Multispectral imaging system based on liquid crystal tunable filtering

Abstract

The invention discloses a multispectral imaging system based on liquid crystal tunable filtering, and relates to the technical field of spectral imaging. The system comprises an initialization module, a light quantum processing module, an imaging control module and a spectrum processing module, and high-quality entanglement source calibration and quantum gain guarantee of the initialization module, sub-nanometer wavelength dynamic compensation of the light quantum processing module and self-adaptive wave band compression acquisition of the imaging control module are realized. A complete closed loop covering photon utilization rate, signal-to-noise ratio improvement, data volume compression and calculation delay / power consumption optimization is formed through quantum accelerated unmixing and frequency domain analysis of the spectrum processing module; the method not only breaks through the minimum detectable radiance limitation of a traditional CCD / CMOS under the condition of weak light, but also improves the signal-to-noise ratio to be higher than the classic limit through the quantum entanglement and quantum information algorithm, meanwhile, the data scale is remarkably reduced under the scene of the hyperspectral channel number, and the time delay and power consumption of demixing and recognition are reduced by means of quantum acceleration.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of spectral imaging technology, in particular to a multispectral imaging system based on liquid crystal tunable filtering. BACKGROUND

[0002] Multispectral imaging refers to collecting images of the same scene or sample at several (usually several to tens of) predetermined narrow wavebands, thereby obtaining rich spectral information, each waveband width is usually between several nanometers to tens of nanometers, and the center wavelength is distributed in the visible light, near-infrared or even short-wave infrared region. Multispectral imaging is widely used in vegetation monitoring, medical diagnosis, food safety detection, archaeological detection and other fields, and material identification and quantitative analysis are realized by virtue of the reflection / absorption differences of different substances at each waveband. Therefore, multispectral imaging is widely used and very important.

[0003] At present, the photon utilization rate and dark noise of the traditional CCD / CMOS sensor limit the minimum detectable radiation of the system in the case of weak light or night vision, and the quantum characteristics of the photons cannot be used to break through the classical signal-to-noise ratio limit. At the same time, the large number of wavebands (tens to hundreds) generated by multispectral or even hyperspectral imaging bring about a large amount of data, and the calculation delay and power consumption are high when the traditional electronic DSP or GPU is used for real-time spectral unmixing and target recognition. SUMMARY

[0004] Therefore, it is necessary to provide a multispectral imaging system based on liquid crystal tunable filtering in view of the problems in the above background technology. The purpose of the present application can be achieved by the following technical scheme: a multispectral imaging system based on liquid crystal tunable filtering, comprising: an initialization module and a light quantum processing module.

[0005] The initialization module starts the spectrometer under a given pumping condition, sets the resolution to be less than or equal to 0.1 nm, records the spectral peak positions of the signal light and the idle light under the forward pumping of the PPKTP crystal (periodically poled potassium titanyl phosphate crystal), confirms the center wavelengths of the two peaks, which are the signal light wavelength (usually 1550 nm) and the idle light wavelength (usually 810 nm), respectively, and uses a single photon counter to count the coincidence count peak value and the random coincidence count of the entangled photon pairs under the same pumping condition. Specifically, the coincidence count of the entangled photon pairs is measured to obtain the maximum coincidence count in four kinds of base combinations, i.e. the coincidence count peak value C max The coincidence count is measured under the condition that there is no entangled photon pair or the system is not aligned to obtain the false coincidence count, i.e. the random coincidence count C rand; from two measurement angles of the angle set {a, a'} and {b, b'}, one angle is taken from each set to form four basic combinations (a, b), (a', b), (a, b') and (a', b'), and the correlations under the four basic combinations are calculated to obtain E(a, b), E(a', b), E(a, b') and E(a', b'), and then the inequality value S is calculated by using the CHSH inequality, and the calculation formula is: S = |E(a, b) + E(a', b) + E(a, b') - E(a', b')|; the quantum enhancement gain is defined as: wherein η det is the quantum efficiency of the detector, which is obtained by independent calibration or comparison light source method, C max and C rand need to be measured under the same pump power and temperature conditions; based on the CHSH inequality value and the quantum enhancement gain, a comprehensive analysis is performed to determine whether the initialization calibration process is successful, if successful, the real-time imaging mode is entered, that is, the light quantum processing module is executed, otherwise the re-calibration is automatically triggered;

[0006] The light quantum processing module obtains the transmittance and electrical signal response data of the liquid crystal tunable filter combined with the quantum dot sensor in the spectral range of 400nm-1700nm, and performs joint modeling of the liquid crystal tunable filter-quantum dot sensor according to the data, and dynamically compensates the incident angle.

[0007] In some embodiments, the initialization calibration judgment process is:

[0008] The inequality value S and the quantum enhancement gain G q According to the formula , the calibration judgment value F is calculated, if the calibration judgment value is greater than or equal to the preset judgment threshold (the preset judgment threshold is e±δ, wherein δ is a very small number, and the value is usually 0.02), the real-time imaging mode is entered, that is, the light quantum processing module is executed; if the calibration judgment value is less than the preset judgment threshold, the re-calibration mode is triggered until the calibration judgment value is greater than or equal to the preset judgment threshold; if the re-calibration times exceed a fixed number, an alarm is triggered.

