Spectrum reconstruction method and system based on dynamic band selection and sparse sampling

Abstract

The invention discloses a spectrum reconstruction method and system based on dynamic band selection and sparse sampling, and the method comprises the steps: selecting a fixed sampling mode or a sparse segmented sampling mode based on target spectrum features, so as to obtain a sampling wavelength sequence; any sampling wavelength in the sampling wavelength sequence is obtained, the theoretical cavity length corresponding to the sampling wavelength is calculated based on the linear relation between the cavity length of the FP cavity and the wavelength, the theoretical cavity length serves as the cavity length of the FP cavity after voltage is applied, the theoretical applied voltage corresponding to the sampling wavelength is calculated and measured, and a theoretical voltage signal sequence corresponding to the sampling wavelength sequence is obtained; according to the method, only the characteristic wave band of the measured target needs to be extracted instead of full-spectrum sampling, redundant information is reduced, the transmission and storage cost is reduced, meanwhile, the imaging speed is increased, the real-time performance is improved, full-spectrum reconstruction is achieved through spectrum reconstruction, and the method is suitable for large-scale popularization and application. And no loss is caused.

Description

Technical Field

[0001] The present invention relates to the technical field of spectrum reconstruction, and in particular to a spectrum reconstruction method and system based on dynamic band selection and sparse sampling. Background Art

[0002] Spectral imaging is a combination of imaging technology and spectral detection technology. Its main technical goal is to obtain spectral images. It is widely used in many fields such as environmental monitoring, industrial testing, and biomedicine. In order to achieve full spectral coverage (usually 400-1700nm), traditional grating spectrometer systems still need to collect data from more than 200 bands, even if the target signal is only concentrated in a few specific bands (such as gas absorption peaks). As a result, more than 90% of the data becomes redundant information. In full-spectrum acquisition mode, data must be collected for the sample under test in each sampling band, and the amount of data generated is extremely large. Although existing solutions rely on data compression algorithms, due to the large amount of raw data, the transmission and storage costs remain high.

[0003] In addition, grating spectrometer systems rely on a sweeping or push-sweeping mechanism, requiring the spectrometer to be constantly moved to acquire spectra at each point and ultimately synthesize a two-dimensional image. This scanning spectral imaging system is not only bulky but also extremely slow in imaging speed. A single imaging session takes several seconds, and invalid bands cannot be skipped, resulting in a significant waste of time. In high-speed moving target monitoring (such as drone dynamic tracking) and real-time industrial inspection scenarios, traditional systems simply cannot meet the millisecond-level response requirements. At the same time, spectroscopic components such as gratings and prisms require precise mechanical support, and the system volume is generally greater than 10 cm³, making miniaturization difficult. Therefore, seeking a new spectroscopic method that can accurately acquire the band data required for spectral analysis, reduce redundant information, lower transmission and storage costs, while simultaneously accelerating imaging speed, improving real-time performance, and achieving system miniaturization has become a key challenge that needs to be overcome in the current field of spectral imaging. Summary of the Invention

[0004] In response to the problems in the related art, the present invention proposes a spectrum reconstruction method based on dynamic band selection and sparse sampling to overcome the technical problems existing in the background technology.

[0005] To this end, the specific technical solutions adopted in the present invention are as follows:

[0006] A spectrum reconstruction method based on dynamic band selection and sparse sampling, comprising:

[0007] S1. Get the starting wavelength λ min and the cutoff wavelength λ max , based on the target spectral characteristics, select a fixed sampling method or a sparse segmented sampling method to obtain the starting wavelength λ min To cut-off wavelength λ maxThe sampling wavelength sequence of

[0008] S2. Obtaining the corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL based on the design parameters of the FP cavity;

