window signal.

[0139] The acquisition and construction module 607 is used to construct a continuous observation operator based on the discrete integral observations acquired from multiple sets of different time-coded sequences and the delay lookup table, and to generate an objective function based on the observation operator and the regularization term, wherein the regularization term is a physical prior based on the exponential decay model.

[0140] The iterative solution module 608 is used to iteratively solve the objective function based on the plug-and-play alternating direction multiplier method, and invert to obtain a high signal-to-noise ratio, grid-free transient response curve and related physical parameters. These physical parameters are specifically used for the fluorescence lifetime and carrier recombination lifetime of the luminescent material.

[0141] In one possible implementation, the generation and writing module 602 is specifically used for:

[0142] For the random coding mode, a random sequence that meets the noncoherence requirement of compressed sensing is generated. This random sequence is used to achieve mixed sampling of strong and weak signals.

[0143] For push-broom mode, a linearly increasing delay sequence is generated, which is used to uniformly step scan on the time axis to reconstruct a high-precision waveform profile.

[0144] In one possible implementation, the read and parse module 603 is specifically used for:

[0145] When a rising edge of the synchronization signal is detected, the read address pointer of the dual ports is driven.

[0146] Read the coarse delay tap value and the fine delay tap value under the current index;

[0147] Store the coarse delay tap value and the fine delay tap value into the register.

[0148] In one possible implementation, the generation and synchronization module 604 is specifically used for:

[0149] Load the fine delay tap value;

[0150] The reference pulse is superimposed with a phase shift to generate a delayed pulse.

[0151] In one possible implementation, the generation module 605 is specifically used for:

[0152] A non-uniform comb gating signal is generated using pulse width shaping logic;

[0153] The coarse delay value is loaded based on the system clock-driven counter, and a target synchronization signal and an exposure window signal are generated when the coarse delay value returns to zero.

[0154] The exposure window signal, the non-uniform comb gating signal, and the target synchronization signal are synchronously calibrated and aligned.

[0155] In one possible implementation, the acquisition and construction module 607 is specifically used for:

[0156] Retrieve the physical timestamp and time offset information from the delay lookup table;

[0157] Based on this physical timestamp, a precise integral mapping relationship is established between continuous physical time signals and discrete integral observations;

[0158] A continuous observation operator is constructed based on this mapping relationship.

[0159] In one possible implementation, the acquisition and construction module 607 is specifically used for:

[0160] The signal is reconstructed based on prior constraints constructed using a parameterized exponential function;

[0161] Based on the reconstructed signal according to the prior constraints, as well as the observation operator, observation time, observed pixel value, and regularization term, an objective function is generated. The formula for the objective function is: ,in, For the observation operator, The transient signal sequence to be reconstructed This is the vector of measured values ​​after compressed sampling. For regularization parameters, To introduce a physical prior regularization term.

[0162] In one possible implementation, the iterative solution module 608 is specifically used for:

[0163] The solution is obtained by alternating iterative steps using the plug-and-play alternating direction multiplier method.

[0164] The least squares method or gradient descent method is used to ensure that the reconstruction results are consistent with the actual observation data;

[0165] Physical model-based proximal operators or exponential fitting operators project intermediate variables onto the physical constraint manifold, update and output physical parameters.

[0166] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0167] For example, such as Figure 7 As shown, the electronic device 700 includes a memory 701 and a processor 702. The memory 701 stores executable program code 703, and the processor 702 is used to call and execute the executable program code 703 to perform a time-coded compressed sensing imaging method based on FPGA.

[0168] Furthermore, embodiments of this application also protect an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to perform the FPGA-based time-coded compressed sensing imaging method provided in embodiments of this application.

[0169] This embodiment can divide the device into functional modules based on the above method example. For example, each module can correspond to a separate function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0170] It should be understood that the apparatus provided in this embodiment is used to execute the above-described FPGA-based time-coded compressed sensing imaging method, and therefore can achieve the same effect as the above-described implementation method.

[0171] When using integrated units, the device may include a processing module and a storage module. When applied to an electronic device, the processing module can be used to control and manage the operation of the electronic device. The storage module can be used to support the execution of relevant program code by the electronic device.

[0172] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits shown in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0173] In addition, the apparatus provided in the embodiments of this application may specifically be a chip, component or module. The chip may include a connected processor and a memory. The memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute the FPGA-based time-coded compressed sensing imaging method provided in the above embodiments.

[0174] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the FPGA-based time-coded compressed sensing imaging method provided in the above embodiment.

[0175] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to realize the FPGA-based time-coded compressed sensing imaging method provided in the above embodiment.

[0176] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0177] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0178] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0179] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A FPGA-based time encoding compressive sensing imaging method, characterized in that, The method includes: The delay cells in the FPGA are calibrated according to the reference clock, the tap step value of the delay cells is obtained, and a delay lookup table is constructed. A time-encoded sequence is generated based on a scanning strategy, and the time-encoded sequence is written into the delayed lookup table. The scanning strategy includes a random encoding mode and a push-scan mode. In response to the synchronization signal, the delay configuration word under the current index is read based on the delay lookup table and parsed into coarse delay tap value and fine delay tap value; A reference pulse is generated based on the coarse delay tap value, and the reference pulse and the fine delay tap value are synchronized to the reference clock. Within the reference clock, the reference pulse is finely delayed using the fine delay tap value to generate a delayed pulse. Based on the delayed pulse, three aligned timing control signals are generated. The timing control signals include a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time. During the validity period of the exposure window signal, the gated image sensor is controlled by the non-uniform comb gating signal to acquire discrete integral observations; A continuous observation operator is constructed based on discrete integral observations collected from multiple sets of different time-coded sequences and the delay lookup table. An objective function is generated based on the observation operator and a regularization term, wherein the regularization term is a physical prior based on an exponential decay model. The objective function is iteratively solved using the plug-and-play alternating direction multiplier method, and the transient response curve with high signal-to-noise ratio and no grid limitation and the related physical parameters are obtained by inversion. The physical parameters include the fluorescence lifetime of the luminescent material and the carrier recombination lifetime. The generation of the objective function based on the observation operator and the regularization term includes: The signal is reconstructed based on prior constraints constructed using a parameterized exponential function; reconstructing a signal based on the prior constraint and generating a target function based on the observation operator, observation time, observation pixel value and regularization term, a formula of the target function is wherein, is an observation operator, is a transient signal sequence to be reconstructed, is a measurement value vector after compression sampling, is a regularization parameter, is a regularization term with introduction of a physical prior.

