Full-spectrum splicing high-resolution imaging system and method for hyperspectral analyzer

Abstract

The invention relates to the technical field of hyperspectral remote sensing spectral analysis, in particular to a full-spectrum splicing high-resolution imaging system and method for a hyperspectral analyzer, and the method comprises the steps: determining a slit switching sequence through an acquisition module; the dynamic collimation module feeds back an aberration result; the adaptive dispersion module carries out wavelength calibration; the high-precision imaging module predicts imaging mirror parameters; the real-time compensation module adopts a Kalman filtering algorithm to predict the defocus compensation amount and executes nanoscale displacement compensation through a piezoelectric ceramic driver; the intelligent splicing module adopts an SIFT feature matching algorithm, a generative adversarial network and a reinforcement learning algorithm to realize image registration and fusion; the cooperative control module generates control signals through Monte Carlo simulation and a Bayesian optimization algorithm to coordinate all the modules. According to the invention, the problems of insufficient spectral resolution, difficulty in data splicing and low real-time measurement efficiency caused by fixed hardware parameters in a wide-spectrum spectral imaging system are solved, and full-spectrum high-resolution imaging is realized.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral remote sensing spectral analysis technology, specifically to a high-resolution imaging system and method for full-spectrum stitching of a hyperspectral analyzer. Background Technology

[0002] The hyperspectral imaging technology employed by the hyperspectral analyzer continuously acquires spectral information of targets in hundreds of narrow bands to form a complete spectral curve, thereby enabling fine spectral analysis of the earth's surface or materials. In remote sensing monitoring, this technology utilizes the differences in spectral characteristics of different materials to accurately identify land cover types and change trends, supporting environmental assessment and disaster early warning. At the same time, in fine spectral analysis scenarios, the high-dimensional data provided by hyperspectral imaging helps reveal the microscopic composition and state of materials, providing a reliable analytical basis for fields such as agriculture, geology, and medicine.

[0003] Existing hyperspectral imaging technologies suffer from the following technical challenges: In broadband spectral imaging systems, hardware parameters such as slit size, dispersive element angle, and optical lens magnification are fixed and cannot be dynamically adapted to remote sensing monitoring scenarios, resulting in insufficient system flexibility. In applications requiring fine spectral analysis, such as monitoring vegetation health, existing instruments struggle to quickly adjust to capture high-resolution data across the entire ultraviolet to near-infrared spectrum, leading to reduced spectral resolution, misalignment of stitched images from different narrowbands, and real-time measurement delays, thus affecting the accurate identification and analysis efficiency of environmental changes. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a high-resolution imaging system and method for full-spectrum stitching of hyperspectral analyzers. This invention solves the technical problems of insufficient spectral resolution, difficulty in data stitching, and low real-time measurement efficiency in remote sensing monitoring caused by fixed hardware parameters in broadband spectral imaging systems that cannot be dynamically adapted.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the hyperspectral analyzer full-spectrum stitching high-resolution imaging system provided by the present invention includes an intelligent acquisition module, a dynamic collimation module, an adaptive dispersion module, a high-precision imaging module, a real-time compensation module, an intelligent stitching module, and a collaborative control module. The intelligent acquisition module acquires broadband ambient light signals, dynamically adjusts the slit width, and outputs a collimated beam to the dynamic collimation module. The dynamic collimation module performs optical path shaping on the collimated beam and outputs a parallel light signal to the adaptive dispersion module; the dynamic collimation module also feeds back the collimation index, which characterizes the parallelism of the light signal, to the intelligent acquisition module; The adaptive dispersion module decomposes the parallel light signal into spectral components and outputs them to the high-precision imaging module; the adaptive dispersion module also feeds back data characterizing the wavelength separation accuracy to the dynamic collimation module; The high-precision imaging module focuses the spectral components to generate preliminary spectral images corresponding to different narrow bands and outputs them to the real-time compensation module; the high-precision imaging module also feeds back the sharpness index, which characterizes the image quality, to the adaptive dispersion module. The real-time compensation module performs focal plane compensation on the preliminary spectral image and outputs a narrowband stable image to the intelligent stitching module; the real-time compensation module also feeds back data characterizing the compensation effect to the high-precision imaging module; The intelligent stitching module registers and fuses multiple narrowband stable images to output a full-spectrum high-resolution spectral image; the intelligent stitching module also feeds back the index characterizing the image alignment accuracy to the real-time compensation module; The collaborative control module communicates bidirectionally with the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. It receives user requirements and module status data, and outputs control signals to the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module.

[0006] Furthermore, in the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention, the intelligent acquisition module integrates a photoelectric sensor; the photoelectric sensor monitors the intensity distribution of incident light, and the intelligent acquisition module drives the slit width adjustment according to the monitored light intensity distribution; the intelligent acquisition module runs Monte Carlo optical simulation before slit switching and determines the slit switching sequence based on the simulation results.

[0007] Furthermore, in the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention, the dynamic collimation module is configured to: calculate the collimation mirror parameters using a particle swarm optimization algorithm; integrate an aberration detection algorithm based on the interference principle for detecting aberrations; and feed back the detected aberration results as a collimation index to the intelligent acquisition module.

[0008] Furthermore, in the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention, the adaptive dispersion module is configured to: optimize the rotation angle of the dispersive element using a genetic algorithm; the adaptive dispersion module queries the grating constant database to match the current spectral range, and performs wavelength calibration based on the query results and the optimized rotation angle to generate wavelength separation accuracy data as wavelength calibration data; the adaptive dispersion module shares the wavelength calibration data with the high-precision imaging module.

[0009] Furthermore, in the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention, the high-precision imaging module is configured as follows: Zernike polynomials were applied to calculate wavefront aberrations, and convolutional neural networks were used to predict imaging mirror parameters. The high-precision imaging module evaluates and generates a sharpness index as an image sharpness index based on the aberration calculation results and the predicted imaging mirror parameters; the high-precision imaging module sends the image sharpness index to the adaptive dispersion module and the real-time compensation module.

[0010] Furthermore, in the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention, the real-time compensation module is configured to: use a Kalman filter algorithm to predict the defocus compensation amount, and use a piezoelectric ceramic actuator to perform the nanometer-level displacement corresponding to the defocus compensation amount for focal plane compensation; the real-time compensation module generates data of the compensation effect after compensation as an image quality evaluation index, and feeds back the image quality evaluation index to the high-precision imaging module.

[0011] Furthermore, in the hyperspectral analyzer full-spectrum stitching high-resolution imaging system of the present invention, the intelligent stitching module is configured to: register multiple narrowband stable images using the SIFT feature matching algorithm; fuse the registered images using a generative adversarial network; optimize the parameters of image registration and fusion using a reinforcement learning algorithm; and calculate a stitching progress index based on the processing progress of image registration and fusion, and feed back the stitching progress index, which represents the processing progress, to the collaborative control module.

[0012] Furthermore, the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention further includes: The collaborative control module receives module status data from the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. The collaborative control module uses Monte Carlo simulation to randomly sample system parameters to simulate module interaction behavior, pre-simulate potential state changes during full-spectrum imaging, and employs a Bayesian optimization algorithm to model the parameter space through a Gaussian process. Based on the acquired state data and simulation results, it updates the probability distribution and converges to the optimal parameter combination through an expectation boosting function to generate optimized parameters. These optimized parameters define the content of the control signal. The collaborative control module generates the control signal by mapping parameters according to the optimized parameters and a pre-built system performance lookup table, and sends the control signal to the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module.

