Liquid crystal filtering multispectral image generation method based on quantum inspiration

Abstract

The invention relates to the technical field of image processing, in particular to a liquid crystal filtering multispectral image generation method based on quantum inspiration. According to the invention, a quantum inspiration optimization structure is combined with liquid crystal filtering hardware, the filtering parameter optimization speed and precision are improved by referring to quantum superposition and parallelism characteristics, and the spectral band dynamic switching is realized by using the electrically-controlled adjustable characteristic of liquid crystal filtering; the problems that a traditional mechanical filtering structure is poor in adjustability and low in integration level and a traditional algorithm is low in high-dimensional data processing efficiency are solved, and high-quality original signals and efficient calculation support are provided for multispectral image generation. According to the method, an image quality analysis and feedback mechanism is constructed, the image quality is evaluated based on parameters such as spectral angular distance and residual noise energy, bad images are fed back to an optimization link to be reprocessed, an optimization frequency threshold value is set to avoid invalid circulation, the quality stability of output images is effectively guaranteed, and the method is suitable for large-scale popularization and application. And the problems of spectrum distortion, noise and difficulty in balancing details are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a liquid crystal filtering multi-spectral image generation method based on quantum inspiration. BACKGROUND

[0002] Quantum inspiration optimizes filter parameter solving and image reconstruction process by drawing on quantum superposition, parallelism and other characteristics, breaks through the efficiency bottleneck of traditional algorithms in high-dimensional data processing, improves parameter optimization speed and reconstruction accuracy, and provides efficient computing support for multi-spectral image generation. Relying on the electrically controllable adjustable characteristics of liquid crystal filtering, dynamic switching of spectral bands is realized, replacing traditional mechanical filtering structures, and at the same time, the hardware volume is reduced through micro-nano integrated design, providing high-quality, high-flexibility original spectral signals for multi-spectral image generation, solving the limitations of traditional filtering in adjustability and integration. In the existing multi-spectral image generation technology, the traditional mechanical filtering structure has the problems of poor adjustability, low integration, difficulty in dynamically switching spectral bands, and large hardware volume. At the same time, image reconstruction relies on traditional algorithms, which are inefficient in high-dimensional data processing, and are prone to spectral distortion such as weakening of red edge effect and insufficient spatial resolution. In addition, in low light environment, noise suppression and detail preservation are difficult to balance, resulting in poor image quality stability and inability to meet high-precision scene requirements. SUMMARY

[0003] The present application provides a liquid crystal filtering multi-spectral image generation method based on quantum inspiration to solve the above technical problems.

[0004] The present application provides a liquid crystal filtering multi-spectral image generation method based on quantum inspiration, comprising the following steps:

[0005] S1: Obtain the hardware module performance report corresponding to the pre-deployed liquid crystal filtering hardware module, and obtain the multi-band original spectral signal set and signal accessory information through the liquid crystal filtering hardware module for original spectral signal acquisition;

[0006] The multi-band original spectral signal set includes a band list, spectral intensity data, and a multi-band original signal data set; the signal accessory information includes signal-to-noise data and noise floor data for each band; the liquid crystal filtering hardware module is an integrated structure containing a liquid crystal filtering unit, a driving circuit, and an optical coupling component; the hardware module performance report includes but is not limited to a dynamically switchable spectral band list, a band switching response time, an integrated volume parameter, and transmittance efficiency data for each band.

[0007] S2: Obtain the signal accessory information and the multi-band original spectral signal set based on the pre-deployed quantum inspiration filtering optimization structure, and perform parameter optimization analysis on the signal accessory information and the multi-band original spectral signal set to output a quantum inspiration parameter optimization model;

[0008] As a further improvement of the present application, the signal auxiliary information and the multi-band original spectrum signal set are subjected to parameter optimization analysis, and the specific analysis content is as follows:

[0009] The quantum-inspired filter optimization structure comprises a parameter optimization triggering unit and a quantum-inspired filter optimization model.

[0010] The parameter optimization triggering unit is used for signal triggering analysis of the signal auxiliary information and the multi-band original spectrum signal set to obtain an optimization start signal.

[0011] The quantum-inspired filter optimization model monitors the optimization start signal through a preset start signal monitor, and when the optimization start signal is monitored, the quantum-inspired filter optimization model performs optimization execution on the signal auxiliary information and the multi-band original spectrum signal set to output a quantum-inspired parameter optimization model.

[0012] Further, the quantum-inspired parameter optimization model specifically comprises: obtaining the monitored signal noise abnormal signal, noise instability signal, band missing signal, spectrum abnormal signal and band redundancy signal corresponding to the optimization start signal; marking the monitored signal as a filter parameter to be optimized, and the quantum-inspired parameter optimization model comprises the filter parameter to be optimized and a quantum-inspired parameter optimization unit.