[0009] S3. Obtain any sampling wavelength λ in the sampling wavelength sequence c , the sampling wavelength λ is calculated based on the linear relationship between the cavity length and wavelength of the FP cavity c The corresponding theoretical cavity length L theory , with the theoretical cavity length L theory is the cavity length of the FP cavity after voltage is applied. According to the corresponding relationship obtained by S2, the measurement sampling wavelength λ is calculated. c The corresponding theoretical applied voltage V theory , the theoretical applied voltage V theory Applied to the FP cavity, the theoretical cavity length L is measured theory The actual theoretical applied voltage V theory Is the error of the cavity length within the error range? If it exceeds the error range, adjust the theoretical applied voltage V theory Until the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, all sampling wavelengths in the sampling wavelength sequence are processed one by one to obtain the theoretical voltage signal sequence corresponding to the sampling wavelength sequence;

[0010] S4, applying the theoretical voltage signal sequence to the FP cavity one by one to complete the spectrum data acquisition of all sampling wavelengths in the sampling wavelength sequence;

[0011] S5, reconstruct the spectral data collected in S4 to obtain the starting wavelength λ min To cut-off wavelength λ max Reconstructed spectral data.

[0012] This spectral reconstruction method based on dynamic band selection and sparse sampling performs wavelength sampling by selecting a fixed sampling method or a sparse segmented sampling method based on the target spectral characteristics. It only needs to extract the characteristic bands of the target to be measured instead of full spectrum sampling, which reduces redundant information and transmission and storage costs, while accelerating imaging speed and improving real-time performance. In addition, full spectrum reconstruction is achieved through spectral reconstruction without causing any loss.

[0013] Furthermore, when a voltage is applied to the FP cavity, the cavity length of the FP cavity will change, and the cavity length change is calculated. The corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL is: the theoretical cavity length change ΔL is positively correlated with the theoretical voltage V.

[0014] Furthermore, the sampling interval or sampling number of the fixed sampling method is a fixed value, which is determined by the starting wavelength λ. minand the cutoff wavelength λ max Generates a uniformly spaced sampling wavelength sequence, which is suitable for preliminary detection scenarios of unknown target spectral characteristics; the sparse segmented sampling method includes dense sampling segments and sparse sampling segments, which is suitable for detection scenarios with known target characteristic bands.

[0015] Furthermore, the sampling interval Δλ of the fixed sampling method is the cutoff wavelength λ max With the starting wavelength λ min Difference and sampling number N samples -1, when there is a non-divisible situation or the ratio is not an integer, it is rounded up.

[0016] Furthermore, the sparse segmented sampling method is to perform dense sampling on the bands within a certain range before and after the target feature band, and perform sparse sampling on the remaining bands. The sampling interval during dense sampling is smaller than the sampling interval during sparse sampling.

[0017] Furthermore, when the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, continue to apply the actual theoretical applied voltage V theory After a certain period of time, the data is collected after the cavity length is kept stable.

[0018] Furthermore, the spectral data is reconstructed through a spectral reconstruction model. The spectral reconstruction model training and construction include:

[0019] A1: Data collection is performed on material samples with known spectral characteristics using fixed sampling, sparse segmented sampling, and full-band spectrum acquisition methods to obtain fixed sampling reconstructed spectra, sparse segmented sampling reconstructed spectra, and standard spectra.

[0020] A2: Randomly generate different band sampling sequences from the fixed sampling reconstructed spectrum, the sparse segmented sampling reconstructed spectrum, and the standard spectrum. Obtain data sequences of several paired fixed sampling reconstructed spectra and corresponding standard spectra, as well as data sequences of several paired sparse segmented sampling reconstructed spectra and corresponding standard spectra, to construct a training dataset.

[0021] A3: Input the training dataset into the deep learning model, change the sampling parameters of the fixed sampling method and the sparse segmented sampling method, and optimize the model parameters by minimizing the error index between the reconstructed spectrum and the standard spectrum to obtain the spectral reconstruction model.

[0022] Furthermore, before reconstructing the spectral data, the spectral data collected by S4 is preprocessed to eliminate the systematic errors and random noise introduced during the data acquisition process. The preprocessed spectral data is then assigned to the positions corresponding to the standard spectrum obtained by full-band spectral acquisition to form a sparse matrix, in which the uncollected positions are filled with zero.