[0013] Furthermore, the hyperspectral analyzer full-spectrum stitched high-resolution imaging system of the present invention further includes: The intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module each include an independent microprocessor for executing the functions of each module, including photoelectric sensor data processing, optimization algorithm calculation, and actuator driving. The collaborative control module includes a main processor for coordinating control and running system-level optimization algorithms. The main processor and the microprocessors of each module exchange data through a master-slave communication protocol using a time-division multiplexing mechanism. The main processor sends global control commands, and the microprocessors return module status information, realizing command-response data synchronization to complete the bidirectional communication.

[0014] Secondly, the present invention provides a method for full-spectrum stitching high-resolution imaging of a hyperspectral analyzer, applied to the aforementioned full-spectrum stitching high-resolution imaging system of a hyperspectral analyzer, comprising: Step 1: Acquire broadband ambient light signals and dynamically adjust the slit width to output a collimated beam; Step 2: Receive the collimated beam, perform optical path shaping on the collimated beam, output a parallel light signal, and feed back the collimation index to Step 1; Step 3: Receive the parallel light signal, decompose the parallel light signal into spectral components, output the spectral components, and feed back the wavelength separation accuracy data to Step 2; Step 4: Receive spectral components, focus the spectral components to generate preliminary spectral images corresponding to different narrowbands, output the preliminary spectral images, and feed back the sharpness index to Step 3; Step 5: Receive the preliminary spectral image, perform focal plane compensation on the preliminary spectral image, output a narrowband stable image, and feed back the compensation effect data to Step 4; Step 6: Receive the narrowband stabilized image, register and fuse multiple narrowband stabilized images, output a full-spectrum high-resolution spectral image, and feed back the image alignment accuracy index to Step 5; Step 7: Interact with Steps 1 to 6, receive the user-inputted requirements parameters and the status data generated in Steps 1 to 6, and send control signals to Steps 1 to 6.

[0015] Beneficial effects of this invention; This invention effectively solves the technical problems of insufficient spectral resolution, difficult data stitching, and low real-time measurement efficiency caused by fixed hardware parameters that cannot be dynamically adapted in broadband spectral imaging systems through modular design and intelligent algorithm collaboration. The intelligent acquisition module monitors the intensity distribution of incident light in real time based on photoelectric sensors and drives the dynamic adjustment of the slit width. Combined with Monte Carlo optical simulation to pre-determine the slit switching sequence, it achieves adaptive optimization of the optical signal acquisition stage. The dynamic collimation module uses a particle swarm optimization algorithm to calculate the collimator parameters, integrates an aberration detection algorithm based on the interference principle, and feeds back the aberration results to the intelligent acquisition module to form a closed-loop adjustment of the optical path shaping to improve beam parallelism. The adaptive dispersion module optimizes the rotation angle of the dispersion element through a genetic algorithm, queries the grating constant database to match the spectral range, performs wavelength calibration, and generates wavelength calibration data to be shared with the high-precision imaging module to achieve wavelength separation accuracy. The high-precision imaging module applies Zernike polynomials to calculate wavefront aberrations and combines convolutional neural networks to predict imaging mirror parameters. The system evaluates the sharpness of generated images and feeds the results back to the adaptive dispersion module and the real-time compensation module to optimize focus quality. The real-time compensation module uses a Kalman filter algorithm to predict the defocus compensation amount, drives a piezoelectric ceramic actuator to perform nanometer-level displacement compensation, and feeds the compensation effect data back to the high-precision imaging module to maintain focal plane stability. The intelligent stitching module uses the SIFT feature matching algorithm to achieve accurate registration of multiple narrowband stable images, uses a generative adversarial network for fusion processing, and dynamically optimizes the registration and fusion parameters through a reinforcement learning algorithm. At the same time, it feeds the stitching progress index back to the collaborative control module to solve the image stitching misalignment problem. The collaborative control module receives the status data of each module, runs a Monte Carlo simulation to preview the system behavior, generates optimization parameters using a Bayesian optimization algorithm, and distributes control signals. It coordinates global resources through a master-slave communication protocol. The synergistic effect of the above technical features enables the system to achieve dynamic hardware parameter adaptation in remote sensing monitoring scenarios, significantly improve spectral resolution, eliminate data stitching difficulties, and improve real-time measurement efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a flowchart of a high-resolution imaging method using a hyperspectral analyzer with full-spectrum stitching.

[0018] Figure 2 This is a schematic diagram of the system imaging principle provided in an embodiment of the present invention.

[0019] Figure 3 The discrete slit adjustment principle provided in the embodiments of the present invention.

[0020] Figure 4 The continuous slit adjustment principle provided in the embodiments of the present invention.

[0021] Figure 5 This is a schematic diagram of the dispersion principle of a rotating grating provided in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram illustrating the dispersion principle of a rotating prism provided in an embodiment of the present invention.

[0023] Figure 7 A schematic diagram illustrating the optical parameter switching principle provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0025] In a first aspect, the hyperspectral analyzer full-spectrum stitching high-resolution imaging system provided by the present invention includes an intelligent acquisition module, a dynamic collimation module, an adaptive dispersion module, a high-precision imaging module, a real-time compensation module, an intelligent stitching module, and a collaborative control module. The intelligent acquisition module acquires broadband ambient light signals, dynamically adjusts the slit width, and outputs a collimated beam to the dynamic collimation module. The dynamic collimation module performs optical path shaping on the collimated beam and outputs a parallel light signal to the adaptive dispersion module; the dynamic collimation module also feeds back the collimation index, which characterizes the parallelism of the light signal, to the intelligent acquisition module; The adaptive dispersion module decomposes the parallel light signal into spectral components and outputs them to the high-precision imaging module; the adaptive dispersion module also feeds back data characterizing the wavelength separation accuracy to the dynamic collimation module; The high-precision imaging module focuses the spectral components to generate preliminary spectral images corresponding to different narrow bands and outputs them to the real-time compensation module; the high-precision imaging module also feeds back the sharpness index, which characterizes the image quality, to the adaptive dispersion module. The real-time compensation module performs focal plane compensation on the preliminary spectral image and outputs a narrowband stable image to the intelligent stitching module; the real-time compensation module also feeds back data characterizing the compensation effect to the high-precision imaging module; The intelligent stitching module registers and fuses multiple narrowband stable images to output a full-spectrum high-resolution spectral image; the intelligent stitching module also feeds back the index characterizing the image alignment accuracy to the real-time compensation module; The collaborative control module communicates bidirectionally with the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. It receives user requirements and module status data, and outputs control signals to the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module.

[0026] The hyperspectral analyzer's full-spectrum stitched high-resolution imaging system achieves broadband ambient light signal acquisition and processing through modular design. The intelligent acquisition module integrates photoelectric sensors to continuously monitor the incident light intensity distribution, driving a dynamic slit width adjustment mechanism based on the monitoring data. Before slit switching, Monte Carlo optical simulations are run to simulate the propagation behavior of light signals through different slit widths. The simulation results are used to determine the slit switching sequence, thereby serializing the switching process to optimize the optical path input. Photoelectric sensor monitoring provides real-time ambient light feedback, slit adjustment response feedback data, and simulation steps to preview the switching effect, logically forming a closed-loop adaptive control for the acquisition process. The adjusted slit outputs a collimated beam to the dynamic collimation module.