[0013] Further, the signal triggering analysis of the signal auxiliary information and the multi-band original spectrum signal set is performed, and the specific analysis content is as follows:

[0014] According to the signal auxiliary information, the current corresponding signal noise data and noise base data are obtained, and the signal noise data and noise base data are analyzed to obtain the signal noise abnormal signal and the noise instability signal; based on the multi-band original spectrum signal set, the current corresponding actual band list, spectrum intensity data and multi-band original signal data set are obtained, and the actual band list, spectrum intensity data and multi-band original signal data set are analyzed to obtain the band missing signal, spectrum abnormal signal and band redundancy signal;

[0015] The current output corresponding signal noise abnormal signal, noise instability signal, band missing signal, spectrum abnormal signal and band redundancy signal are obtained, and the signals are monitored through a pre-set signal set unit, and when any signal is monitored, an optimization start signal is generated, that is, when one or more signals are detected, the optimization start signal is generated.

[0016] Further, the signal noise data and noise base data are analyzed, and the specific analysis content is as follows:

[0017] According to the signal-to-noise data, the signal intensity and signal noise corresponding to each wave band are obtained, the actual signal intensity corresponding to each detection time point in a preset unit analysis period is obtained based on the signal intensity, the signal intensity of each detection time point is calculated to obtain the mean value of the signal intensity, the noise standard deviation of each wave band is obtained according to the signal noise, the original classical signal of each wave band is obtained based on the database, and the quantum state probability concentration is obtained through quantization mapping; the mean value of the signal intensity, the noise standard deviation and the quantum state probability concentration corresponding to each wave band are substituted into the quantum probability distribution bilateral filtering fusion formula The signal-to-noise ratio SNR corresponding to the i-th wave band is calculated i ; wherein P xz,i ,σ noise,i respectively represent the mean value of the signal intensity, the quantum state probability concentration and the noise standard deviation corresponding to the i-th wave band; η xz represents a preset quantum probability weight coefficient, and the value is 0.05-0.2; B(σ kj ,σ hd ) is a bilateral filtering factor; the signal-to-noise ratio corresponding to each wave band is compared with the corresponding preset signal-to-noise ratio threshold value, when the signal-to-noise ratio is greater than the preset signal-to-noise ratio threshold value, the corresponding wave band is marked as a signal-to-noise abnormal wave band, the ratio of the number of the signal-to-noise abnormal wave band to the total number of wave bands is calculated to obtain an abnormal wave band ratio, and when the preset signal-to-noise signal output unit detects that the abnormal wave band ratio is greater than the corresponding preset abnormal wave band ratio threshold value, an abnormal signal-to-noise signal is output.

[0018] Based on the noise base data, the actual noise mean value, quantum noise data and driving voltage data of each wave band are obtained, and the variance of the actual noise mean value is calculated; based on the quantum noise data, the quantum noise amplitude corresponding to each time point is obtained, and the quantum noise autocorrelation coefficient C t is calculated through the quantum adjacent time autocorrelation coefficient calculation formula ; wherein, T is a preset time window length corresponding to autocorrelation calculation, t is a corresponding value time; Zx t represents the quantum noise amplitude at time t, represents the variance of the quantum noise; based on the driving voltage data, the driving voltage is obtained, and the liquid crystal voltage entropy H V is calculated by using the discrete form formula of Shannon entropy ; wherein, the value of the driving voltage is [0, V max ], and P(V) is the probability distribution of the driving voltage; the quantum noise autocorrelation coefficient and the liquid crystal voltage entropy are substituted into the liquid crystal hysteresis-quantum fluctuation entropy fusion algorithm NS=exp(-(H V ×C t) and a noise stability threshold value is obtained according to a pre-design, and when the noise stability value is less than the noise stability threshold value, a noise instability signal is output through a preset noise signal output unit.

[0019] Further, the actual waveband list, the spectral intensity data and the multi-waveband original signal data set are analyzed, specifically:

[0020] The actual waveband list is compared with a corresponding preset waveband target list to obtain a missing waveband list, and a missing waveband index set is constructed according to the missing waveband list, and is denoted as Φ que ; the actual waveband list is reconstructed based on the GNN to obtain a waveband feature vector of the kth waveband, and is denoted as The output kth waveband is analyzed in association with other wavebands by using the GNN to output an association strength, and is denoted as GNN(G gl , k); the fusion graph neural network and the pre-similarity model are used to calculate a waveband integrity anomaly quantity BI; wherein, α k represents a preset importance weight of the kth waveband; represents a standard waveband feature vector corresponding to the kth waveband; when the waveband integrity anomaly quantity is greater than a preset upper limit value of the integrity anomaly, a waveband missing signal is output through a preset integrity signal unit.