[0023] A spectrum reconstruction system based on dynamic band selection and sparse sampling, for implementing any of the above-mentioned spectrum reconstruction methods based on dynamic band selection and sparse sampling, comprising:

[0024] FP cavity tunable filter module, spectral data acquisition module and spectral data processing module. When the FP cavity tunable filter module is powered on, the cavity length changes to change the central wavelength of its transmission spectrum. It is used to screen the wavelength sequence to be collected according to the preset sampling method. The spectral data acquisition module collects the spectral data screened by the FP cavity tunable filter module to obtain the starting wavelength λ min To cut-off wavelength λ max The spectrum data processing module is connected to the spectrum data acquisition module to process the spectrum data collected by the spectrum data acquisition module. The processing process includes:

[0025] S2. Obtaining the corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL based on the design parameters of the FP cavity;

[0026] S3. Obtain any sampling wavelength λ in the sampling wavelength sequence c , the sampling wavelength λ is calculated based on the linear relationship between the cavity length and wavelength of the FP cavity c The corresponding theoretical cavity length L theory , with the theoretical cavity length L theory is the cavity length of the FP cavity after voltage is applied. According to the corresponding relationship obtained by S2, the measurement sampling wavelength λ is calculated. c The corresponding theoretical applied voltage V theory , the theoretical applied voltage V theory Applied to the FP cavity, the theoretical cavity length L is measured theory The actual theoretical applied voltage V theory Is the error of the cavity length within the error range? If it exceeds the error range, adjust the theoretical applied voltage V theory Until the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, all sampling wavelengths in the sampling wavelength sequence are processed one by one to obtain the theoretical voltage signal sequence corresponding to the sampling wavelength sequence;

[0027] S4, applying the theoretical voltage signal sequence to the FP cavity one by one to complete the spectrum data acquisition of all sampling wavelengths in the sampling wavelength sequence;

[0028] S5, reconstruct the spectral data collected in S4 to obtain the starting wavelength λ min To cut-off wavelength λ max Reconstructed spectral data.

[0029] Furthermore, it also includes a light source module, an optical acquisition module and an FP cavity driving module. The light source module provides a light source for the optical acquisition module to collect target spectral information. The FP cavity driving module is used to apply voltage to the FP cavity tunable filter module to drive the cavity length change of the FP cavity. The spectral data acquisition module is connected to the FP cavity driving module. When receiving the trigger signal of the FP cavity driving module, the spectral data acquisition module performs spectral data acquisition.

[0030] The beneficial effects of the present invention are:

[0031] 1. This spectral reconstruction method based on dynamic band selection and sparse sampling performs wavelength sampling by selecting a fixed sampling method or a sparse segmented sampling method based on the target spectral characteristics. It does not require full spectrum sampling for any target, reduces redundant information, lowers transmission and storage costs, and speeds up imaging and improves real-time performance. In addition, full spectrum reconstruction is achieved through spectral reconstruction without causing any loss.

[0032] 2. This spectral reconstruction method based on dynamic band selection and sparse sampling provides two dynamic sampling modes: fixed sampling and sparse segmented sampling, to meet the needs of different application scenarios. The fixed sampling method can achieve a global preliminary scan of the spectrum without prior spectral knowledge, which is suitable for scenarios such as environmental pollutant screening and system full spectrum calibration. The sparse segmented sampling method dynamically adjusts the sampling strategy based on prior knowledge of spectral characteristics or real-time feedback, densely sampling key areas and sparsely sampling non-key areas. While ensuring the accuracy of key band data, it effectively reduces the amount of data collected, saves resources, and improves sampling efficiency.