[0009] In some embodiments, the re-calibration is specifically:

[0010] Step one, compare the inequality value S with the entanglement judgment value, if S is greater than or equal to the entanglement judgment value, directly jump to step two; otherwise, adjust the crystal temperature and pump power, optimize the entanglement generation rate according to the CHSH theoretical relationship curve, until S is greater than or equal to the entanglement judgment value, and then jump to step two;

[0011] Step two, compare the quantum enhancement gain G q with the enhancement threshold, if G q≥ enhancement threshold, then directly jump to step three, otherwise increase the detector calibration or optimize the coupling and collimation of the light path, optimize the collection efficiency of Cmax, until G q ≥ 10, and then jump to step three;

[0012] Step three, recalculate the inequality value S and the quantum enhancement gain G q , and quantitatively calculate the calibration segment judgment value according to this, if the calibration segment judgment value ≥ the preset judgment threshold, then end the calibration and enter the real-time imaging mode, that is, execute the light quantum processing module; otherwise, the re-calibration number is increased by one, and returns to step one; if the re-calibration number exceeds a fixed number, an alarm is triggered.

[0013] In some embodiments, the modeling compensation process is: 4-1, use a monochromator as an incident light source, output monochromatic light in the range of 400nm to 1700nm, step 1nm, LCTF is adjusted according to the control voltage to ensure that the center wavelength is aligned with the current incident wavelength, measure the LCTF transmittance at each wavelength by an optical power meter, denoted as T(λ), and collect the output response signal of the quantum dot photoelectric sensor at the wavelength, denoted as R(λ), the response signal here is a current signal, construct a continuous spectral response data table = {λ i ,T(λ i ),R(λ i )}, i is any one wavelength in 400 to 1700nm;

[0014] 4-2, adjust the incident light beam angle θ from 0° to the maximum angle θ max in steps, with a step size of 5°; at each angle θ j , measure the actual LCTF transmission center wavelength denoted as λ c (θ j ), where j is any one angle from 0° to the maximum angle θ max , j = 0°, 5°, …, θ max ; calculate the wavelength shift Δλ(θ j ) according to the formula Δλ(θ c ) = λ j (θ c ) - λ j (0°); use the least squares method to fit the relationship between the angle and the wavelength shift, assuming that the relationship is a quadratic function:

[0015] Δλ(θ) = A·θ 2 +B·θ+C, where Δλ(θ) is the wavelength shift at a certain incident angle θ, A, B and C are constant coefficients of the quadratic equation obtained by fitting, reflecting the LCTF characteristics, and θ is the incident angle;

[0016] 4-3, the least square optimization objective function is constructed as: where Δλ measured (θ j ) is the wavelength shift value corresponding to the jth incident angle actually measured; at all sampling angle points, a set of coefficients (A, B, C) is found to minimize the sum of squared errors between the fitting model output A·θ j 2 +B·θ j +C and the actual measured Δλ measured (θ j ), so as to obtain the optimal compensation model, and the incident angle caused transmission center wavelength shift is corrected in real time according to the offset compensation model, so as to ensure that even if the incident angle changes, the actual transmission wavelength of the LCTF can be kept near the expected value, and the stability and accuracy of the system are enhanced.

[0017] In some embodiments, further comprising an imaging control module and a spectral processing module;

[0018] The imaging control module performs waveband adaptive selection based on fast pre-scanning, adaptive screening and accurate scanning;

[0019] The spectral processing module combines high-resolution spectral data with quantum computing, performs fast unmixing and frequency domain analysis under a large-scale channel space, and then performs visual processing.

[0020] In some embodiments, the waveband adaptive selection process is as follows: 6-1, fast pre-scanning: drive the quantum dot sensor to measure the light intensity in the wavelength range in turn with a step of 5 nm, to obtain low-resolution spectral data {λ k ,I(λ k )}, λ k = λ min +5k, where k=0,1,2……K, K represents the index of the pre-scanning wavelength point, K is the index when the maximum wavelength is scanned, and I(λ k ) is the photocurrent measured at wavelength λ k ;

[0021] 6-2, adaptive screening: calculate the signal-to-noise index for each pre-scanning waveband to obtain the signal-to-noise ratio, and the calculation formula is where μ sig (λ k ) represents the average signal strength measured at wavelength λ k , μ niose (λ k ) and σ niose (λ k ) represent the average noise and standard deviation measured under shielding or dark field conditions; set the signal-to-noise ratio threshold SNR min, the waveband is reserved if the signal-to-noise ratio SNR of the waveband is greater than or equal to a signal-to-noise ratio threshold SNR min , and then the correlation of each reserved waveband with the characteristic spectrum P(λ) of the target substance is calculated, and the correlation value r(λ k ) is calculated by the Pearson correlation coefficient, and the calculation formula is:

[0022] , wherein P(λ) is the standard absorption or reflection spectrum value of the target substance; and are the average values of I(λ) and P(λ) in the window λ k +5m, respectively, for de-meaning, m is a window index offset, and M is a window half-width for controlling the correlation calculation, and the unit is step; the calculated correlation value r(λ k ) has a value range of [-1, 1], and is used to quantify the similarity of the pre-scanning signal and the target spectrum in the window; a preset correlation threshold r min , the waveband that satisfies r(λ k )≥r min is reserved, and further waveband screening is completed; 6-3, accurate scanning: the wavebands that have passed the double screening of the signal-to-noise ratio and the correlation degree are compressed in number, the signal-to-noise ratio SNR and the correlation value r are calculated according to the formula H=(SNR) 1.25 ×(r) 1.72 to obtain a waveband matching value H, each waveband is sorted in descending order according to the waveband matching value, and a preset proportion of the wavebands ranked at the top is selected as a key waveband set, and each waveband in the key waveband set is sorted in descending order according to the waveband matching value; the LCTF is driven to accurately scan in the order of each waveband in the key waveband set with a resolution of 1 nm.

