CN116164841A - Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip - Google Patents

Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip Download PDF

Info

Publication number
CN116164841A
CN116164841A CN202310456976.9A CN202310456976A CN116164841A CN 116164841 A CN116164841 A CN 116164841A CN 202310456976 A CN202310456976 A CN 202310456976A CN 116164841 A CN116164841 A CN 116164841A
Authority
CN
China
Prior art keywords
spectrum
network
reconstruction
imaging chip
calculation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202310456976.9A
Other languages
Chinese (zh)
Other versions
CN116164841B (en
Inventor
胡长虹
薛栋林
吕宝林
肖树林
哈清华
薛旭成
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changchun Institute of Optics Fine Mechanics and Physics of CAS
Original Assignee
Changchun Institute of Optics Fine Mechanics and Physics of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changchun Institute of Optics Fine Mechanics and Physics of CAS filed Critical Changchun Institute of Optics Fine Mechanics and Physics of CAS
Priority to CN202310456976.9A priority Critical patent/CN116164841B/en
Publication of CN116164841A publication Critical patent/CN116164841A/en
Application granted granted Critical
Publication of CN116164841B publication Critical patent/CN116164841B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28Investigating the spectrum
    • G01J3/2823Imaging spectrometer
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/02Details
    • G01J3/0205Optical elements not provided otherwise, e.g. optical manifolds, diffusers, windows
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28Investigating the spectrum
    • G01J2003/283Investigating the spectrum computer-interfaced
    • G01J2003/284Spectral construction
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00Road transport of goods or passengers
    • Y02T10/10Internal combustion engine [ICE] based vehicles
    • Y02T10/40Engine management systems

Landscapes

  • Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Spectrometry And Color Measurement (AREA)

Abstract

The invention discloses a spectrum reconstruction method based on a calculation enhanced pixel spectral imaging chip, which relates to the field of spectrum imaging detectors. The spectrum reconstruction method based on the calculation enhanced pixel spectroscopic spectrum imaging chip comprises a calculation enhanced spectrum reconstruction mode and a calculation direct spectrum reconstruction mode, has high signal-to-noise ratio, wide spectrum range and high spectrum precision, and has the accuracy and precision of scientific metering level.