[0027] After receiving the collimated beam, the dynamic collimation module uses a particle swarm optimization algorithm to iteratively search for the collimator lens parameter combination to minimize the beam divergence angle. The module integrates an aberration detection algorithm based on interferometry principles to analyze beam parallelism deviation in real time and detect aberration results. The dynamic collimation module feeds back the aberration results as a collimation index to the intelligent acquisition module for coordinated adjustment of slit parameters. The optimization algorithm dynamically calibrates the collimator lens, aberration detection provides beam quality assessment, and the feedback mechanism connects to the preceding modules, logically achieving real-time optimization of the optical path shaping. The shaped beam outputs a parallel light signal to the adaptive dispersion module.

[0028] The adaptive dispersion module decomposes the parallel light signal into spectral components and uses a genetic algorithm to optimize the rotation angle of the dispersive element to enhance wavelength separation linearity. The module queries a grating constant database to match the current spectral imaging range, and performs wavelength calibration by combining the optimized angle and database information, generating wavelength calibration data. The adaptive dispersion module shares the wavelength calibration data with the high-precision imaging module and feeds back data characterizing wavelength separation accuracy to the dynamic collimation module, achieving dynamic calibration of wavelength separation accuracy. Algorithm optimization improves dispersion accuracy, the database ensures accurate parameter matching, the calibration steps establish a wavelength reference, and the sharing mechanism promotes collaborative work in subsequent imaging stages.

[0029] After receiving the spectral components, the high-precision imaging module uses Zernike polynomials to calculate wavefront aberrations and quantify the degree of optical system distortion. The module employs a convolutional neural network to learn from historical imaging data and predict imaging mirror parameter settings. Based on the aberration calculation results and predicted parameters, the high-precision imaging module evaluates and generates an image sharpness index, which is then sent to the adaptive dispersion module and the real-time compensation module. Aberration analysis diagnoses focusing defects, the neural network optimizes imaging conditions, the evaluation step monitors image quality, and index sharing enables a cross-module quality feedback loop. The module focuses the spectral components to generate preliminary spectral images corresponding to different narrowbands and outputs them to the real-time compensation module.

[0030] The real-time compensation module performs focal plane compensation on the preliminary spectral image, employing a Kalman filter algorithm to process historical defocus data sequences and predict the amount of defocus compensation at future moments. The module uses a piezoelectric ceramic actuator to receive compensation commands and execute precise nanometer-level displacement adjustments, driving changes in the focal plane position to compensate for defocus. After the compensation operation is complete, the real-time compensation module generates compensation effect data as an image quality evaluation index and feeds this index back to the high-precision imaging module, forming a closed-loop logic for focal plane stability control. The compensated image is then output as a narrowband stabilized image to the intelligent stitching module.

[0031] The intelligent stitching module receives multiple narrowband stable images and uses the SIFT feature matching algorithm to extract scale-invariant feature transform descriptors from the images, achieving accurate registration of key point correspondences between the images. The registered images are then input into a generative adversarial network for fusion processing. The generator network synthesizes a seamless stitched image, while the discriminator network evaluates image authenticity to optimize fusion quality. The module employs a reinforcement learning algorithm to dynamically adjust the registration threshold and fusion parameters based on the fusion results, iteratively optimizing stitching accuracy through a reward mechanism. The intelligent stitching module feeds back indicators representing image alignment accuracy to the real-time compensation module and stitching progress indicators to the collaborative control module, maintaining the continuity of the system workflow. The fused image outputs a full-spectrum high-resolution spectral image.

[0032] The collaborative control module communicates bidirectionally with all modules, receiving user-inputted parameters and status data from the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. The module runs Monte Carlo simulations, randomly sampling system parameter combinations to simulate module interaction behavior and rehearse potential state changes during full-spectrum imaging. Simulation results are combined with status data and input into a Bayesian optimization algorithm. This algorithm models the parameter space using Gaussian processes, efficiently searches for the optimal combination of control parameters, and generates optimized parameters. The collaborative control module maps these optimized parameters to a pre-built system performance lookup table, generates control signals, and distributes these signals to the microprocessors of each module. Each of the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module includes an independent microprocessor to execute the module's functions, including photoelectric sensor data processing, optimization algorithm calculations, and actuator driving. The collaborative control module includes a main processor for coordinating control and running system-level optimization algorithms. The main processor and the microprocessor exchange data through a master-slave communication protocol using a time-division multiplexing mechanism. The main processor sends global control commands, and the microprocessor returns module status information, realizing command-response data synchronization and coordinating global resources.

[0033] This invention's intelligent acquisition module integrates a photoelectric sensor. The photoelectric sensor monitors the incident light intensity distribution in real time, and the intelligent acquisition module drives a dynamic adjustment mechanism for the slit width based on the monitored light intensity distribution data. Before slit switching, the intelligent acquisition module runs Monte Carlo optical simulation to simulate the propagation behavior of light signals through different slit widths. The simulation results are used to determine the slit switching sequence, thereby serializing the switching process to optimize the optical path input. The photoelectric sensor provides real-time feedback on ambient light, slit adjustment response feedback data, and Monte Carlo simulation previews the switching effect, logically forming a closed-loop adaptive control for the acquisition process.

[0034] This invention's dynamic collimation module employs a particle swarm optimization algorithm to iteratively search for collimator lens parameter combinations to minimize the beam divergence angle. The module integrates an aberration detection algorithm based on interferometry principles to analyze beam parallelism deviation in real time and detect aberration results. The detected aberration results are fed back to the intelligent acquisition module as a collimation index for coordinated adjustment of slit parameters. The particle swarm optimization algorithm dynamically calibrates the collimator lens, aberration detection provides beam quality assessment, and the feedback mechanism connects to preceding modules, enabling real-time optimization of optical path shaping.

[0035] This invention's adaptive dispersion module employs a genetic algorithm to optimize the rotation angle of the dispersive element, thereby enhancing wavelength separation linearity. The adaptive dispersion module queries a grating constant database to match the current spectral imaging range, and combines the optimized rotation angle with database information to perform wavelength calibration, generating wavelength calibration data. The adaptive dispersion module shares this wavelength calibration data with a high-precision imaging module and feeds back data characterizing wavelength separation accuracy to a dynamic collimation module. The genetic algorithm improves dispersion accuracy, the database ensures accurate parameter matching, wavelength calibration establishes a wavelength benchmark, and the sharing mechanism promotes collaborative work in subsequent imaging stages.

[0036] This invention's high-precision imaging module applies Zernike polynomials for wavefront aberration calculation. Zernike polynomials decompose wavefront aberrations to quantify the degree of optical system distortion. The module uses a convolutional neural network to learn historical imaging data and predict imaging mirror parameter settings. Based on the aberration calculation results and predicted imaging mirror parameters, the high-precision imaging module evaluates and generates a sharpness index as an image sharpness indicator. Aberration analysis diagnoses focusing defects, the neural network optimizes imaging conditions, and the evaluation step monitors image quality. The high-precision imaging module sends the image sharpness index to the adaptive dispersion module and the real-time compensation module. Index sharing enables cross-module quality feedback loops to optimize the focusing process.