[0023] In some embodiments, the fast unmixing and frequency domain analysis are:

[0024] In the real-time spectral processing flow, the spectral vector collected from the LCTF-quantum dot sensor is taken as an N-dimensional vector, the whole vector is normalized by calculating its Euclidean norm, so that the sum of squares of each component is 1, the set of normalized amplitudes is directly mapped to the amplitudes of n qubits by using amplitude encoding technology, the corresponding quantum state is constructed in the optical quantum processor through controlled rotation gates and C-NOT gates, and then the quantum Fourier transform is applied to the quantum state, so that the fast switching from time domain to frequency domain is realized under the logarithmic level of quantum gate number; the aliasing matrix obtained by experience is mapped into the Hamiltonian for quantum phase estimation, the eigenvalue thereof is read by QPE, and the eigenvalue is mapped into the reciprocal by means of the controlled rotation gate, so that the inverse operation of the aliasing matrix is equivalent completed; finally, the de-aliasing spectral vector is encoded back to the quantum register through inverse QPE and the corresponding multi-bit flip operation, and the de-aliasing and accurately separated spectral result is output in real time under the exponential acceleration.

[0025] In some embodiments, the visualization processing is:

[0026] The quantum register is subjected to a measurement operation, the result of each measurement is an n-bit binary value, corresponding to a certain specific ground state, by performing a plurality of independent measurements on the quantum register, the number of times falling on each ground state is counted to calculate the probability distribution of its occurrence, and the probability distribution reflects the amplitude square of each component in the normalized spectral vector; the square root of each probability value is taken to obtain the approximate amplitude value, and then the amplitude value is multiplied by the Euclidean norm of the original input spectral vector, so that the de-aliasing result in the true proportion is obtained; finally, the de-aliasing result in the true proportion is used for image display, specifically, the reaction result of each waveband is mapped into the intensity value of a pixel, the endmember abundance map is generated according to the endmember represented by different wavebands, and then the true color image is superimposed.

[0027] Compared with the prior art, the beneficial effects of the present application are:

[0028] 1. The initialization module can generate a highly reliable spectral database and quantum source performance index report at the beginning of system startup by high-resolution spectral line scanning of the entangled photon source under given pumping conditions, strict measurement of coincidence counting peaks and random coincidence counting of entangled photon pairs by a single-photon counter, calculation of CHSH inequality values using correlation functions under four basis combinations, and calculation of quantum enhancement gain combined with quantum efficiency of the detector. The initialization module can accurately determine the high-quality conditions of the entangled source and fine-tune the crystal temperature and pumping power to optimize the performance of the quantum source in the least number of adjustments. In this way, the entire multispectral imaging system has stable, repeatable, and significantly noise-reduced quantum entangled photon pairs before entering the subsequent acquisition and processing process, providing a solid quantum gain foundation for imaging in low-light and weak-signal environments; 2. By organically combining a liquid crystal tunable filter (LCTF) with a quantum dot photoelectric sensor and scanning at 1 nm steps in the 400 nm-1700 nm range, the filter transmittance and sensor response current signal at each wavelength are obtained, and the transmission center wavelength shift under different incident angles is fitted by a second-order polynomial least squares method to generate a compensation model in real time. The module solidifies the fitting model into the FPGA driver, automatically adjusts the control voltage using the real-time readings of the angle sensor, so that the LCTF can output the target wavelength under any incident deviation. Not only does this significantly reduce the spectral drift error caused by changes in viewing angle, but it also achieves closed-loop self-stabilization of the optical path at the hardware level, ensuring that the system always maintains sub-nanometer transmission wavelength accuracy in dynamic scenarios, providing high-fidelity and repeatable spectral input for subsequent quantum and classical combined imaging.

[0029] 3. The four-step adaptive band selection strategy of pre-scanning, signal-to-noise ratio, correlation, and compression screening quickly reduces dozens to hundreds of traditional hyperspectral bands to one-third of the most information-rich bands. First, low-resolution spectra are obtained at 5 nm steps and the signal-to-noise ratio of each segment is calculated, and low signal-to-noise segments are removed. Then, the Pearson correlation of the remaining bands with the target material standard absorption spectrum is calculated, and only bands with signal-to-noise ratio and correlation higher than the preset threshold are retained. Finally, the candidate bands are sorted in descending order according to the weighted matching value, and the first 1 / 3 are taken as the key band set, driving the LCTF to quickly scan at 1 nm accuracy on these selected bands, significantly improving the scanning rate and reducing the data volume while retaining the most critical spectral features, greatly improving the imaging efficiency and signal-to-noise performance in low-light environments and real-time target recognition tasks.