[0021] A multispectral image with a preset sampling requirement is obtained based on the spectral intensity data, an actual spectral intensity histogram of the multispectral image is extracted by using a generative adversarial network GAN, a normal dynamic range histogram entropy corresponding to the spectral image is generated, and the actual spectral intensity histogram and the normal dynamic range histogram entropy are discriminated based on a discriminator of the generative adversarial network GAN to output a corresponding actual signal histogram entropy; pixels corresponding to the multispectral image are identified to obtain an invalid pixel proportion value; a fusion GAN and a histogram entropy model DRC=|XH real -XH GAN |·xs w ·D GAN are used to calculate a spectral anomaly quantity DRC; wherein, XH real , XH GAN respectively represent the actual signal histogram entropy and the normal dynamic range histogram entropy; xs w represents the invalid pixel proportion value; D GAN is an abnormal confidence degree output by the GAN discriminator, and the value is 0-1; a pre-set spectral anomaly reference quantity is obtained, a difference between the current spectral anomaly quantity and the spectral anomaly reference quantity is calculated to obtain an anomaly quantity fluctuation value, and when the anomaly quantity fluctuation value is greater than a preset fluctuation threshold value, a spectral anomaly signal is output through a preset spectral signal unit.

[0022] Based on the multi-band raw signal dataset, two bands are randomly selected as analysis bands and denoted as B. i B j Obtain the dot product of the band vectors and the magnitude of the band vectors corresponding to the two analysis bands; use the cosine similarity formula... The cosine similarity CS(B) between the two analysis bands was calculated. i B j ); where xB i xB j These are represented as the dot product of the band vectors corresponding to the two analysis wavelengths, |xB i |、|xB j | These represent the band vector magnitudes corresponding to the two analysis wavelengths; calculated using the balanced mutual information formula. The statistical correlation values ​​of the two analysis bands were calculated; the cosine similarity and statistical correlation values ​​of the two analysis bands were substituted into BR=1-[τ×MI(B i B j )+(1-τ)×|CS(B i B j The band redundancy BR is obtained by comprehensive calculation; where τ represents the preset weighting coefficient with a value of 0.6; when the band redundancy is less than the preset redundancy threshold, the band redundancy signal is output through the band redundancy signal unit.

[0023] S3: Obtain the quantum-inspired parameter optimization model and the multi-band original spectral signal set, substitute the multi-band original spectral signal set into the quantum-inspired parameter optimization model to optimize and output the optimal filter parameters;

[0024] S4: Obtain the optimal filtering parameters and input the optimal filtering parameters into the pre-deployed multispectral image generation model, and output the initial multispectral image through the multispectral image generation model.

[0025] S5: Acquire the initial multispectral image and perform image parameter acquisition on the multispectral image to output image quality parameters. Perform quality analysis on the image quality parameters to obtain the image quality status. When the image quality status corresponds to an excellent state, mark the initial multispectral image as the final multispectral image; otherwise, when the quality status corresponds to an undesirable state, feed back to step S2 for optimization generation, and count the number of optimization generation corresponding to the initial multispectral image. When the number of optimization generation exceeds the preset optimization number threshold, generate an optimization error report and provide feedback based on the optimization error report.

[0026] As a further improvement of the present invention, a quality analysis is performed on the image quality state, and the specific analysis content is as follows:

[0027] The image quality parameters include a spectral angle distance, a local window parameter and residual noise energy; when the spectral angle distance is greater than a preset spectral angle threshold, a part of the spectral angle distance exceeding the spectral angle distance is marked as a spectral angle distance outlier; a local window of a preset size is obtained based on the local window parameter, a spatial variance corresponding to the local window is calculated to obtain a variance deviation together with a spatial variance of a normal region; the residual noise energy is divided into a plurality of noise energy intervals, a residual influence coefficient is set for each noise energy interval, the current residual noise energy is matched with each noise energy interval to obtain a corresponding residual influence coefficient, and the residual noise energy is multiplied by the residual influence coefficient to obtain a residual noise outlier.

[0028] The spectral angle distance outlier, the variance deviation and the residual noise outlier are normalized and the values thereof are taken, a coupling calculation model is used to calculate a quality abnormality index GI. X c The SA, LV and RN represent the normalized values of the spectral angle distance outlier, the variance deviation and the residual noise outlier respectively. c The χ1, χ2 and χ3 represent preset weight factors corresponding to the SA, LV and RN. GI The β is a preset cross-influence factor, and the value thereof is 0.1-0.3; when the quality abnormality index is greater than a preset quality abnormality threshold, the image quality state is generated as a poor state, otherwise, the image quality state is an excellent state.