[0033] 3. This spectral reconstruction method based on dynamic band selection and sparse sampling achieves subnanometer-level cavity length adjustment accuracy for the FP cavity through precise wavelength-cavity length-voltage mapping, combined with a closed-loop voltage feedback mechanism based on a PID control algorithm. This ensures the FP cavity remains stable at the desired state and provides precise filtering conditions for spectral acquisition. Simultaneously, the spectral data processing module utilizes advanced spectral reconstruction algorithms to reconstruct a complete full spectrum from sparsely sampled data, preserving key spectral features and improving spectral resolution and analytical accuracy.

[0034] 4. The spectral reconstruction system based on dynamic band selection and sparse sampling adopts a miniaturized structural design to reduce hardware costs and adapt to lightweight application scenarios. It achieves millisecond-level fast imaging and adaptive sampling based on the FP cavity tunable filter, breaking through the limitations of slow imaging in traditional systems. Through dynamic band selection and sparse sampling, data redundancy is reduced. Combined with the compressed sensing algorithm, the accuracy of spectral analysis is guaranteed while reducing the amount of data, effectively solving the problems of bloated data, slow response, and expensive equipment in traditional systems, and promoting the development of spectral analysis technology towards lightweight and intelligent directions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 This is a schematic flow chart of the steps of a spectrum reconstruction method based on dynamic band selection and sparse sampling according to the first embodiment of the present invention;

[0037] Figure 2 This is a module diagram of a spectrum reconstruction system based on dynamic band selection and sparse sampling according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Example 1

[0040] A spectrum reconstruction method based on dynamic band selection and sparse sampling, such as Figure 1 Shown, including:

[0041] Get the starting wavelength λ min and the cutoff wavelength λ max , based on the target spectral characteristics, select a fixed sampling method or a sparse segmented sampling method to obtain the starting wavelength λ min To cut-off wavelength λ max The sampling wavelength sequence, fixed sampling mode or sparse segmented sampling mode are all interval sampling, and full-band sampling is not required. Specifically:

[0042] The sampling interval or number of samples in the fixed sampling mode is a fixed value, which is determined by the starting wavelength λ. min and the cutoff wavelength λ max Generates a sequence of evenly spaced sampling wavelengths. When there is no prior spectral knowledge or a global preliminary scan of the spectrum is required, a fixed sampling method is used for sampling. For example, in the full spectrum range (400-1700nm), assuming a sampling interval of 5nm, the system will start from the starting wavelength of 400nm and generate a sampling point every 5nm until it reaches or exceeds the cutoff wavelength of 1700nm, and finally output the wavelength sampling sequence. When the input is the number of samples, the sampling interval can be calculated using the formula: . It should be noted that during the calculation process, if there is a non-divisible situation or the obtained wavelength is not an integer, it will be rounded to ensure that the determination of the sampling point is in line with actual operability. The fixed sampling method is suitable for preliminary detection scenarios of unknown target spectral characteristics. For example, when screening pollutants in environmental monitoring, since the spectral characteristics of the pollutants are not clear at the beginning, a fixed sampling method can be used for preliminary scanning detection. In addition, in the full spectrum calibration work during the system calibration stage, the fixed sampling method can help determine the basic situation of the entire spectral range.

[0043] The sparse segmented sampling method, which includes dense sampling segments and sparse sampling segments, dynamically adjusts the sampling strategy based on prior knowledge of spectral characteristics or real-time feedback to achieve "dense sampling in key areas + sparse sampling in non-key areas." When the user enters a list of target characteristic bands, for example, if the CO2 absorption peak is known to be at 1530nm and the CH4 absorption peak is at 1665nm, the system will perform dense sampling within a ±5nm range of these characteristic peaks, achieving dense sampling in key areas and reducing the interval to 1nm or even 0.1nm. In other areas, the interval is relaxed to 10nm, achieving sparse sampling in non-key areas, and finally outputting a wavelength sampling sequence. This method ensures that sufficient data is collected in key characteristic bands, improving accuracy, while reducing sampling points in non-key areas, conserving resources. It is suitable for detection scenarios where the target characteristic bands are known.