[0030] 4. By normalizing the high-resolution spectral vector by 2-norm on the classical side, mapping it onto the logarithmic-level qubits by amplitude encoding, constructing the quantum state, then switching from time domain to frequency domain by means of quantum Fourier transform quantum gate complexity, and implementing exponentially accelerated matrix inversion on the aliasing matrix obtained by experience through HHL algorithm, the clean spectrum is quickly unmixed; After measurement, reverse normalization is performed to restore the true physical quantity, and output to the visualization pipeline, making full use of quantum parallelism and exponential acceleration characteristics, completely solving the real-time unmixing bottleneck of classical DSP / GPU under high-channel large data volume, significantly reducing the calculation delay and power consumption. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0032] Figure 1 It is a schematic diagram of the system module of the present application.

[0033] Figure 2 It is a schematic diagram of the initialization calibration process of the present application. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below. In the following description, a lot of specific details are set forth in order to fully understand the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the scope of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0035] As shown in Figure 1 , the multispectral imaging system based on liquid crystal tunable filter includes an initialization module, a light quantum processing module, an imaging control module, and a spectral processing module.

[0036] As shown in Figure 2 , under the given pump condition, the initialization module determines the center wavelength of the entangled photon pair, coincidence count and CHSH inequality (Clauser-Horne-Shimony-Holt inequality, which is a core criterion for measuring quantum entanglement) measurement and evaluation of quantum enhancement gain through spectral measurement and single photon counting; Specifically:

[0037] Start the spectrometer, set the resolution ≤ 0.1 nm, under the forward pumping of the PPKTP crystal (periodically poled potassium titanyl phosphate crystal), record the spectral peak positions of the signal light and the idle light respectively, confirm the center wavelengths of the two peaks, which are the signal light wavelength (usually 1550 nm) and the idle light wavelength (usually 810 nm) respectively, and use a single photon counter to count the coincidence count peak value and the random coincidence count of the entangled photon pairs under the same pumping condition. Specifically, coincidence counting measurement is performed on the entangled photon pairs to obtain the maximum coincidence count, i.e. the coincidence count peak value C max ; coincidence counting measurement is performed under the condition that there is no entangled photon pair or the system is misaligned to obtain the false coincidence count, i.e. the random coincidence count C rand ; from the two angle sets {a, a'} and {b, b'}, one angle is taken from each set to form four base combinations, which are (a, b), (a', b), (a, b') and (a', b'), and the correlations under the four base combinations are calculated to obtain E(a, b), E(a', b), E(a, b') and E(a', b'); then the inequality value S is calculated using the CHSH inequality, and the calculation formula is: S = |E(a, b) + E(a', b) + E(a, b') - E(a', b')|; the quantum enhancement gain is defined as: where η det is the quantum efficiency of the detector, which is obtained by independent calibration or comparison light source method, C max and C rand need to be measured under the same pumping power and temperature conditions;

[0038] The quantum mechanical upper bound (i.e. Tsirelson upper bound) is When S>2, it is considered that the classical local realism has been violated, indicating the existence of entanglement, but in actual operation, setting the boundary value as 2 is too broad, and it cannot be guaranteed that the obtained entanglement source is of high quality, repeatable and with small noise. However, it is almost impossible to achieve perfect quantum mechanical theory upper bound in the actual process; therefore, the entanglement judgment threshold is set to 2.7 by the person skilled in the art; if S<entanglement judgment threshold (2.7), it indicates that the quality of the entanglement source is not enough; there is an enhancement threshold (set to 10), if quantum enhancement gain G q <10, it indicates that the enhancement ratio is insufficient, and if quantum enhancement gain G q ≥10, it indicates that the quantum noise compression effect is significant; the inequality value S and the quantum enhancement gain G q are calculated according to the formula The calibration judgment value F is calculated, and if the calibration judgment value is greater than or equal to a preset judgment threshold (the preset judgment threshold is e±δ, where δ is a very small number, and is usually 0.02), it indicates that the calibration is successful, and the real-time imaging mode is entered, that is, the light quantum processing module is executed; if the calibration judgment value is greater than or equal to the preset judgment threshold, it indicates that the re-calibration mode is reached, and the calibration judgment value is greater than or equal to the preset judgment threshold; if the re-calibration times exceed 5, an alarm is triggered to inform the engineer to overhaul; the re-calibration is specifically as follows:

[0039] Step one, compare the inequality value S with the entanglement judgment value, if S is greater than or equal to the entanglement judgment value, directly jump to step two; otherwise, adjust the crystal temperature and the pump power, optimize the entanglement generation rate according to the CHSH theoretical relationship curve, until S is greater than or equal to the entanglement judgment value, and then jump to step two;

[0040] Step two, calculate the quantum enhancement gain G q , and compare G q with the enhancement threshold, if G q is greater than or equal to the enhancement threshold, directly jump to step three, otherwise, increase the detector calibration or optimize the light path coupling and collimation to improve the collection efficiency of Cmax, until G q is greater than or equal to 10, and then jump to step three;