[0029] Compared with the prior art, the technical scheme provided by the present application has the beneficial effects that:

[0030] 1. The present application combines the quantum heuristic optimization structure with the liquid crystal filter hardware, learns from the quantum superposition and parallelism characteristics to improve the filter parameter optimization speed and precision, simultaneously uses the electrically controllable adjustable characteristics of the liquid crystal filter to realize the dynamic switching of the spectral band, solves the problems of poor adjustability, low integration and low high-dimensional data processing efficiency of the traditional mechanical filtering structure, and provides high-quality original signals and efficient calculation support for multispectral image generation.

[0031] 2. The present application constructs an image quality analysis and feedback mechanism, evaluates the image quality based on the spectral angle distance, residual noise energy and other parameters, feeds back the poor state image to the optimization link for reprocessing, sets a threshold of the optimization number to avoid invalid cycles, effectively guarantees the quality stability of the output image, and solves the problems of spectral distortion, noise and detail balance. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. The following drawings are not drawn in scale, and the emphasis is on showing the main idea of the present application.

[0033] Fig. 1 Flow chart of the method of the present application;

[0034] Fig. 2 Principle block diagram of the parameter optimization triggering unit of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0036] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figs. 1-2 In the embodiments of the present application, one embodiment of the liquid crystal filtering multi-spectral image generation method based on quantum inspiration includes:

[0037] S1, pre-treatment dynamic regulation: obtaining a hardware module performance report corresponding to a pre-deployed liquid crystal filtering hardware module, and obtaining a multi-band original spectral signal set and signal accessory information through original spectral signal acquisition by the liquid crystal filtering hardware module; the multi-band original spectral signal set includes a waveband list, spectral intensity data, and a multi-band original signal data set; the signal accessory information includes signal-to-noise data and noise floor data of each waveband;

[0038] The liquid crystal filtering hardware module is a pre-deployed integrated structure containing a liquid crystal filtering unit, a driving circuit, and an optical coupling component; the hardware module performance report includes but is not limited to a dynamically switchable spectral waveband list, a waveband switching response time, an integrated volume parameter, and waveband transmittance efficiency data.

[0039] S2, quantum inspiration filtering parameter optimization model construction: obtaining the signal accessory information and the multi-band original spectral signal set based on a pre-deployed quantum inspiration filtering optimization structure, and performing parameter optimization analysis on the signal accessory information and the multi-band original spectral signal set to output a quantum inspiration parameter optimization model.

[0040] The parameter optimization analysis on the signal accessory information and the multi-band original spectral signal set includes the following specific analysis contents:

[0041] The quantum inspiration filtering optimization structure includes a parameter optimization triggering unit and a quantum inspiration filtering optimization model.

[0042] The parameter optimization triggering unit is configured to perform signal triggering analysis on the signal auxiliary information and the multi-band original spectrum signal set to obtain an optimized starting signal.

[0043] The quantum-inspired filter optimization model is configured to monitor the optimized starting signal through a preset starting signal monitor, and when the optimized starting signal is monitored, the quantum-inspired filter optimization model performs output of a quantum-inspired parameter optimization model on the signal auxiliary information and the multi-band original spectrum signal set. The quantum-inspired filter optimization model comprises a parameter encoding module, a fitness function expression, and a quantum evolution rule.

[0044] The quantum-inspired parameter optimization model specifically comprises: obtaining a monitored signal of a signal-to-noise abnormality, a noise instability, a band absence, a spectrum abnormality, and a band redundancy corresponding to the optimized starting signal; and marking the monitored signal as a filter parameter to be optimized, and the quantum-inspired parameter optimization model comprises the filter parameter to be optimized and a quantum-inspired parameter optimization unit.

[0045] The signal triggering analysis on the signal auxiliary information and the multi-band original spectrum signal set specifically comprises the following steps:

[0046] According to the signal auxiliary information, current corresponding signal-to-noise data and noise base data are obtained; according to the signal-to-noise data, signal intensity and signal noise corresponding to each band are obtained, actual signal intensity corresponding to each detection time point in a preset unit analysis period is obtained based on the signal intensity, and the signal intensities of the detection time points are calculated to obtain a signal intensity mean value; noise standard deviation of each band is obtained according to the signal noise; original classical signal of each band is obtained based on a database, and quantum state probability concentration is obtained through quantumization mapping; the signal intensity mean value, the noise standard deviation, and the quantum state probability concentration corresponding to each band are substituted into a quantum probability distribution bilateral filter fusion formula to calculate a signal-to-noise ratio SNR of the i-th band i ; wherein P xz,i ,σ noise,i respectively represent the signal intensity mean value, the quantum state probability concentration, and the noise standard deviation corresponding to the i-th band; η xz represents a preset quantum probability weight coefficient, and takes a value of 0.05-0.2, and is used to balance the classical signal and the quantum characteristics; B(σ kj ,σ hd ) is a bilateral filter factor, wherein d kj ,Δhd respectively represent a spatial distance and a gray difference corresponding to the current band, σ kj ,σ hdThe standard deviation of the space / grey domain corresponding to the current wave band is represented; the signal-to-noise ratio corresponding to each wave band is compared with the corresponding preset signal-to-noise ratio threshold value; when the signal-to-noise ratio is greater than the preset signal-to-noise ratio threshold value, the corresponding wave band is marked as a signal-to-noise abnormal wave band; the number of the signal-to-noise abnormal wave band is calculated by ratio to obtain an abnormal wave band ratio; when the preset signal-to-noise signal output unit detects that the abnormal wave band ratio is greater than the corresponding preset abnormal wave band ratio threshold value, a signal-to-noise abnormal signal is output.