[0037] This invention's real-time compensation module employs a Kalman filter algorithm to process historical defocus data sequences, predicting the defocus compensation amount for future moments. The Kalman filter algorithm iteratively updates the predicted values ​​based on a state-space model to reduce the impact of noise. The real-time compensation module uses a piezoelectric ceramic actuator to receive compensation commands and execute precise nanometer-level displacement adjustments, driving changes in the focal plane position to compensate for defocusing. The real-time compensation module generates compensation effect data as an image quality evaluation index, assessing image stability after the compensation operation is complete. The real-time compensation module feeds back the image quality evaluation index to the high-precision imaging module, forming a closed-loop logic for focal plane stability control, maintaining image quality.

[0038] This invention's intelligent stitching module employs the SIFT feature matching algorithm to register multiple narrowband stable images. The SIFT algorithm extracts scale-invariant feature transform descriptors from the images, achieving accurate registration of key point correspondences between images. The intelligent stitching module uses a generative adversarial network (GAN) to fuse the registered images. The generator network synthesizes a seamless stitched image, while the discriminator network evaluates image authenticity to optimize fusion quality. The intelligent stitching module uses a reinforcement learning algorithm to optimize image registration and fusion parameters. Based on the fusion results, the reinforcement learning algorithm dynamically adjusts the registration threshold and fusion parameters, iteratively optimizing stitching accuracy through a reward mechanism. The intelligent stitching module calculates a stitching progress index based on the processing progress of image registration and fusion, and feeds this index back to the collaborative control module to maintain the continuity of the system workflow.

[0039] This invention's collaborative control module receives module status data from an intelligent acquisition module, a dynamic collimation module, an adaptive dispersion module, a high-precision imaging module, a real-time compensation module, and an intelligent stitching module. It employs Monte Carlo simulation with random sampling of system parameter combinations to simulate module interaction behavior and pre-simulate potential state changes during full-spectrum imaging. The simulation results and state data are combined and input into a Bayesian optimization algorithm. The Bayesian optimization algorithm models the parameter space using a Gaussian process, updates the probability distribution based on the acquired state data and simulation results, and converges to the optimal parameter combination through an expectation boosting function to generate optimized parameters. These optimized parameters define the content of the control signal. The collaborative control module maps the optimized parameters to a pre-built system performance lookup table to generate the control signal and sends it to each module, achieving dynamic prediction and optimized control of system behavior.

[0040] The intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module of this invention all include independent microprocessors. The microprocessors are used to execute the functions of each module, including photoelectric sensor data processing, optimization algorithm calculation, and actuator driving. The collaborative control module includes a main processor, which is used to coordinate control and run system-level optimization algorithms. The main processor and the microprocessors of each module exchange data through a master-slave communication protocol using a time-division multiplexing mechanism. The main processor sends global control commands, and the microprocessors return module status information, realizing command-response data synchronization to complete bidirectional communication and achieve efficient collaboration of hardware resources.

[0041] Secondly, please refer to Figure 1 This invention provides a method for full-spectrum stitching high-resolution imaging of a hyperspectral analyzer, applied to the full-spectrum stitching high-resolution imaging system of the hyperspectral analyzer as described above, comprising: Step 1: Acquire broadband ambient light signals and dynamically adjust the slit width to output a collimated beam; Step 2: Receive the collimated beam, perform optical path shaping on the collimated beam, output a parallel light signal, and feed back the collimation index to Step 1; Step 3: Receive the parallel light signal, decompose the parallel light signal into spectral components, output the spectral components, and feed back the wavelength separation accuracy data to Step 2; Step 4: Receive spectral components, focus the spectral components to generate preliminary spectral images corresponding to different narrowbands, output the preliminary spectral images, and feed back the sharpness index to Step 3; Step 5: Receive the preliminary spectral image, perform focal plane compensation on the preliminary spectral image, output a narrowband stable image, and feed back the compensation effect data to Step 4; Step 6: Receive the narrowband stabilized image, register and fuse multiple narrowband stabilized images, output a full-spectrum high-resolution spectral image, and feed back the image alignment accuracy index to Step 5; Step 7: Interact with Steps 1 to 6, receive the user-inputted requirements parameters and the status data generated in Steps 1 to 6, and send control signals to Steps 1 to 6.

[0042] In step 1 of this invention, the intelligent acquisition module integrates a photoelectric sensor to monitor the incident light intensity distribution in real time. Based on the monitored light intensity distribution data, a dynamic adjustment mechanism for the slit width is driven. Before slit switching, a Monte Carlo optical simulation is run to simulate the propagation behavior of the light signal through different slit widths. The simulation results are used to determine the slit switching sequence, thereby serializing the switching process to optimize the optical path input. The photoelectric sensor monitors and provides real-time feedback of ambient light, slit adjustment response feedback data, and simulation steps to preview the switching effect. Logically, a closed-loop adaptive control of the acquisition link is formed. The adjusted slit outputs a collimated beam to step 2.

[0043] In step 2 of this invention, after receiving the collimated beam through the dynamic collimation module, a particle swarm optimization algorithm is used to iteratively search for the collimating mirror parameter combination to minimize the beam divergence angle. The module integrates an aberration detection algorithm based on the interference principle to analyze the beam parallelism deviation in real time and detect aberration results. The dynamic collimation module feeds back the aberration results as a collimation index to step 1 for collaborative adjustment of slit parameters, optimization algorithm to dynamically calibrate the collimating mirror, aberration detection to provide beam quality assessment, and feedback mechanism to connect to the preceding steps. Logically, this achieves real-time optimization of optical path shaping. The shaped beam outputs a parallel light signal to step 3.

[0044] In step 3 of this invention, the parallel light signal is decomposed into spectral components through an adaptive dispersion module. A genetic algorithm is used to optimize the rotation angle of the dispersive element to enhance the linearity of wavelength separation. The module queries the grating constant database to match the current spectral imaging range. Wavelength calibration is performed by combining the optimized angle and database information to generate wavelength calibration data. The adaptive dispersion module shares the wavelength calibration data to step 4 and feeds back the data characterizing the wavelength separation accuracy to step 2, realizing dynamic calibration of wavelength separation accuracy. Algorithm optimization improves dispersion accuracy, the database ensures accurate parameter matching, the calibration step establishes a wavelength reference, and the sharing mechanism promotes collaborative work in subsequent imaging stages.

[0045] In step 4 of this invention, after receiving the spectral components through the high-precision imaging module, wavefront aberration calculation is performed using Zernike polynomials to quantify the degree of optical system distortion. The module uses a convolutional neural network to learn historical imaging data and predict imaging mirror parameter settings. Based on the aberration calculation results and predicted parameters, the high-precision imaging module evaluates and generates an image sharpness index and sends the sharpness index to steps 3 and 5. Aberration analysis diagnoses focusing defects, the neural network optimizes imaging conditions, the evaluation step monitors image quality, and index sharing realizes a cross-step quality feedback loop. The module focuses the spectral components to generate preliminary spectral images corresponding to different narrow bands and outputs them to step 5.