2. The method according to claim 1, characterized in that, The generation of time-encoded sequences based on the barcode scanning strategy includes: For the random coding mode, a random sequence that meets the noncoherence requirement of compressed sensing is generated, and the random sequence is used to realize mixed sampling of strong and weak signals; For push-broom mode, a linearly increasing delay sequence is generated, which is used to uniformly step scan on the time axis to reconstruct a high-precision waveform profile.

3. The method according to claim 1, characterized in that, In response to the synchronization signal, the delay configuration word under the current index is read based on the delay lookup table and parsed into coarse delay tap values ​​and fine delay tap values, including: When a rising edge of the synchronization signal is detected, the read address pointer of the dual ports is driven. Read the coarse delay tap value and the fine delay tap value under the current index; The coarse delay tap value and the fine delay tap value are stored in a register.

4. The method according to claim 1, characterized in that, The step of finely delaying the reference pulse using the fine delay tap value to generate a delayed pulse includes: Load the fine delay tap value; The reference pulse is superimposed with a phase shift to generate a delayed pulse.

5. The method according to claim 1, characterized in that, The generation of three aligned timing control signals based on the delayed pulse includes: A non-uniform comb gating signal is generated using pulse width shaping logic; The system clock drives the counter to load a coarse delay value, and when the coarse delay value returns to zero, a target synchronization signal and an exposure window signal are generated. The exposure window signal, the non-uniform comb gating signal, and the target synchronization signal are synchronously calibrated and aligned.

6. The method according to claim 1, characterized in that, The construction of a continuous observation operator based on discrete integral observations acquired from multiple sets of different time-coded sequences and the delay lookup table includes: Retrieve the physical timestamp and time offset information from the delay lookup table; Based on the physical timestamp, establish an accurate integral mapping relationship between continuous physical time signals and discrete integral observations; A continuous observation operator is constructed based on the mapping relationship.

7. The method according to claim 1, characterized in that, The objective function is iteratively solved using the plug-and-play alternating direction multiplier method, and the resulting high signal-to-noise ratio, mesh-free transient response curve and related physical parameters are obtained through inversion, including: The solution is obtained by alternating iterative steps using the plug-and-play alternating direction multiplier method. The least squares method or gradient descent method is used to ensure that the reconstruction results are consistent with the actual observation data; Proximal operators or exponential fitting operators based on physical models project intermediate variables onto the physical constraint manifold, update and output the physical parameters.

8. A time-coded compressed sensing imaging device based on FPGA, characterized in that, The device includes: A module for obtaining and building is used to calibrate delay cells in the FPGA according to a reference clock, obtain the tap step value of the delay cell, and build a delay lookup table. A generation and writing module is used to generate a time-encoded sequence based on a scanning strategy and write the time-encoded sequence into the delay lookup table. The scanning strategy includes a random encoding mode and a push-broom mode. The read and parse module is used to read the delay configuration word under the current index based on the delay lookup table in response to the synchronization signal, and parse it into coarse delay tap value and fine delay tap value; A generation and synchronization module is used to generate a reference pulse based on the coarse delay tap value, and synchronize the reference pulse and the fine delay tap value to the reference clock. Within the reference clock, the fine delay tap value is used to finely delay the reference pulse to generate a delayed pulse. The generation module is used to generate three aligned timing control signals based on the delayed pulse. The timing control signals include a target synchronization signal, a non-uniform comb gating signal, and an exposure window signal. The target synchronization signal is used to trigger the pulse excitation source, the non-uniform comb gating signal is used to control the gating window of the gated image intensifier, and the exposure window signal is used to control the camera exposure time. The control and acquisition module is used to control the gated image sensor and acquire discrete integral observations using the non-uniform comb gating signal during the validity period of the exposure window signal. The acquisition and construction module is used to construct a continuous observation operator based on discrete integral observations acquired from multiple sets of different time-coded sequences and the delay lookup table, and to construct a priori-constrained reconstructed signal based on a parameterized exponential function; based on the prior-constrained reconstructed signal, the observation operator, the observation time, the observed pixel value, and the regularization term, an objective function is generated, the formula of which is: ,in, For the observation operator, The transient signal sequence to be reconstructed This is the vector of compressed sampled measurements. For regularization parameters, To introduce a regularization term for physical priors; the regularization term is a physical prior based on an exponential decay model; The iterative solution module is used to iteratively solve the objective function based on the plug-and-play alternating direction multiplier method, and invert to obtain a high signal-to-noise ratio, grid-free transient response curve and related physical parameters, including the fluorescence lifetime of the luminescent material and the carrier recombination lifetime.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the electronic device to perform the method as described in any one of claims 1 to 7.

Images