[0047] The actual noise mean value, quantum noise data and driving voltage data of each wave band are obtained based on the noise base data, and the variance of the actual noise mean value is calculated; the quantum noise amplitude corresponding to each time is obtained based on the quantum noise data, and the quantum noise autocorrelation coefficient C is calculated by the quantum adjacent time autocorrelation coefficient calculation formula . t ; wherein T is the preset time window length corresponding to the autocorrelation calculation, and t is the corresponding value time; Zx t represents the quantum noise amplitude at time t, represents the variance of the quantum noise; the driving voltage is obtained based on the driving voltage data, and the liquid crystal voltage entropy H is calculated by the discrete form formula of Shannon entropy . V ; wherein the value of the driving voltage is [0, V max ], P(V) is the probability distribution of the occurrence of the driving voltage, which is represented by the discrete probability mass function, and satisfies ∑P(V)=1, that is, the sum of all voltage probabilities is 1; the quantum noise autocorrelation coefficient and the liquid crystal voltage entropy are substituted into the liquid crystal hysteresis-quantum fluctuation entropy fusion algorithm NS=exp(-(H V ×C t ) to calculate the noise stability value NS, and the noise stability threshold value given in advance is obtained; when the noise stability value is less than the noise stability threshold value, a noise instability signal is output through the preset noise signal output unit.

[0048] The actual wave band list, spectral intensity data and multi-wave band original signal data set corresponding to the current are obtained based on the multi-wave band original spectral signal set; the actual wave band list is compared with the corresponding preset wave band target list to obtain a missing wave band list, the missing wave band corresponding to the missing wave band list is constructed into a missing wave band index set, and is denoted as Φ que ; if there is no missing wave band list, the missing wave band index set is empty; the wave band feature vector of the kth wave band is obtained by reconstructing the actual wave band list based on GNN, and is denoted as ; the correlation strength is output by using GNN to associate and analyze the output kth wave band with other wave bands, and is denoted as GNN(G gl , k); the fusion graph neural network and the pre-similarity model The band integrity anomaly amount BI is calculated, wherein a k represents the preset importance weight of the kth band, for example, the red edge band a k is 0.8. represents the standard band feature vector corresponding to the kth band; GNN represents a deep learning model for processing graph structure data, and the core function is to model the topological correlation between bands and reconstruct the missing band features; when the band integrity anomaly amount is greater than the preset upper limit of the integrity anomaly, a band missing signal is output through a preset integrity signal unit.

[0049] Based on the spectral intensity data, a multispectral image with a preset sampling requirement is obtained, an actual spectral intensity histogram of the multispectral image is extracted through a generative adversarial network (GAN), a normal dynamic range histogram entropy corresponding to the spectral image is generated, and the actual spectral intensity histogram and the normal dynamic range histogram entropy are discriminated based on a discriminator of the generative adversarial network (GAN) to output a corresponding actual signal histogram entropy; the pixels corresponding to the multispectral image are identified to obtain an invalid pixel ratio, and the invalid pixels include the sum of overexposed and underexposed pixels; a fusion GAN and a histogram entropy model DRC = |XH real -XH GAN |·xs w ·D GAN The spectral anomaly amount DRC is calculated, wherein XH real , XH GAN respectively represent the actual signal histogram entropy and the normal dynamic range histogram entropy; xs w represents the invalid pixel ratio; D GAN is the abnormal confidence output by the GAN discriminator, and the value is 0-1, and the higher the value, the greater the difference between the actual signal and the normal distribution; a preset spectral anomaly reference amount is obtained, the current spectral anomaly amount is compared with the spectral anomaly reference amount to obtain an anomaly amount fluctuation value, and when the anomaly amount fluctuation value is greater than a preset fluctuation threshold, a spectral anomaly signal is output through a preset spectral signal unit; the GAN is a deep learning model composed of a generator and a discriminator, and data generation and discrimination are realized through the adversarial training of the two.