[0044] The corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL is obtained based on the design parameters of the FP cavity. Specifically:

[0045] In the FP cavity of this embodiment, when a voltage is applied to it, the cavity length of the FP cavity changes, and the cavity length change is calculated. The FP cavity typically uses piezoelectric ceramics or micro-electromechanical systems to achieve cavity length adjustment. In theory, the corresponding relationship between the theoretical voltage V applied to the FP cavity and the theoretical cavity length change ΔL satisfies the following: the theoretical cavity length change ΔL is positively correlated with the theoretical voltage V. Due to the non-ideal driving process of piezoelectric ceramics and electrostatic forces, the relationship between the two is determined by actual system measurements.

[0046] And according to the multi-beam interference principle of the FP cavity, when the resonance condition is met, the relationship between the cavity length L and the target wavelength is: , where m is the resonance order and θ is the incident angle. The theoretical cavity length L corresponding to the target wavelength can be calculated by this formula theory The initial cavity length L0 of the FP cavity is actually measurable. The length of the piezoelectric ceramic will change after voltage is applied. The FP cavity indirectly drives the change of the cavity length by energizing the piezoelectric ceramic. The change pattern or value of the length of the piezoelectric ceramic after voltage is applied can be obtained by consulting the piezoelectric ceramic manual, thereby obtaining the theoretical voltage signal sequence to achieve precise control of the tunable optical path of the FP cavity.

[0047] Get the sampling wavelength λ in the sampling wavelength sequence c , calculate the sampling wavelength λ c A corresponding voltage signal is applied to the FP cavity, so each voltage signal corresponds to a target cavity length. A capacitive displacement sensor integrated into the FP cavity monitors the actual cavity length in real time. The error between the monitored value and the target value is input into a PID controller for dynamic adjustment to eliminate cavity length errors. If the cavity length does not meet subnanometer stability accuracy (i.e., error ≥ 0.1 nm or standard deviation of fluctuation σ ≥ 0.05 nm), the voltage signal is adjusted until the cavity length meets subnanometer stability accuracy. Once the cavity length stabilizes, spectral data acquisition begins. A module clock synchronization mechanism ensures that acquisition is strictly synchronized with cavity length stability, preventing aliasing of optical signal acquisition and cavity length fluctuations due to timing deviations and ensuring data reliability. This feedback adjustment and triggered acquisition process is repeated until all sampling wavelengths in the sampling wavelength sequence have been acquired, completing spectral data acquisition for the entire sequence.

[0048] Reconstruct the collected spectral data to obtain the starting wavelength λ min To cut-off wavelength λ max The reconstructed spectral data, specifically:

[0049] First, the collected spectral data is preprocessed. Normalization, noise filtering, and baseline correction are performed to eliminate systematic errors and random noise introduced during data acquisition. After preprocessing, the sampling points are assigned values ​​at corresponding positions in the full-band spectral sequence, and unsampled positions are filled with zeros, forming a representation similar to a sparse matrix.

[0050] The data is then fed into a pre-built spectral reconstruction model to reconstruct the complete full spectrum. The reconstruction algorithm can employ an end-to-end model based on deep learning or a traditional compressed sensing reconstruction algorithm (such as OMP and TVAL3). Through an optimized solution process, efficient mapping from sparse sampling to the full spectrum is achieved. This processing approach preserves the key features of the original spectrum while improving spectral resolution through mathematical optimization, ultimately achieving high-precision spectral reconstruction.

[0051] The construction of the spectral reconstruction model includes:

[0052] In a laboratory environment, a dual-system synchronous acquisition strategy was adopted, and fixed sampling, sparse segmented sampling and full-band spectrum acquisition methods were used to collect data on material samples with known spectral characteristics to obtain fixed sampling reconstructed spectra, sparse segmented sampling reconstructed spectra and standard spectra.