[0041] Step three, recalculate the inequality value S and the quantum enhancement gain G q , and calculate the calibration segment judgment value according to the calculation, if the calibration segment judgment value is greater than or equal to the preset judgment threshold, the calibration is ended, and the real-time imaging mode is entered, that is, the light quantum processing module is executed; otherwise, the re-calibration times are accumulated by one, and the process returns to step one; if the re-calibration times exceed 5, an alarm is triggered; by performing high-resolution spectral line scanning on the entangled photon source under the given pump condition, and combining the coincidence counting peak value of the entangled photon pairs and the random coincidence counting measured by the single photon counter, the initialization module can generate a highly reliable spectral database and quantum source performance index report at the beginning of system startup, calculate the CHSH inequality value according to the correlation function under four kinds of basis combinations, and calculate the quantum enhancement gain according to the quantum efficiency of the detector, the initialization module can accurately determine the high-quality condition of the entangled source, and through the crystal temperature and pump power fine tuning, the best quantum source performance is achieved in the least number of debugging times; in this way, the entire multi-spectral imaging system has stable, repeatable and significant noise compression quantum entangled photon pairs before entering the subsequent acquisition and processing process, providing a solid quantum gain foundation for imaging in low light and weak signal environments. The light quantum processing module obtains the transmittance and electrical signal response data of the liquid crystal tunable filter (LCTF) combined with the quantum dot sensor in the 400nm-1700nm spectral range, and performs LCTF-quantum dot sensor joint modeling according to the data, dynamically compensates the incident angle, and ensures the precise controllability of the system transmission wavelength; specifically as follows:

[0042] Using a high-stability monochromator as an incident light source, monochromatic light with an output range of 400-1700 nm is stepped by 1 nm, the LCTF is adjusted according to the control voltage to ensure that the center wavelength is aligned with the current incident wavelength, the LCTF transmittance at each wavelength is measured by a light power meter and recorded as T(λ), and the output response signal R(λ) of the quantum dot photoelectric sensor at the wavelength is collected at the same time, where the response signal is a current signal, and a continuous spectral response data table = {λ i ,T(λ i ),R(λ i )} is constructed, i is any one wavelength in 400-1700 nm, i = 400, 401, 402, …, 1700; the incident light beam angle θ is gradually adjusted from 0° to the maximum angle θ max , with a step size of 5°; at each angle θ j , the actual LCTF transmittance center wavelength is measured and recorded as λ c (θ j ), where j is any one angle from 0° to the maximum angle θ max , j = 0°, 5°, …, θ max ; the wavelength shift Δλ(θ j ) is calculated according to the formula Δλ(θ c ) = λ j (θ c )-λ j (0°); the least squares method is used to fit the relationship between the angle and the wavelength shift, assuming that the relationship is a quadratic function: Δλ(θ) = A·θ 2 +B·θ+C, where Δλ(θ) is the wavelength shift at a certain incident angle θ, A, B and C are constant coefficients of the quadratic equation obtained by fitting, reflecting the LCTF characteristics, and θ is the incident angle;

[0043] The least squares optimization objective function is constructed as: where Δλ measured (θ j ) is the wavelength shift value corresponding to the jth incident angle actually measured; at all sampling angle points, a set of coefficients (A, B, C) is found to minimize the sum of squared errors between the fitted model output A·θ j 2 +B·θ j +C and the actual measured Δλ measured (θ j ), thereby obtaining the optimal compensation model, and the incident angle-induced transmittance center wavelength shift is corrected in real time according to the offset compensation model, ensuring that the LCTF actual transmittance wavelength can be kept near the expected value even if the incident angle changes, enhancing the stability and accuracy of the system;

[0044] By organically combining liquid crystal tunable filter (LCTF) with quantum dot photoelectric sensor, and scanning at 1 nm step in the range of 400 nm-1700 nm, the filter transmittance and sensor response current signal at each wavelength are obtained, and the transmission center wavelength shift at different incident angles is fitted by a second-order polynomial least squares, and a compensation model is generated in real time. The model is solidified into the FPGA driver, and the control voltage is automatically adjusted by using the real-time reading of the angle sensor, so that the LCTF can output the target wavelength under any incident deviation. Not only the spectral drift error caused by the change of viewing angle is significantly reduced, but also the closed loop self-stabilization of the optical path is realized at the hardware level, so that the system as a whole always maintains sub-nanometer level transmission wavelength accuracy in dynamic scenes, and provides high-fidelity and repeatable spectral input for subsequent quantum and classical combined imaging.