[0046] In step 5 of this invention, the initial spectral image is compensated for focal plane using a real-time compensation module. The Kalman filter algorithm is used to process the historical defocus data sequence and predict the defocus compensation amount at future moments. The module uses a piezoelectric ceramic actuator to receive the compensation amount command and perform precise nanometer-level displacement adjustment to drive the focal plane position change to compensate for the defocus phenomenon. After the compensation operation is completed, the real-time compensation module generates data on the compensation effect as an image quality evaluation index and feeds this index back to step 4 to form a closed-loop logic for focal plane stability control. The compensated image is then output as a narrowband stable image to step 6.

[0047] In step 6 of this invention, multiple narrowband stable images are received through an intelligent stitching module. The SIFT feature matching algorithm is used to extract scale-invariant feature transform descriptors from the images to achieve accurate registration of key point correspondences between images. The registered images are input into a generative adversarial network for fusion processing. The generator network synthesizes a seamless stitched image, and the discriminator network evaluates the image authenticity to optimize the fusion quality. The module uses a reinforcement learning algorithm to dynamically adjust the registration threshold and fusion parameters based on the fusion results. The stitching accuracy is iteratively optimized through a reward mechanism. The intelligent stitching module feeds back the index characterizing the image alignment accuracy to step 5 and the stitching progress index to step 7 to maintain the continuity of the system workflow. The fused image is output as a full-spectrum high-resolution spectral image.

[0048] In step 7 of this invention, the collaborative control module communicates bidirectionally with steps 1 to 6, receiving user-inputted demand parameters and module state data generated in steps 1 to 6. The module runs a Monte Carlo simulation method, randomly sampling system parameter combinations to simulate step interaction behaviors, and pre-simulating potential state changes during the full-spectrum imaging process. The simulation results are combined with the state data and input into a Bayesian optimization algorithm. The parameter space is modeled through a Gaussian process, efficiently searching for the optimal combination of control parameters to generate optimized parameters. The collaborative control module performs parameter mapping based on the optimized parameters and a pre-built system performance lookup table, generating control signals and sending the control signals to steps 1 to 6, thereby achieving dynamic coordination and optimized control of global resources.

[0049] This invention addresses the technical problems of insufficient spectral resolution, data stitching difficulties, and low real-time measurement efficiency caused by fixed hardware parameters in broadband spectral imaging systems through modular design and intelligent algorithms. The intelligent acquisition module integrates photoelectric sensors to monitor the incident light intensity distribution in real time, driving dynamic adjustment of the slit width. Combined with Monte Carlo optical simulation to pre-determine the slit switching sequence, it achieves adaptive optimization of the light signal acquisition process, thereby improving spectral resolution. The dynamic collimation module uses a particle swarm optimization algorithm to calculate collimator parameters and integrates an aberration detection algorithm based on interference principles. The detected aberration results are fed back to the intelligent acquisition module as a collimation index, forming a closed-loop adjustment for optical path shaping and enhancing beam parallelism. The adaptive dispersion module optimizes the rotation angle of the dispersive element using a genetic algorithm, queries a grating constant database to match the current spectral range, performs wavelength calibration to generate wavelength calibration data, and shares it with the high-precision imaging module to achieve dynamic calibration of wavelength separation accuracy. The high-precision imaging module applies Zernike polynomials to calculate wavefront aberrations, combines convolutional neural networks to predict imaging mirror parameters, evaluates and generates image sharpness indices, and feeds them back to the adaptive dispersion module and real-time compensation module to optimize focusing quality. The real-time compensation module uses a Kalman filter algorithm to predict the defocus compensation amount and performs nanometer-level displacement compensation through a piezoelectric ceramic actuator. The resulting compensation effect data is fed back to the high-precision imaging module as an image quality evaluation index to maintain focal plane stability. The intelligent stitching module uses the SIFT feature matching algorithm to register multiple narrowband stable images, combines them with a generative adversarial network for fusion processing, and dynamically optimizes the registration and fusion parameters through a reinforcement learning algorithm to solve the image stitching misalignment problem. The collaborative control module receives status data from each module, runs a Monte Carlo simulation to preview system behavior, generates optimized parameters using a Bayesian optimization algorithm, distributes control signals through a master-slave communication protocol, coordinates global resources, and improves real-time measurement efficiency. The bidirectional feedback and intelligent algorithm collaboration between these modules enable dynamic adaptation of hardware parameters, significantly improving spectral resolution, data stitching accuracy, and measurement efficiency.

[0050] Please see Figures 2 to 7 The hyperspectral analyzer full-spectrum stitching high-resolution imaging system of the present invention addresses the problems of insufficient spectral resolution, data stitching difficulties, and low real-time measurement efficiency caused by the fixed hardware parameters of existing broadband spectral imaging systems that cannot be dynamically adapted. It achieves dynamic optimization through modular design and intelligent algorithms. The specific implementation of the system is as follows.

[0051] The system consists of an intelligent acquisition module, a dynamic collimation module, an adaptive dispersion module, a high-precision imaging module, a real-time compensation module, an intelligent stitching module, and a collaborative control module. The intelligent acquisition module integrates photoelectric sensors to continuously monitor the intensity distribution of incident light and drives a slit width adjustment mechanism based on the monitoring data. Before slit switching, a Monte Carlo optical simulation is run to simulate the propagation behavior of the light signal, and the slit switching sequence is determined based on the simulation results to optimize the optical path input. Slit adjustment can be performed discretely or continuously. Discrete adjustment switches between slits of different widths by rotating a wheel, while continuous adjustment controls the opening size by moving a block.

[0052] After receiving the collimated beam, the dynamic collimation module uses a particle swarm optimization algorithm to iteratively search for the collimation mirror parameter combination to minimize the beam divergence angle. The module integrates an aberration detection algorithm based on the interference principle to analyze the beam parallelism deviation in real time and feeds back the aberration results to the intelligent acquisition module to form a closed-loop adjustment for optical path shaping.

[0053] The adaptive dispersion module uses a genetic algorithm to evolve the rotation angle of the dispersive element to enhance wavelength separation linearity. The module queries a grating constant database to match the current spectral imaging range and combines the optimized angle to perform wavelength calibration, generating calibration data that is shared with the high-precision imaging module. The dispersive element employs a rotating grating or prism structure. Grating-type dispersion uses a segmented grating to rotate a specific blaze direction into the optical path, while prism-type dispersion disperses the spectral range through rotation.

[0054] The high-precision imaging module applies Zernike polynomial decomposition of wavefront aberrations to quantify the distortion of the optical system. It uses a convolutional neural network to learn historical imaging data and predict imaging mirror parameters. Based on aberration calculations and parameter predictions, it evaluates image sharpness metrics and sends these metrics to the adaptive dispersion module and the real-time compensation module. The real-time compensation module uses a Kalman filter algorithm to process historical defocus data sequences to predict defocus compensation amounts. It then uses a piezoelectric ceramic actuator to perform nanometer-level displacement adjustment of the focal plane position, generating a compensated image quality evaluation metric that is fed back to the high-precision imaging module.