[0050] Based on a multi-band original signal data set, any two bands are taken as analysis bands, and are respectively denoted as B i and B j The band vector dot product and the band vector length corresponding to the two analysis bands are obtained; the cosine similarity formula is used to calculate the cosine similarity CS(B i , B j ) of the two analysis bands; wherein xB i , xB jrespectively represent the band vector dot product corresponding to two analysis wavelengths, |xB i |xB j | respectively represent the band vector module length corresponding to two analysis wavelengths; the balance mutual information calculation formula The statistical correlation value of the two analysis bands is calculated; the cosine similarity and the statistical correlation value corresponding to the two analysis bands are substituted into BR = 1-[τ×MI(B i , B j )+(1-τ)×|CS(B i , B j )|] to obtain the band redundancy BR; wherein τ represents a preset weight coefficient, and the value is 0.6; when the band redundancy is less than the preset redundancy threshold, the band redundancy signal unit outputs the band redundancy signal.

[0051] The current output corresponding signal-to-noise anomaly signal, noise instability signal, band missing signal, spectral anomaly signal and band redundancy signal are obtained, and the signals are monitored by the pre-set signal set unit. When any signal is monitored, an optimization start signal is generated, that is, when one or more signals are detected, an optimization start signal is generated.

[0052] S3 filter parameter solving: obtain the quantum heuristic parameter optimization model and the multi-band original spectrum signal set, substitute the multi-band original spectrum signal set into the quantum heuristic parameter optimization model, obtain the quantum heuristic parameter optimization unit and the to-be-optimized filter parameter through the quantum heuristic parameter optimization model, and optimize the to-be-optimized filter parameter and the multi-band original spectrum signal set through the quantum heuristic parameter optimization unit to output the optimal filter parameter.

[0053] S4 multi-spectral image generation: obtain the optimal filter parameter and input the optimal filter parameter into the pre-deployed multi-spectral image generation model, convert the optimal filter parameter into an initial multi-spectral image through the pre-deployed multi-spectral image generation model, realize the conversion of the spectral signal to the visual image; utilize the optimal parameter to realize the efficient conversion of the spectral signal to the image, and ensure that the image can accurately reflect the spectral characteristics of each band.

[0054] S5 image optimization output: obtain the initial multi-spectral image and perform image parameter acquisition on the multi-spectral image to output image quality parameters, perform quality analysis on the image quality parameters to obtain an image quality state, and when the image quality state corresponds to an optimal state, mark the initial multi-spectral image as a final multi-spectral image; otherwise, when the image quality state corresponds to a poor state, feedback to step S2 for optimization generation, and count the number of optimization generations corresponding to the initial multi-spectral image. When the number of optimization generations exceeds a preset optimization number threshold, an optimization error report is generated, and feedback is performed based on the optimization error report.

[0055] Quality analysis is performed on the image quality state quality, and specific analysis contents are as follows:

[0056] The image quality parameters include a spectral angle distance, a local window parameter, and residual noise energy; when the spectral angle distance is greater than a preset spectral angle threshold, a part of the spectral angle distance that exceeds the spectral angle distance is marked as a spectral angle distance outlier, the spectral angle distance is an angle between a spectral curve of an initial multi-spectral image and a reference spectrum, and reflects a matching degree of a spectral shape, for example, original high signal-to-noise ratio band data and real ground object spectral library data; a local window of a preset size is obtained based on the local window parameter, a spatial variance corresponding to the local window is calculated to obtain a variance deviation, and the spatial variance is obtained from coupling calculation in three dimensions of brightness, contrast, and structure; the residual noise energy is divided into a plurality of noise energy intervals, a residual influence coefficient is set for each noise energy interval, the current residual noise energy is matched with each noise energy interval to obtain a corresponding residual influence coefficient, and the residual noise energy is multiplied by the residual influence coefficient to obtain a residual noise outlier; the residual influence coefficient is greater than zero, and increases with an increase of a maximum value of the noise energy interval; the residual noise energy represents a noise component of the generated image separated through a high-pass filter, a noise energy proportion is calculated, and a noise suppression effect is evaluated.

[0057] The spectral angle distance outlier, the variance deviation, and the residual noise outlier are normalized and the values are taken, and a coupling calculation model is used to calculate a quality anomaly index GI; wherein SA, LV, and RN represent normalized values of the spectral angle distance outlier, the variance deviation, and the residual noise outlier respectively; χ c ∈(χ1, χ2, χ3) represents preset weight factors corresponding to SA, LV, and RN; β GI is a preset cross-influence factor, and the value is 0.1-0.3, which is used to amplify the synergistic effect of simultaneous abnormality of the three indexes, for example, when the spectral angle distance outlier, the variance deviation, and the residual noise outlier are all abnormal, SA=LV=RN=0.6, the cross-term contribution is approximately 0.1x0.6=0.06, and the overall anomaly index is enhanced; when the quality anomaly index is greater than a preset quality anomaly threshold, the generated image quality state is a poor state, otherwise, the image quality state is an optimal state.