[0045] The imaging control module is based on adaptive waveband selection, which greatly improves the imaging efficiency and signal-to-noise performance on the premise of ensuring key spectral information. Specifically:

[0046] The quantum dot sensor is driven to measure the light intensity in the wavelength range with a step of 5 nm, and low-resolution spectral data {λ k ,I(λ k )} is obtained, λ k =λ min +5k, where k=0,1,2……K, K represents the index of the pre-scanned wavelength point, K is the index when the maximum wavelength is scanned, and I(λ k ) is the measured photocurrent at wavelength λ k ; the signal-to-noise index is calculated for each pre-scanned waveband λ k , and the measurement reliability is quantified, and the calculation formula is where μ sig (λ k ) represents the average signal strength of multiple measurements at wavelength λ k , μ niose (λ k ) and σ niose (λ k ) represent the average noise value and standard deviation measured under shielding or dark field conditions; set the signal-to-noise ratio threshold SNR min , and keep the wavebands greater than or equal to the signal-to-noise ratio threshold SNR min , and then calculate the correlation of each retained waveband with the target substance characteristic spectrum P(λ); the Pearson correlation coefficient is used to calculate the correlation value r(λ k ) by those skilled in the art, and the calculation formula is: where P(λ) is the standard absorption or reflection spectrum value of the target substance; and I(λ) and P(λ) are respectively the average value of I(λ) and P(λ) in the window λ k + 5m, for de-meaning, m is the window index offset, M is the window half-width for controlling the correlation calculation, and the unit is step, 5nm per step according to the description above; according to the formula, the correlation value r(λ k ) calculated is in the range of [-1, 1], which is used to quantify the similarity of the pre-scanned signal and the target spectrum in the window; a preset correlation threshold r min , the wavelength band that satisfies r(λ k ) ≥ r min is reserved, and further wavelength band screening is completed; the wavelength bands that have passed the double screening of signal-to-noise ratio and correlation degree are compressed in number, and the signal-to-noise ratio SNR and the correlation value r are calculated according to the formula H = (SNR) 1.25 × (r) 1.72 to obtain the wavelength band matching value H, and the larger the wavelength band matching value, the better the wavelength band in signal instruction and target feature matching performance; the wavelength bands are sorted in descending order according to their wavelength band matching values, and a preset proportion (the preset proportion is set to one-third by the present technician, and this value can be adjusted according to actual conditions) of the top-ranked wavelength bands is selected as a key wavelength band set, and the wavelength bands in the key wavelength band set are sorted in descending order according to their wavelength band matching values; the LCTF is driven to scan accurately in sequence according to the order of the wavelength bands in the key wavelength band set with a resolution of 1nm, and high-resolution data is obtained; compared with full-spectrum 1nm scanning, the scanning wavelength bands are reduced to about one-third, which can significantly improve the imaging speed and data processing efficiency, and focus on the most valuable wavelength bands;

[0047] Through the four-step adaptive wavelength selection strategy of pre-scanning, signal-to-noise ratio, correlation degree and compression screening, dozens to hundreds of traditional hyperspectral wavelength bands are rapidly reduced to one-third of the most information content; first, low-resolution spectra are obtained with a step of 5nm and the signal-to-noise ratio of each segment is calculated, and low signal-to-noise segments are removed; then, Pearson correlation degree calculation is performed on the remaining wavelength bands and the standard absorption spectrum of the target substance, and only the wavelength bands with signal-to-noise ratio and correlation degree higher than the preset threshold are retained; finally, the candidate wavelength bands are sorted in descending order according to the weighted matching value, and the first 1 / 3 is taken as a key wavelength band set, and the LCTF is driven to scan quickly on these selected wavelength bands with a precision of 1nm, which significantly improves the scanning rate, reduces the data volume, and retains the most critical spectral features, greatly improving the imaging efficiency and signal-to-noise performance in low-light environments and real-time target recognition tasks.

[0048] The spectral processing module combines high-resolution spectral data with quantum computing to realize fast unmixing and frequency domain analysis in a large-scale channel space, and then performs visual processing; specifically:

[0049] In the real-time spectral processing flow, the high-resolution spectral data (i.e., high-resolution spectral vector) collected from the LCTF-quantum dot sensor is first normalized as an N-dimensional vector by calculating its Euclidean norm (i.e., 2-norm, square root of the sum of squares of all channel signal values) to make the sum of squares of its components equal to 1; then the amplitude encoding technique is used to directly map this set of normalized amplitudes to the amplitudes of n qubits (where N = 2 n ), and a corresponding quantum state is constructed in the optical quantum processor through a series of controlled rotation gates and C-NOT gates, followed by the quantum Fourier transform on this quantum state to achieve fast switching from the time domain (channel index) to the frequency domain (Fourier coefficient) with a logarithmic number of quantum gates; then for the spectral crosstalk problem caused by aliasing or convolution, the empirically obtained aliasing matrix is mapped to the Hamiltonian that can be used for quantum phase estimation, and its eigenvalues are read through QPE, and these eigenvalues are mapped to their reciprocals through controlled rotation gates, thus equivalent to completing the inverse operation of the aliasing matrix; finally, through inverse QPE and the corresponding multi-bit flip operation, the de-aliased spectral vector is encoded back into the quantum register, and the de-aliased and accurately separated spectral results are output in real time with exponential acceleration; after quantum de-aliasing calculation, the spectral information in the system has been encoded in the quantum state. In order to obtain these results, multiple measurement operations need to be performed on the quantum register, and the result of each measurement is an n-bit binary value corresponding to a certain specific ground state. By performing multiple independent measurements on the quantum register, the number of times falling into each ground state is counted to calculate its probability distribution, which reflects the amplitude square of each component in the normalized spectral vector; however, the HHL algorithm calculates the de-aliased spectrum after normalization, so the measured probability needs to be denormalized to recover the actual physical quantity; first take the square root of each probability value to get the approximate amplitude value, then multiply these amplitude values by the Euclidean norm of the original input spectral vector calculated earlier, to get the de-aliased result in the true proportion, which ensures that the spectral intensity output by the system has actual physical meaning, rather than just the amplitude value in the quantum state; finally, the de-aliased result in the true proportion is used for image display, specifically mapping the reaction result of each waveband into the intensity value of the pixel, generating an endmember abundance map according to the endmember represented by different wavebands, and then superimposing a true color (RGB) image to realize the fusion of spectral data and real visual image, enhancing the interpretability and intuitive effect of imaging;