[0055] The intelligent stitching module employs the SIFT feature matching algorithm to extract scale-invariant feature transform descriptors from multiple narrowband stable images for registration. After registration, the images are input into a generative adversarial network for fusion processing. The generator network synthesizes a seamless stitched image, while the discriminator network evaluates the realism and optimizes the fusion quality. The module uses a reinforcement learning algorithm to dynamically adjust the registration threshold and fusion parameters based on the fusion results and iteratively optimizes the stitching accuracy through a reward mechanism. Simultaneously, the image alignment accuracy index is fed back to the real-time compensation module, and the stitching progress index is sent to the collaborative control module. The collaborative control module receives the status data of each module, runs Monte Carlo simulation, randomly samples system parameters, and pre-demonstrates the module interaction behavior. Combined with the Bayesian optimization algorithm, it generates optimized control signals by modeling the parameter space through Gaussian processes and distributes them to each module. The system adopts a master-slave communication protocol. The main processor of the collaborative control module exchanges data with the microprocessors of each module. The microprocessors independently execute local control logic, such as the intelligent acquisition module processing sensor data to drive slit adjustment and the dynamic collimation module running particle swarm optimization algorithm.

[0056] In remote sensing monitoring scenarios such as vegetation health analysis, the system dynamically adapts hardware parameters through the above-described implementation methods. After the user inputs the spectral range and resolution requirements, the collaborative control module matches parameter combinations from a pre-built lookup table to drive the switching of each module, achieving high-resolution imaging across the entire spectrum. A bidirectional feedback mechanism between modules ensures improved spectral resolution through algorithmic optimization, data stitching achieves alignment through intelligent fusion technology, and real-time measurement efficiency is enhanced through parallel processing and predictive control.

[0057] When implemented in remote sensing monitoring scenarios such as vegetation health analysis, this invention utilizes a hyperspectral analyzer's full-spectrum mosaic high-resolution imaging system. Through an intelligent acquisition module integrating photoelectric sensors, it monitors the incident light intensity distribution in real time. Based on the monitoring data, it drives a dynamic adjustment mechanism for the slit width. Before slit switching, a Monte Carlo optical simulation is run to simulate the light signal propagation behavior. The simulation results determine the slit switching sequence to optimize the optical path input. The photoelectric sensors provide real-time feedback on ambient light and slit adjustment response feedback data. The simulation previews the switching effect, forming a closed-loop adaptive control in the acquisition process, thus improving spectral resolution. After receiving the collimated beam, the dynamic collimation module uses a particle swarm optimization algorithm to iteratively search for the collimating lens parameter combination to minimize the beam divergence angle. It integrates an aberration detection algorithm based on interferometry to analyze beam parallelism deviation in real time. The detected aberration results are fed back to the intelligent acquisition module as a collimation index, achieving real-time optimization of the optical path shaping. The adaptive dispersion module optimizes the rotation angle of the dispersive element using a genetic algorithm to enhance wavelength separation linearity. It queries a grating constant database to match the current spectral imaging range and, combined with the optimized angle, performs wavelength calibration to generate wavelength calibration data shared with the high-precision imaging module. Wavelength separation accuracy data is also fed back to the dynamic collimation module for dynamic calibration. The high-precision imaging module uses Zernike polynomials to calculate wavefront aberrations and quantify the optical system distortion. It employs a convolutional neural network to learn historical imaging data and predict imaging mirror parameters. Based on the aberration calculation results and predicted parameters, it evaluates and generates an image sharpness index, which is sent to the adaptive dispersion module and the real-time compensation module to optimize focus quality. The real-time compensation module uses a Kalman filter algorithm to process historical defocus data sequences to predict defocus compensation. It performs nanometer-level displacement adjustment of the focal plane position using a piezoelectric ceramic actuator, generating compensation effect data as an image quality evaluation index, which is fed back to the high-precision imaging module to maintain focal plane stability. The intelligent stitching module employs the SIFT feature matching algorithm to extract scale-invariant feature transform descriptors from multiple narrowband stable images for registration. After registration, the images are input into a generative adversarial network for fusion processing. The generator network synthesizes a seamless stitched image, while the discriminator network evaluates realism and optimizes fusion quality. A reinforcement learning algorithm dynamically adjusts the registration threshold and fusion parameters based on the fusion results and iteratively optimizes stitching accuracy through a reward mechanism. The image alignment accuracy index is fed back to the real-time compensation module, and the stitching progress index is sent to the collaborative control module to resolve image stitching misalignment issues. The collaborative control module receives status data from each module, runs Monte Carlo simulation, randomly samples system parameters, and pre-demonstrates module interaction behavior. It combines a Bayesian optimization algorithm to model the parameter space using Gaussian processes, generating optimized control signals that are distributed to each module. Data exchange is achieved through a master-slave communication protocol using a time-division multiplexing mechanism. The master processor sends global control commands, and the microprocessor returns module status information, achieving command-response data synchronization and coordinating global resources to improve real-time measurement efficiency. This enables accurate identification of changing trends in vegetation health analysis through full-spectrum high-resolution imaging.

[0058] When implemented in geological exploration scenarios such as mineral identification, this invention utilizes a hyperspectral analyzer's full-spectrum stitched high-resolution imaging system. This system acquires broadband ambient light signals via an intelligent acquisition module and dynamically adjusts the slit width, outputting a collimated beam to a dynamic collimation module. A photoelectric sensor monitors the incident light intensity distribution, driving slit adjustment, and Monte Carlo optical simulation pre-optimizes the slit sequence input. The dynamic collimation module shapes the collimated beam, outputting a parallel light signal. It uses a particle swarm optimization algorithm to calculate collimator parameters and integrates an aberration detection algorithm to detect aberrations as a collimation index feedback, forming a closed-loop optical path shaping mechanism to improve beam quality. The adaptive dispersion module decomposes the parallel light signal into spectral components, optimizes the rotation angle of the dispersive element using a genetic algorithm, queries a grating constant database to match the mineral's spectral range, performs wavelength calibration to generate wavelength calibration data shared with the high-precision imaging module, and feeds back wavelength separation accuracy data to the dynamic collimation module, achieving wavelength accuracy. The high-precision imaging module focuses spectral components to generate a preliminary spectral image. It applies Zernike polynomials to calculate wavefront aberrations and combines this with a convolutional neural network to predict imaging mirror parameters. The resulting sharpness index is evaluated and fed back to the adaptive dispersion module and the real-time compensation module to optimize the imaging process. The real-time compensation module performs focal plane compensation on the preliminary image, using a Kalman filter algorithm to predict defocus compensation. It then performs nanometer-level displacement compensation using a piezoelectric ceramic actuator, generating compensation effect data as a quality indicator to maintain image stability. The intelligent stitching module registers and fuses multiple stable images. It uses the SIFT feature matching algorithm for accurate registration, a generative adversarial network for fusion processing, and a reinforcement learning algorithm to optimize parameters and resolve stitching misalignments, outputting a high-resolution spectral image across the entire spectrum. The collaborative control module receives user requests and module status data, runs a Monte Carlo simulation to preview system behavior, and uses a Bayesian optimization algorithm to generate control signals. These signals coordinate the microprocessors of each module through a master-slave communication protocol, enabling efficient data stitching and real-time measurement in mineral identification.