[0053] Fixed sampling reconstructed spectra, sparse segmented sampling reconstructed spectra and standard spectra are randomly generated with different band sampling sequences to obtain data sequences of several paired fixed sampling reconstructed spectra and corresponding standard spectra, as well as data sequences of several paired sparse segmented sampling reconstructed spectra and corresponding standard spectra to construct training data sets.

[0054] The training dataset was fed into a deep learning model, and the sampling parameters were varied between fixed and sparse segmented sampling methods. Model parameters were optimized by minimizing error metrics (such as mean square error and spectral angular distance) between the reconstructed spectrum and the standard spectrum. During the training process, a stochastic gradient descent algorithm was used to iteratively update the model weights, gradually improving the accuracy of spectral reconstruction. Ultimately, a high-precision, highly generalizable spectral reconstruction model was formed, enabling reliable conversion from sparsely sampled data to detailed full spectra.

[0055] Example 2

[0056] A spectrum reconstruction system based on dynamic band selection and sparse sampling, such as Figure 2 As shown, it includes: a light source module 1, an optical acquisition module 2, an FP cavity tunable filter module 3, an FP cavity driving module 4, a spectrum data acquisition module 5 and a spectrum data processing module 6.

[0057] Light source module 1 provides a wide-band light source, covering visible light to near-infrared. Precision circuitry and temperature control technology ensure stable output light intensity, with a fluctuation error of less than 0.5%. This stable light source provides the system with a reliable raw signal, preventing detection errors caused by varying lighting conditions and adapting to target detection needs in diverse scenarios.

[0058] Optical acquisition module 2 utilizes optical components such as an object plane lens. Through optimized optical path design, it efficiently focuses light reflected or transmitted from the object's surface and introduces it into the system. Precision alignment of the optical components minimizes aberrations and light energy loss, fully preserving the target's spectral information and providing high-quality optical signals for subsequent processing.

[0059] The FP cavity tunable filter module 3 consists of a pair of parallel highly reflective mirrors forming a resonant cavity. By actively adjusting the cavity length (the distance between the two mirrors), the central wavelength of its transmission spectrum is dynamically changed. The cavity length is adjusted by adjusting the device, and the actual cavity length of the filter is monitored through its circuit and internal structure.

[0060] The FP cavity drive module 4 is the core control unit of the system. It uses the input voltage as a reference to build a precise dynamic control system. It monitors the actual cavity length of the FP cavity in real time at a very high frequency through the built-in capacitor of the FP cavity and the external circuit, and compares it with the theoretical applied voltage V theory The corresponding theoretical cavity length L theory Continuous comparison is performed. Once a deviation between the actual cavity length and the ideal cavity length is detected, the FP cavity driver module 4 will quickly initiate a closed-loop regulation mechanism based on the PID control algorithm: the proportional link quickly adjusts the output voltage according to the current deviation, the integral link eliminates the system static error, and the differential link predicts the deviation change trend to make corrections in advance. These three links work together to form a "monitoring-comparison-correction-re-monitoring" closed loop, ensuring that the FP cavity can accurately reach the target cavity length and ultimately achieve sub-nanometer adjustment accuracy. In actual practice, when analyzing a specific spectral band, even if the cavity length changes slightly due to external environmental interference, the module can complete the adjustment within milliseconds, stabilizing the FP cavity in the desired state and providing precise filtering conditions for subsequent spectral acquisition.

[0061] The spectral data acquisition module 5, the system's core acquisition unit, works in conjunction with the FP cavity driver module 4. Upon receiving a trigger signal from the FP cavity driver module 4, it selects a specific wavelength range or a single characteristic wavelength band for optical signal acquisition based on the signal level characteristics, supporting both continuous scanning and multi-sequence sampling modes. During the acquisition process, the optical signal output by the FP cavity tunable filter module 3 is converted into a digital signal, and data accuracy is ensured through temperature compensation and multiple sampling averaging techniques. The collected digital signal is then transmitted to the spectral data acquisition module 5, ensuring complete and accurate spectral information.