[0058] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for quantum-inspired liquid crystal filtered multispectral image generation, characterized in that, Comprising the following steps: S1: Obtain the hardware module performance report corresponding to the pre-deployed liquid crystal filtering hardware module, obtain the multi-band original spectrum signal set and signal auxiliary information by the liquid crystal filtering hardware module for original spectrum signal acquisition; S2: Obtain the signal auxiliary information and the multi-band original spectrum signal set based on the pre-deployed quantum heuristic filtering optimization structure, and perform parameter optimization analysis on the signal auxiliary information and the multi-band original spectrum signal set to output a quantum heuristic parameter optimization model; S3: Obtain the quantum heuristic parameter optimization model and the multi-band original spectrum signal set, and substitute the multi-band original spectrum signal set into the quantum heuristic parameter optimization model for optimization to output optimal filtering parameters; S4: Obtain the optimal filtering parameters and input the optimal filtering parameters into the pre-deployed multi-spectral image generation model, and output an initial multi-spectral image through the multi-spectral image generation model; S5: Obtain the initial multi-spectral image and perform image parameter acquisition on the multi-spectral image to output image quality parameters, perform quality analysis on the image quality parameters to obtain an image quality state, and when the image quality state corresponds to an optimal state, mark the initial multi-spectral image as a final multi-spectral image; otherwise, when the quality state corresponds to a poor state, feedback to step S2 for optimization generation, and count the number of optimization generations corresponding to the initial multi-spectral image, and when the number of optimization generations exceeds a preset optimization number threshold, generate an optimization error report, and feedback based on the optimization error report.

2. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 1, wherein, The parameter optimization analysis on the signal auxiliary information and the multi-band original spectrum signal set includes the following specific analysis contents: The quantum heuristic filtering optimization structure includes a parameter optimization trigger unit and a quantum heuristic filtering optimization model; The parameter optimization trigger unit is used for signal trigger analysis on the signal auxiliary information and the multi-band original spectrum signal set to obtain an optimization start signal; the quantum heuristic filtering optimization model monitors the optimization start signal through a preset start signal monitor, and when the optimization start signal is monitored, the quantum heuristic filtering optimization model performs optimization on the signal auxiliary information and the multi-band original spectrum signal set to output a quantum heuristic parameter optimization model.

3. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 2, wherein, The signal trigger analysis on the signal auxiliary information and the multi-band original spectrum signal set includes the following specific analysis contents: According to the signal auxiliary information, obtain current corresponding signal-to-noise data and noise base data, and analyze the signal-to-noise data and the noise base data to obtain signal-to-noise abnormal signals and noise instability signals; Based on the multi-band original spectrum signal set, obtain current corresponding actual band list, spectral intensity data and multi-band original signal data set, and analyze the actual band list, spectral intensity data and multi-band original signal data set to obtain band missing signals, spectral abnormal signals and band redundant signals; Obtain the current output corresponding signal-to-noise abnormal signals, noise instability signals, band missing signals, spectral abnormal signals and band redundant signals, and monitor the signals through a pre-set signal set unit, and when any signal is monitored, generate an optimization start signal.

4. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 3, wherein, The analysis on the signal-to-noise data and the noise base data includes the following specific analysis contents: According to the signal-to-noise data, the signal intensity and the signal noise corresponding to each wave band are obtained, the actual signal intensity corresponding to each detection time point in a preset unit analysis period is obtained based on the signal intensity, the signal intensity of each detection time point is calculated to obtain the mean value of the signal intensity; the noise standard deviation of each wave band is obtained according to the signal noise; the original classical signal of each wave band is obtained based on the database, and the quantum state probability concentration is obtained through quantization mapping; the mean value of the signal intensity, the noise standard deviation and the quantum state probability concentration corresponding to each wave band are substituted into the quantum probability distribution bilateral filtering fusion formula to obtain the signal-to-noise ratio SNR corresponding to the i th wave band i ; the signal-to-noise ratio corresponding to each wave band is compared with the corresponding preset signal-to-noise ratio threshold value, when the signal-to-noise ratio is greater than the preset signal-to-noise ratio threshold value, the corresponding wave band is marked as a signal-to-noise abnormal wave band, the ratio of the number of the signal-to-noise abnormal wave band to the total number of wave bands is calculated to obtain an abnormal wave band ratio, and when the preset signal-to-noise signal output unit detects that the abnormal wave band ratio is greater than the corresponding preset abnormal wave band ratio threshold value, an abnormal signal-to-noise signal is output. Based on the noise base data, actual noise mean, quantum noise data and driving voltage data of each waveband are obtained, and the variance of the actual noise mean is calculated; Based on the quantum noise data, the quantum noise amplitude corresponding to each time is obtained, and the quantum noise autocorrelation coefficient is calculated through the quantum adjacent time autocorrelation coefficient calculation formula; The quantum noise autocorrelation coefficient and liquid crystal voltage entropy are substituted into the liquid crystal hysteresis-quantum fluctuation entropy fusion algorithm to calculate the noise stability value, and the noise stability threshold given in advance is obtained. When the noise stability value is less than the noise stability threshold, the noise instability signal is output through the preset noise signal output unit.

5. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 4, wherein, The actual waveband list, spectral intensity data and multi-waveband original signal data set are analyzed, which is specifically: The actual waveband list is compared with the corresponding preset waveband target list to obtain a missing waveband list, and the missing waveband list is constructed into a missing waveband index set. Based on GNN, the waveband feature vector of the kth waveband is reconstructed, and the output kth waveband is associated with other wavebands by GNN to output the correlation strength. The waveband integrity anomaly quantity is calculated by using the fusion graph neural network and the pre-similarity model. When the waveband integrity anomaly quantity is greater than the preset upper limit of the integrity anomaly, the waveband missing signal is output through the preset integrity signal unit. Based on the spectral intensity data, a multi-spectral image with a preset sampling requirement is obtained, the actual spectral intensity histogram of the multi-spectral image is extracted by a generative adversarial network (GAN), and the normal dynamic range histogram entropy corresponding to the spectral image is generated. The actual spectral intensity histogram and the normal dynamic range histogram entropy are discriminated by the discriminator of the generative adversarial network (GAN) to output the corresponding actual signal histogram entropy. The invalid pixel ratio is obtained by identifying the pixels of the multi-spectral image. The spectral anomaly is calculated by using the fusion GAN and the histogram entropy model. The pre-set spectral anomaly reference quantity is obtained, the current spectral anomaly quantity is compared with the spectral anomaly reference quantity, and the anomaly quantity fluctuation value is calculated. When the anomaly quantity fluctuation value is greater than the pre-set fluctuation threshold, the spectral anomaly signal is output through the pre-set spectral signal unit. Based on the multi-waveband original signal data set, two wavebands are selected as analysis wavebands, and the waveband vector dot product and waveband vector length corresponding to the two analysis wavebands are obtained. The cosine similarity of the two analysis wavebands is calculated by the cosine similarity formula. The statistical correlation value of the two analysis wavebands is calculated by the balanced mutual information calculation formula. The waveband redundancy is calculated by comprehensively calculating the cosine similarity and statistical correlation value corresponding to the two analysis wavebands. When the waveband redundancy is less than the pre-set redundancy threshold, the waveband redundancy signal is output through the waveband redundancy signal unit.

6. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 5, wherein, The quantum heuristic parameter optimization model, specifically: including the to-be-optimized filtering parameter and the quantum heuristic parameter optimization unit obtaining the signal noise anomaly signal, the noise instability signal, the waveband missing signal, the spectral anomaly signal and the waveband redundancy signal corresponding to the monitoring of the optimization start signal; the monitored signal is marked as the to-be-optimized filtering parameter.

7. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 1, wherein, The multi-band original spectral signal set comprises a band list, spectral intensity data and a multi-band original signal data set; the signal accessory information comprises signal-to-noise data and noise floor data of each band; The liquid crystal filter hardware module is an integrated structure comprising a liquid crystal filter unit, a driving circuit and an optical coupling component; the hardware module performance report comprises a dynamically switchable spectral band list, a band switching response time, an integrated volume parameter and transmittance efficiency data of each band.

8. The quantum-inspired liquid crystal filtered multispectral image generation method of claim 1, wherein, The quality analysis of the image quality state quality comprises the following specific analysis contents: The image quality parameters comprise a spectral angle distance, a local window parameter and a residual noise energy; when the spectral angle distance is greater than a preset spectral angle threshold, a part of the spectral angle distance exceeding the spectral angle distance is marked as a spectral angle distance abnormal value; a local window of a preset size is obtained based on the local window parameter, a spatial variance of the local window is calculated with a spatial variance of a normal region to obtain a variance deviation; the residual noise energy is divided into a plurality of noise energy intervals, a residual influence coefficient is set for each noise energy interval, a current residual noise energy is matched with each noise energy interval to obtain a corresponding residual influence coefficient, and the residual noise energy is multiplied by the residual influence coefficient to obtain a residual noise abnormal value; The spectral angle distance abnormal value, the variance deviation and the residual noise abnormal value are normalized and their values are taken, a quality abnormality index is calculated by using a coupling calculation model; when the quality abnormality index is greater than a preset quality abnormality threshold, the image quality state is generated as a poor state, otherwise, the image quality state is an excellent state.

Images