[0059] The technical terms used in this invention are explained below: Monte Carlo optical simulation is a numerical simulation method based on random sampling and statistical analysis. It simulates the physical processes of an optical system, such as light signal propagation and energy distribution, by generating a large number of random samples, thereby evaluating system performance or predicting behavior. In a full-spectrum stitched high-resolution imaging system for a hyperspectral analyzer, Monte Carlo optical simulation is applied to the intelligent acquisition module. Before slit switching, it simulates the propagation behavior of light signals through different slit widths. Based on the simulation results, the slit switching sequence is determined, and the optical path input is optimized to improve spectral resolution and acquisition efficiency.

[0060] Particle swarm optimization (PSO) is a swarm intelligence optimization algorithm that simulates the social behavior of flocks of birds or schools of fish. It iteratively updates particle positions and velocities based on individual and group experience to search for the optimal solution in the problem space. In this invention, the PSO algorithm is used by the dynamic collimation module to calculate collimator parameters. It iteratively minimizes the beam divergence angle to shape the optical path, improve beam parallelism, and feeds the optimization results back to the preceding module to achieve closed-loop adjustment.

[0061] Interferometric aberration detection is a technique that uses the interference phenomenon of light to detect aberrations in optical systems. It quantifies beam parallelism deviation and optical defects by analyzing the phase difference and fringe deformation in the interference pattern. In a hyperspectral analyzer system, the dynamic collimation module integrates an aberration detection algorithm based on the interferometric principle. It monitors the aberration results of the collimated beam in real time and feeds the data back to the intelligent acquisition module as a collimation index, collaboratively adjusting the slit parameters to enhance system stability.

[0062] Genetic algorithm optimization is an evolutionary algorithm that simulates the natural selection process. It uses operations such as selection, crossover, and mutation to evolve unsets and solve complex optimization problems. In this invention, genetic algorithm optimization is applied to the adaptive dispersion module to optimize the rotation angle of the dispersive elements, enhance wavelength separation linearity, and perform wavelength calibration using a grating constant database. The generated wavelength calibration data is then shared with subsequent modules to achieve accuracy and consistency in full-spectrum imaging.

[0063] Zernike polynomial calculation is a mathematical tool used to describe optical wavefront aberrations. By decomposing wavefront distortion using orthogonal polynomials, it quantifies the degree of aberrations in an optical system, such as spherical aberration, coma, and astigmatism. In a high-resolution imaging system with full-spectrum stitching in a hyperspectral analyzer, the high-precision imaging module applies Zernike polynomials to calculate wavefront aberrations, analyzes the distortion of the beam after passing through optical components, thereby evaluating the optical quality of the imaging system and providing a quantitative basis for subsequent parameter optimization.

[0064] Convolutional neural network prediction is a deep learning algorithm that automatically extracts data features through multiple convolutional layers to achieve end-to-end pattern recognition and parameter regression. In the technical solution of this invention, the high-precision imaging module uses a convolutional neural network to learn historical imaging data, predict the optimal parameter settings of the imaging lens, such as focal length and magnification, and dynamically adjusts the imaging conditions based on the input spectral components to improve image clarity and spectral resolution.

[0065] The Kalman filter algorithm is a recursive state estimation algorithm that optimizes the prediction of the dynamic system state in real time by combining system model and noise statistics. In the real-time compensation module, the Kalman filter algorithm processes the historical defocus data sequence, predicts the defocus compensation amount at future moments, reduces the influence of measurement noise through iterative updates, achieves accurate calibration of the focal plane position, and maintains imaging stability.

[0066] A piezoelectric ceramic actuator is a micro-displacement actuator based on the piezoelectric effect, which can convert electrical signals into nanometer-level mechanical displacement. The real-time compensation module of this invention uses a piezoelectric ceramic actuator to receive the defocus compensation command predicted by the Kalman filter algorithm, drives the focal plane to make precise displacement adjustment, compensates for defocusing caused by environmental changes or system vibration, and realizes the stability and resolution of the image focal plane.

[0067] SIFT feature matching algorithm is a feature detection and description algorithm in computer vision. Through steps such as scale space extremum detection, key point localization, orientation assignment, and feature descriptor generation, it extracts scale-invariant and rotation-invariant feature points in images to achieve stable matching between images. In the full-spectrum high-resolution imaging system of a hyperspectral analyzer, the intelligent stitching module uses SIFT feature matching algorithm to register multiple narrowband stable images, extracts scale-invariant feature transformation descriptors from the images, establishes key point correspondences, solves the problem of misalignment in stitching different narrowband images, and improves data alignment accuracy.

[0068] Generative Adversarial Network (GAN) fusion is a deep learning framework based on adversarial training between a generator and a discriminator. The generator is responsible for synthesizing data, while the discriminator evaluates the authenticity of the data. The generation quality is optimized by minimizing the adversarial loss. In the technical solution of this invention, the intelligent stitching module uses a GAN to fuse registered narrowband stable images. The generator network synthesizes a seamless full-spectrum high-resolution spectral image, and the discriminator network evaluates the authenticity of the fused image and dynamically adjusts the network parameters to eliminate stitching artifacts, thereby achieving the naturalness and continuity of image fusion.

[0069] Reinforcement learning algorithm optimization is a machine learning method that learns the optimal policy through interaction between an agent and its environment. It iteratively updates policy parameters based on reward signals to maximize cumulative rewards. In a full-spectrum high-resolution imaging system for hyperspectral analyzers, the intelligent stitching module uses reinforcement learning algorithms to optimize the parameters of image registration and fusion. It sets a reward function based on the fusion results, dynamically adjusts the registration threshold and fusion weights, and improves stitching efficiency and quality through a trial-and-error mechanism, thereby achieving adaptive parameter adjustment.

[0070] Monte Carlo simulation is a numerical method based on random sampling and statistical simulation. It simulates the uncertain behavior of a system through a large number of random experiments for probabilistic prediction and decision support. In the technical solution of this invention, the collaborative control module adopts Monte Carlo simulation to randomly sample system parameters and simulate the interaction behavior of the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. It simulates the potential state changes in the full-spectrum imaging process, provides a probability distribution basis for Bayesian optimization algorithms, and realizes the dynamic generation and optimization of control signals.

[0071] Bayesian optimization is a global optimization strategy based on Bayesian theory, particularly suitable for scenarios where the objective function has high computational cost or is a black-box function. The algorithm models the uncertainty of the parameter space through a Gaussian process, using the desired improvement of the acquisition function to balance exploration and utilization, gradually converging to the optimal parameter combination. In a full-spectrum mosaic high-resolution imaging system for a hyperspectral analyzer, the collaborative control module uses Bayesian optimization to process module state data and Monte Carlo simulation results, updates the parameter probability distribution, generates optimized control signals, and dynamically adjusts various hardware parameters to improve imaging efficiency and quality.

[0072] Two-way communication is an interactive mode that supports bidirectional data flow between the sender and receiver, commonly found in distributed control systems. It employs a master-slave protocol and time-division multiplexing technology to achieve efficient synchronous exchange of commands and status information, enabling real-time coordination. In a full-spectrum mosaic high-resolution imaging system for a hyperspectral analyzer, the collaborative control module interacts with the intelligent acquisition module and dynamic collimation module through a two-way communication mechanism. The main processor sends global control commands, and the microprocessors of each module return status data, enabling dynamic resource allocation and system behavior optimization.