[0062] The spectral data processing module 6 serves as the data center of the system. It is responsible for receiving the spectral data output by the spectral data acquisition module 5. It first preprocesses the data, including noise filtering and baseline correction, and then uses spectral reconstruction to reconstruct the sparsely sampled data into complete full-spectral data.

[0063] This spectral reconstruction system based on dynamic band selection and sparse sampling solves the problems of large data redundancy caused by the need for traditional grating spectrometry systems to cover the entire spectrum, and its scanning imaging relies on precise mechanical structures, has slow imaging speeds, and is bulky, making it difficult to meet the needs of high-speed target monitoring and real-time industrial detection, as well as the inability to achieve miniaturization.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A spectrum reconstruction method based on dynamic band selection and sparse sampling, characterized in that: include: S1. Get the starting wavelength λ min and the cutoff wavelength λ max , based on the target spectral characteristics, select a fixed sampling method or a sparse segmented sampling method to obtain the starting wavelength λ min To cut-off wavelength λ max The sampling wavelength sequence of S2. Obtaining the corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL based on the design parameters of the FP cavity; S3. Obtain any sampling wavelength λ in the sampling wavelength sequence c , the sampling wavelength λ is calculated based on the linear relationship between the cavity length and wavelength of the FP cavity c The corresponding theoretical cavity length L theory , with the theoretical cavity length L theory is the cavity length of the FP cavity after voltage is applied. According to the corresponding relationship obtained by S2, the measurement sampling wavelength λ is calculated. c The corresponding theoretical applied voltage V theory , the theoretical applied voltage V theory Applied to the FP cavity, the theoretical cavity length L is measured theory The actual theoretical applied voltage V theory Is the error of the cavity length within the error range? If it exceeds the error range, adjust the theoretical applied voltage V theory Until the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, all sampling wavelengths in the sampling wavelength sequence are processed one by one to obtain the theoretical voltage signal sequence corresponding to the sampling wavelength sequence; S4, applying the theoretical voltage signal sequence to the FP cavity one by one to complete the spectrum data acquisition of all sampling wavelengths in the sampling wavelength sequence; S5, reconstruct the spectral data collected in S4 to obtain the starting wavelength λ min To cut-off wavelength λ max Reconstructed spectral data.

2. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 1, characterized in that: When a voltage is applied to the FP cavity, the cavity length of the FP cavity changes, and the cavity length change is calculated. The corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL is: the theoretical cavity length change ΔL is positively correlated with the theoretical voltage V.

3. A spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 1 or 2, characterized in that: The sampling interval or number of samples in the fixed sampling mode is a fixed value, which is determined by the starting wavelength λ. min and the cutoff wavelength λ max Generates a uniformly spaced sampling wavelength sequence, which is suitable for preliminary detection scenarios of unknown target spectral characteristics; the sparse segmented sampling method includes dense sampling segments and sparse sampling segments, which is suitable for detection scenarios with known target characteristic bands.

4. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 3, characterized in that: The sampling interval Δλ of the fixed sampling method is the cutoff wavelength λ max With the starting wavelength λ min Difference and sampling number N samples -1, when there is a non-divisible situation or the ratio is not an integer, it is rounded up.

5. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 3, characterized in that: The sparse segmented sampling method is to perform dense sampling on the bands within a certain range before and after the target feature band, and perform sparse sampling on the remaining bands. The sampling interval during dense sampling is smaller than that during sparse sampling.

6. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 1, characterized in that: When the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, continue to apply the actual theoretical applied voltage V theory After a certain period of time, the data is collected after the cavity length is kept stable.

7. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 1, characterized in that: The reconstruction of spectral data is achieved through the spectral reconstruction model. The spectral reconstruction model training and construction include: A1: Data collection is performed on material samples with known spectral characteristics using fixed sampling, sparse segmented sampling, and full-band spectrum acquisition methods to obtain fixed sampling reconstructed spectra, sparse segmented sampling reconstructed spectra, and standard spectra. A2: Randomly generate different band sampling sequences from the fixed sampling reconstructed spectrum, the sparse segmented sampling reconstructed spectrum, and the standard spectrum. Obtain data sequences of several paired fixed sampling reconstructed spectra and corresponding standard spectra, as well as data sequences of several paired sparse segmented sampling reconstructed spectra and corresponding standard spectra, to construct a training dataset. A3: Input the training dataset into the deep learning model, change the sampling parameters of the fixed sampling method and the sparse segmented sampling method, and optimize the model parameters by minimizing the error index between the reconstructed spectrum and the standard spectrum to obtain the spectral reconstruction model.

8. The spectrum reconstruction method based on dynamic band selection and sparse sampling according to claim 1, characterized in that: Before reconstructing the spectral data, the spectral data collected by S4 were preprocessed to eliminate the systematic errors and random noise introduced during the data acquisition process. The preprocessed spectral data were then assigned values ​​at the positions corresponding to the standard spectra obtained by full-band spectral acquisition to form a sparse matrix, in which the uncollected positions were filled with zero.

9. A spectrum reconstruction system based on dynamic band selection and sparse sampling, used to implement a spectrum reconstruction method based on dynamic band selection and sparse sampling as described in any one of claims 1 to 8, characterized in that: The invention comprises an FP cavity tunable filter module (3), a spectrum data acquisition module (5) and a spectrum data processing module (6). When the FP cavity tunable filter module (3) is powered on, the cavity length changes to change the central wavelength of its transmission spectrum, and is used to screen the wavelength sequence to be collected according to a preset sampling method. The spectrum data acquisition module (5) collects the spectrum data screened by the FP cavity tunable filter module (3) to obtain the starting wavelength λ min To cut-off wavelength λ max The spectrum data processing module (6) is connected to the spectrum data acquisition module (5) to process the spectrum data collected by the spectrum data acquisition module (5). The processing process includes: S2. Obtaining the corresponding relationship between the theoretical voltage V and the theoretical cavity length change ΔL based on the design parameters of the FP cavity; S3. Obtain any sampling wavelength λ in the sampling wavelength sequence c , the sampling wavelength λ is calculated based on the linear relationship between the cavity length and wavelength of the FP cavity c The corresponding theoretical cavity length L theory , with the theoretical cavity length L theory is the cavity length of the FP cavity after voltage is applied. According to the corresponding relationship obtained by S2, the measurement sampling wavelength λ is calculated. c The corresponding theoretical applied voltage V theory , the theoretical applied voltage V theory Applied to the FP cavity, the theoretical cavity length L is measured theory The actual theoretical applied voltage V theory Is the error of the cavity length within the error range? If it exceeds the error range, adjust the theoretical applied voltage V theory Until the theoretical cavity length L theory The actual theoretical applied voltage V theory After the error of the cavity length is within the error range, all sampling wavelengths in the sampling wavelength sequence are processed one by one to obtain the theoretical voltage signal sequence corresponding to the sampling wavelength sequence; S4, applying the theoretical voltage signal sequence to the FP cavity one by one to complete the spectrum data acquisition of all sampling wavelengths in the sampling wavelength sequence; S5, reconstruct the spectral data collected in S4 to obtain the starting wavelength λ min To cut-off wavelength λ max Reconstructed spectral data.

10. The spectrum reconstruction system based on dynamic band selection and sparse sampling according to claim 9, characterized in that: The optical acquisition module (2) and the FP cavity driving module (4) are further included. The optical acquisition module (1) provides a light source for the optical acquisition module (2) to acquire target spectral information. The FP cavity driving module (4) is used to apply a voltage to the FP cavity tunable filter module (3) to drive the cavity length of the FP cavity to change. The spectral data acquisition module (5) is connected to the FP cavity driving module (4). When receiving a trigger signal from the FP cavity driving module (4), the spectral data acquisition module (5) performs spectral data acquisition.

Images