[0050] By normalizing the high-resolution spectral vector by 2-norm on the classical side, mapping it to the logarithmic level qubit by amplitude coding, constructing the quantum state, then switching the time domain to the frequency domain by the quantum Fourier transform quantum gate complexity, and implementing the exponentially accelerated matrix inversion of the aliasing matrix obtained by experience through the HHL algorithm, the clean spectrum is quickly demixed; After measurement, the real physical quantity is recovered by reverse normalization, and output to the visualization pipeline, making full use of the characteristics of quantum parallelism and exponential acceleration, completely solving the real-time demixing bottleneck of classical DSP / GPU under high channel and large data volume, significantly reducing the calculation delay and power consumption. The technical features of the above embodiments can be combined arbitrarily, in order to make the description simple, not all possible combinations of each technical feature in the above embodiments are described, however, as long as the combination of these technical features does not exist contradictory, it should be considered as the scope of the present application.

[0051] The above formula is an example of an empirical formula defined by a person skilled in the art, as long as the formula conforms to the relationship between the parameters of the present application, the specific size of the weight factor in the formula is reasonably set by a person skilled in the art according to actual use.

[0052] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1. A multispectral imaging system based on liquid crystal tunable filtering, characterized in that: include: Initialization module and photon processing module; The initialization module starts the spectrometer under given pumping conditions, scans the spectrum peaks of the signal light and idle light of the entangled photon pair under the forward pump output of the periodically polarized potassium titanyl phosphate crystal, determines the central wavelength of the two photons, and uses a single-photon counter to count the coincidence counts of the entangled photon pair under the same pump power and temperature conditions, obtains the highest coincidence count under four measurement basis combinations, and counts the random coincidence counts when the system is misaligned or there are no entangled photons. The quantum enhancement gain is calculated using these two sets of data and the detector quantum efficiency. The correlation function is calculated by counting the four types of coincidence and non-coincidence events under each pair of basis combinations, and the CHSH inequality value is obtained based on this. A comprehensive analysis based on the CHSH inequality value and the quantum enhancement gain is performed to determine whether the initialization calibration process is successful. If successful, the real-time imaging mode is entered, that is, the optical quantum processing module is executed. Otherwise, recalibration is automatically triggered. The photon processing module obtains the transmittance and electrical signal response data of the liquid crystal tunable filter combined with the quantum dot sensor in the 400nm-1700nm spectral range, and based on this, performs joint modeling of the liquid crystal tunable filter and quantum dot sensor to dynamically compensate for the incident angle.

2. The multispectral imaging system based on liquid crystal tunable filtering according to claim 1, characterized in that: The initialization calibration judgment process is: The inequality value and the quantum enhancement gain are quantized and calculated to obtain a calibration judgment value. If the calibration judgment value is greater than or equal to the preset judgment threshold, the real-time imaging mode is entered, that is, the photon processing module is executed; If the calibration judgment value is less than the preset judgment threshold, the recalibration mode is triggered until the calibration judgment value is greater than or equal to the preset judgment threshold; If the number of recalibration times exceeds a fixed number, an alarm is triggered.

3. The multispectral imaging system based on liquid crystal tunable filtering according to claim 2, characterized in that: The recalibration is as follows: Step 1: Compare the inequality value S with the entanglement judgment value. If S ≥ the entanglement judgment value, jump directly to step 2. Otherwise, adjust the crystal temperature and pump power according to the CHSH theoretical relationship curve to optimize the entanglement generation rate until S ≥ the entanglement judgment value, then jump to step 2. Step 2: The quantum enhancement gain G q Compared with the enhancement threshold, if G q ≥ Enhancement Threshold, then jump directly to step 3, otherwise increase detector calibration or optimize optical path coupling and collimation, optimize Cmax collection efficiency, until G q ≥10, then jump to step 3; Step 3: Recalculate the inequality value S and quantum enhancement gain G q , and quantify the calibration segment judgment value based on it. If the calibration segment judgment value is ≥ the preset judgment threshold, the calibration is terminated and the real-time imaging mode is entered, that is, the photon processing module is executed; otherwise, the number of recalibrations is accumulated and returns to step 1; if the number of recalibrations exceeds the fixed number, an alarm is triggered.