[0073] Feedback control loop is a closed-loop regulation structure that feeds back the system output signal to the input, compares it with a reference value to generate an error signal, and drives the controller to adjust the system behavior. This loop enhances the system's stability, robustness, and adaptability. In a full-spectrum stitched high-resolution imaging system for a hyperspectral analyzer, feedback control loop runs through multiple modules. For example, the dynamic collimation module feeds back the collimation index to the intelligent acquisition module, and the real-time compensation module feeds back the compensation effect data to the high-precision imaging module, forming a closed-loop optimization for optical path shaping and focal plane compensation.

[0074] Grating constant database query is a grating parameter matching method based on database retrieval technology. Grating constants define the spacing between grating lines, and the database stores constant information for different gratings. The query operation can quickly match the current spectral range requirements. In a full-spectrum mosaic high-resolution imaging system of a hyperspectral analyzer, the adaptive dispersion module queries the grating constant database, combines it with a genetic algorithm to optimize the rotation angle of the dispersive element, and performs wavelength calibration, achieving dynamic adaptation of wavelength separation accuracy and spectral imaging range.

Claims

1. A high-resolution imaging system for full-spectrum stitching of a hyperspectral analyzer, characterized in that, It includes an intelligent acquisition module, a dynamic collimation module, an adaptive dispersion module, a high-precision imaging module, a real-time compensation module, an intelligent stitching module, and a collaborative control module; The intelligent acquisition module acquires broadband ambient light signals, dynamically adjusts the slit width, and outputs a collimated beam to the dynamic collimation module. The dynamic collimation module performs optical path shaping on the collimated beam and outputs a parallel light signal to the adaptive dispersion module; the dynamic collimation module also feeds back the collimation index, which characterizes the parallelism of the light signal, to the intelligent acquisition module; The adaptive dispersion module decomposes the parallel light signal into spectral components and outputs them to the high-precision imaging module; the adaptive dispersion module also feeds back data characterizing the wavelength separation accuracy to the dynamic collimation module; The high-precision imaging module focuses the spectral components to generate preliminary spectral images corresponding to different narrow bands and outputs them to the real-time compensation module; the high-precision imaging module also feeds back the sharpness index, which characterizes the image quality, to the adaptive dispersion module. The real-time compensation module performs focal plane compensation on the preliminary spectral image and outputs a narrowband stable image to the intelligent stitching module; the real-time compensation module also feeds back data characterizing the compensation effect to the high-precision imaging module; The intelligent stitching module registers and fuses multiple narrowband stable images to output a full-spectrum high-resolution spectral image; the intelligent stitching module also feeds back the index characterizing the image alignment accuracy to the real-time compensation module; The collaborative control module communicates bidirectionally with the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. It receives user requirements and module status data, and outputs control signals to the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module.

2. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The intelligent acquisition module integrates a photoelectric sensor; the photoelectric sensor monitors the intensity distribution of incident light, and the intelligent acquisition module drives the slit width adjustment according to the monitored light intensity distribution; the intelligent acquisition module runs Monte Carlo optical simulation before slit switching and determines the slit switching sequence based on the simulation results.

3. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The dynamic collimation module is configured to: calculate the collimation mirror parameters using a particle swarm optimization algorithm; integrate an aberration detection algorithm based on the interference principle for detecting aberrations; and feed back the detected aberration results as a collimation index to the intelligent acquisition module.

4. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The adaptive dispersion module is configured to: optimize the rotation angle of the dispersive element using a genetic algorithm; query the grating constant database to match the current spectral range, and perform wavelength calibration based on the query results and the optimized rotation angle to generate wavelength separation accuracy data as wavelength calibration data; and share the wavelength calibration data with the high-precision imaging module.

5. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The high-precision imaging module is configured as follows: Zernike polynomials were applied to calculate wavefront aberrations, and convolutional neural networks were used to predict imaging mirror parameters. The high-precision imaging module evaluates and generates a sharpness index as an image sharpness index based on the aberration calculation results and the predicted imaging mirror parameters; the high-precision imaging module sends the image sharpness index to the adaptive dispersion module and the real-time compensation module.

6. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The real-time compensation module is configured to: use a Kalman filter algorithm to predict the defocus compensation amount, and use a piezoelectric ceramic actuator to perform nanometer-level displacement corresponding to the defocus compensation amount for focal plane compensation; the real-time compensation module generates data of the compensation effect after compensation as an image quality evaluation index, and feeds the image quality evaluation index back to the high-precision imaging module.

7. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, The intelligent stitching module is configured to: register multiple narrowband stable images using the SIFT feature matching algorithm; fuse the registered images using a generative adversarial network; optimize the parameters of image registration and fusion using a reinforcement learning algorithm; calculate a stitching progress index based on the processing progress of image registration and fusion, and feed the stitching progress index, which represents the processing progress, back to the collaborative control module.

8. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, Also includes: The collaborative control module receives module status data from the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module. The collaborative control module uses Monte Carlo simulation to simulate the interaction behavior of modules by combining random sampling system parameters, pre-simulating potential state changes during the full-spectrum imaging process. It also employs a Bayesian optimization algorithm to model the parameter space through a Gaussian process, updates the probability distribution based on the acquired state data and simulation results, and converges to the optimal parameter combination through an expectation boosting function to generate optimized parameters. These optimized parameters define the content of the control signal. The collaborative control module generates control signals by mapping parameters according to the optimized parameters and a pre-built system performance lookup table, and sends the control signals to the intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module.

9. The hyperspectral analyzer full-spectrum stitched high-resolution imaging system as described in claim 1, characterized in that, Also includes: The intelligent acquisition module, dynamic collimation module, adaptive dispersion module, high-precision imaging module, real-time compensation module, and intelligent stitching module each include an independent microprocessor for executing the functions of each module, including photoelectric sensor data processing, optimization algorithm calculation, and actuator driving. The collaborative control module includes a main processor for coordinating control and running system-level optimization algorithms. The main processor and the microprocessors of each module exchange data through a master-slave communication protocol using a time-division multiplexing mechanism. The main processor sends global control commands, and the microprocessors return module status information, realizing command-response data synchronization to complete the bidirectional communication.

10. A method for full-spectrum stitching high-resolution imaging of a hyperspectral analyzer, applied to the full-spectrum stitching high-resolution imaging system of a hyperspectral analyzer as described in any one of claims 1 to 9, characterized in that, include: Step 1: Acquire broadband ambient light signals and dynamically adjust the slit width to output a collimated beam; Step 2: Receive the collimated beam, perform optical path shaping on the collimated beam, output a parallel light signal, and feed back the collimation index to Step 1; Step 3: Receive the parallel light signal, decompose the parallel light signal into spectral components, output the spectral components, and feed back the wavelength separation accuracy data to Step 2; Step 4: Receive spectral components, focus the spectral components to generate preliminary spectral images corresponding to different narrowbands, output the preliminary spectral images, and feed back the sharpness index to Step 3; Step 5: Receive the preliminary spectral image, perform focal plane compensation on the preliminary spectral image, output a narrowband stable image, and feed back the compensation effect data to Step 4; Step 6: Receive the narrowband stabilized image, register and fuse multiple narrowband stabilized images, output a full-spectrum high-resolution spectral image, and feed back the image alignment accuracy index to Step 5; Step 7: Interact with Steps 1 to 6, receive the user-inputted requirements parameters and the status data generated in Steps 1 to 6, and send control signals to Steps 1 to 6.

Images