4. The multispectral imaging system based on liquid crystal tunable filtering according to claim 3, characterized in that: The modeling compensation process is: 4-1. A monochromator is used as the incident light source, and the output range is monochromatic light from 400 nm to 1700 nm, with a step of positive integers. The LCTF is adjusted according to the control voltage to ensure that its central wavelength is aligned with the current incident wavelength. The LCTF transmittance at each wavelength is measured by an optical power meter and recorded as T(λ). At the same time, the output response signal R(λ) of the quantum dot photoelectric sensor at that wavelength is collected. Here, the response signal is a current signal. A continuous spectral response data table is constructed = {λ i ,T(λ i ), R(λ i )}, i is any wavelength between 400 and 1700 nm; 4-2, gradually adjust the incident beam angle θ from 0° to the maximum angle θ max , the step size is a positive integer; at each angle, the actual LCTF transmission center wavelength is measured, and the wavelength offset is calculated by subtracting it from the transmission center wavelength when the incident beam angle is 0°. The relationship between the angle and the wavelength offset is fitted using the least squares method to construct a quadratic equation with respect to the incident beam angle θ, which reflects the LCTF characteristics; 4-3. Construct the least squares optimization objective function and find a set of constant coefficients of the quadratic equation at all sampling angle points so that the sum of the square errors between the fitted quadratic equation and the actual measured wavelength offset value is minimized, thereby obtaining the optimal compensation model. Based on this offset compensation model, the transmission center wavelength offset caused by the incident angle is corrected in real time.

5. The multispectral imaging system based on liquid crystal tunable filtering according to claim 4, characterized in that: Also includes an imaging control module and a spectrum processing module; The imaging control module performs adaptive band selection based on fast pre-scanning, adaptive screening and precise scanning; The spectral processing module combines high-resolution spectral data with quantum computing to perform rapid unmixing and frequency domain analysis in large-scale channel space, and then visualizes it.

6. The multispectral imaging system based on liquid crystal tunable filtering according to claim 5, characterized in that: The band adaptive selection process is: 6-1, Fast pre-scan: Drive the quantum dot sensor to measure the light intensity within the wavelength range in a fixed step size, and obtain low-resolution spectral data {λ k ,I(λ k )},λ k =λ min +5k, where k = 0, 1, 2 ... K, K represents the index of the pre-scan wavelength point, K is the index when the maximum wavelength is scanned, I (λ k ) is the wavelength λ k The photocurrent measured at 6-2, Adaptive screening: Calculate the signal and noise index for each pre-scan band to obtain the signal-to-noise ratio, and set the signal-to-noise ratio threshold SNR min , retaining a value greater than or equal to the signal-to-noise ratio threshold SNR min The bands of each retained band are then calculated for their correlation with the characteristic spectrum P(λ) of the target substance. The technicians in this field use the Pearson correlation coefficient to calculate the correlation value r(λ k ), the calculated correlation value r(λ k ) takes a value in the range of [-1,1] and is used to quantify the similarity between the pre-scan signal and the target spectrum within the window; Preset correlation threshold r min , will satisfy r(λ k )≥r min The bands are retained to complete further band screening; 6-3, Precision Scanning: Compress the number of bands after double screening of signal-to-noise ratio and correlation, quantify the signal-to-noise ratio and correlation value to obtain the band matching value, sort the bands in descending order according to their band matching value, select the top preset proportion as the key band set, and sort the bands in the key band set in descending order according to their band matching value; drive LCTF to scan the bands in the key band set in sequence with the preset resolution.

7. The multispectral imaging system based on liquid crystal tunable filtering according to claim 6, characterized in that: Fast unmixing and frequency domain analysis: In the real-time spectral processing flow, the spectral vector collected from the LCTF-quantum dot sensor is treated as an N-dimensional vector. The entire vector is normalized by calculating its Euclidean norm so that the sum of the squares of its components is 1. Using amplitude encoding technology, this set of normalized amplitudes is directly mapped to the amplitudes of n quantum bits. The corresponding quantum state is constructed in the optical quantum processor through controlled rotation gates and C-NOT gates. A quantum Fourier transform is then applied to this quantum state, achieving rapid switching from the time domain to the frequency domain with a logarithmic number of quantum gates. The empirically obtained aliasing matrix is ​​mapped to the Hamiltonian for quantum phase estimation, and its eigenvalues ​​are read through QPE. The eigenvalues ​​are mapped to reciprocals with the help of controlled rotation gates, thereby equivalently completing the inverse operation of the aliasing matrix. Finally, through inverse QPE and corresponding multi-bit flipping operations, the de-aliased spectral vector is encoded back into the quantum register, and the de-aliased and precisely separated spectral results are output in real time with exponential acceleration.

8. The multispectral imaging system based on liquid crystal tunable filtering according to claim 7, characterized in that: The visualization process is: A measurement operation is performed on the quantum register. The result of each measurement is an n-bit binary value, corresponding to a specific ground state. By performing several independent measurements on the quantum register, the number of times it falls on each ground state is counted to calculate the probability distribution of its occurrence. This probability distribution reflects the square of the amplitude of each component in the normalized spectral vector; the square root of each probability value is taken to obtain an approximate amplitude value, and then the amplitude value is multiplied by the Euclidean norm of the original input spectral vector to obtain the unmixing result at the true scale; finally, the unmixing result at the true scale is used for image display. Specifically, the response result of each band is mapped to the intensity value of the pixel, and an endmember abundance map is generated according to the endmembers represented by different bands, and then a true color image is superimposed.

Images