Description

Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip
Technical Field
The invention relates to the technical field of spectrum imaging detectors, in particular to a spectrum reconstruction method based on a calculation enhanced pixel spectral imaging chip.
Background
The optical filter for the spectrum imaging chip mainly comprises a pixel narrow-band optical filter and a pixel random optical filter, and the spectrum imaging chip based on the two optical filters has the characteristics of simple structure, small volume and small spectrum image distortion. When the spectrum imaging chip based on the two filters is used, separately, the spectrum imaging chip based on the pixel narrowband filter has the problems of low signal-to-noise ratio and narrow coverage spectrum range when the spectrum imaging chip based on the pixel narrowband filter is used because the light energy utilization rate of the pixel narrowband filter is low; in addition, when the spectrum imaging chip based on the pixel random filter is used, the broadband pixel random filter is firstly used for encoding an incident spectrum, and then an algorithm is used for multi-spectrum reconstruction, and although the spectrum imaging chip based on the pixel random filter has the advantages of high signal-to-noise ratio and wide single-chip coverage spectrum range when in use, a large amount of training data is required to achieve multi-spectrum reconstruction precision when the algorithm is used for multi-spectrum reconstruction, the data stability is insufficient, and the method reliability is reduced.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a spectrum reconstruction method based on a calculation enhanced pixel spectral imaging chip.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the invention relates to a calculation enhancement type pixel spectroscopic spectrum imaging chip which comprises a plurality of spectrum detection units, wherein each spectrum detection unit consists of a plurality of narrow-band optical filters and a plurality of random optical filters, and the plurality of narrow-band optical filters and the plurality of random optical filters are etched on the surface of a detector by using a photoetching process to form the spectrum detection units.
Further, each spectrum detection unit consists of 4 narrow-band filters and 12 random filters, wherein the 4 random filters are arranged in the center, and a circle of 12 narrow-band filters are arranged at the periphery of the random filters.
The invention relates to a spectrum reconstruction method based on a calculation enhanced pixel spectroscopic imaging chip, which is realized by adopting the calculation enhanced pixel spectroscopic imaging chip, and comprises the following steps of:
the original spectrum is subjected to narrow-band filter and random filter on the calculation enhanced pixel spectroscopic imaging chip
Figure SMS_1
Performing coding acquisition, sensing coded light intensity by a light intensity detector, performing analog-digital conversion on the light intensity to obtain corresponding light intensity digital quantity, inputting all the light intensity digital quantity into a spectrum reconstruction network for network training and optimization, and optimizing and adjusting network parameters of the spectrum reconstruction network to ensure that a reconstructed spectrum is #, is #, and the like>
Figure SMS_2
Continuously approximates the original spectrum +.>
Figure SMS_3
High-precision, wide-spectrum and high-signal-to-noise ratio spectrum information reconstruction is realized.
Further, the specific operation steps of the network training and optimizing are as follows:
first forward propagating to obtain a reconstructed spectrum
Figure SMS_4
Calculating a loss term value and its relation to the network parameter +.>
Figure SMS_5
According to which the network parameters are updated by a back propagation algorithm +.>
Figure SMS_6
Repeating the processThe process converges and approaches zero the drop in the loss term value, thereby achieving optimization of the spectral reconstruction network. />
Further, the network parameter optimization expression of the spectrum reconstruction network is as follows:
Figure SMS_7
in the method, in the process of the invention,
Figure SMS_10
for regularization factor, ++>
Figure SMS_14
For network parameters +.>
Figure SMS_17
Reconstructing spectrum +.>
Figure SMS_9
Is a regular term; spectral reconstruction network->
Figure SMS_12
The network parameter optimization expression of (2) is divided into two items,/->
Figure SMS_15
Reconstruction error term, representing reconstruction spectrum +.>
Figure SMS_18
Is +.>
Figure SMS_8
Fitting degree of (3); />
Figure SMS_13
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure SMS_16
Overfitting with +.>
Figure SMS_19
Regularization ofRestraint, make->
Figure SMS_11
The two terms are added to form a penalty term, which is optimized such that the penalty term value approaches zero.
The invention relates to a spectrum reconstruction method based on a calculation enhanced pixel spectroscopic imaging chip, which is realized by adopting the calculation enhanced pixel spectroscopic imaging chip, and comprises the following steps of calculating a straight-through spectrum reconstruction mode:
the original spectrum is subjected to narrow-band filter and random filter on the calculation enhanced pixel spectroscopic imaging chip
Figure SMS_20
Performing coding acquisition, namely sensing coded light intensity by a light intensity detector, and performing analog-digital conversion on the light intensity to obtain a corresponding light intensity digital quantity; inputting the light intensity digital quantity coded by the narrow-band filter into a spectrum restoration function to obtain a restored spectrum; inputting the light intensity digital quantity encoded by the random optical filter into a spectrum reconstruction network for network training and optimization, and optimizing and adjusting network parameters of the spectrum reconstruction network to reconstruct a spectrum +.>
Figure SMS_21
Continuously approximates the original spectrum +.>
Figure SMS_22
Finally, a high-precision recovery spectrum and a high-signal-to-noise specific gravity spectrum are obtained.
Further, the spectral restoration function is s1=g ([ I ] 1 ,I 12 ]) G is a Gaussian function.
Further, the specific operation steps of the network training and optimizing are as follows:
first forward propagating to obtain a reconstructed spectrum
Figure SMS_23
Calculating a loss term value and its relation to the network parameter +.>
Figure SMS_24
According to which the network parameters are updated by a back propagation algorithm +.>
Figure SMS_25
Repeating this process causes the loss term drop to converge and approach zero, thereby achieving optimization of the spectral reconstruction network.
Further, the network parameter optimization expression of the spectrum reconstruction network is as follows:
Figure SMS_26
in the method, in the process of the invention,
Figure SMS_28
for regularization factor, ++>
Figure SMS_33
For network parameters +.>
Figure SMS_36
Reconstructing spectrum +.>
Figure SMS_30
Is a regular term; spectral reconstruction network->
Figure SMS_31
The network parameter optimization expression of (2) is divided into two items,/->
Figure SMS_34
Reconstruction error term, representing reconstruction spectrum +.>
Figure SMS_37
Is +.>
Figure SMS_27
Fitting degree of (3); />
Figure SMS_32
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure SMS_35
Overfitting with +.>
Figure SMS_38
Regularizing the constraint such that ∈ ->
Figure SMS_29
The two terms are added to form a penalty term, which is optimized such that the penalty term value approaches zero.
The beneficial effects of the invention are as follows:
according to the spectrum reconstruction method based on the calculation enhanced pixel spectral imaging chip, the spectrum reconstruction is realized through designing the calculation enhanced pixel spectral imaging chip and the spectrum reconstruction mode, and the spectrum reconstruction method has high signal-to-noise ratio, wide spectrum range and high spectrum precision, and has the accuracy and precision of scientific metering level.
The spectrum reconstruction method based on the calculation enhanced pixel spectral imaging chip has two spectrum reconstruction modes, namely a calculation enhanced spectrum reconstruction mode and a calculation through spectrum reconstruction mode, and can solve the problems of low light energy utilization rate, low signal to noise ratio and narrow coverage spectrum range of the existing pixel narrowband filter-based spectrum imaging chip in use; the problems that a large amount of training data is needed to achieve the reconstruction accuracy of the multispectral spectrum, the data stability is insufficient and the reliability of the method is reduced when the spectrum imaging chip based on the pixel random filter is used can be solved.
Drawings
Fig. 1 is a schematic structural diagram of a spectrum detecting unit.
FIG. 2 is a schematic diagram of a spectral detection unit area array on a computational enhancement pixel spectral imaging chip.
Fig. 3 is a schematic diagram of a spectrum acquisition and reconstruction process in a computationally enhanced spectrum reconstruction mode.
Fig. 4 is a schematic diagram of a spectrum acquisition and reconstruction process in a calculation-through spectrum reconstruction mode.
In the figure: 1. a narrow-band filter and a random filter.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
The invention relates to a calculation enhanced pixel spectroscopic imaging chip which mainly comprises a plurality of spectroscopic detection units, wherein the spectroscopic detection units are mainly used for imaging and spectrum acquisition, and each spectroscopic detection unit is equivalent to one pixel of an imaging spectrometer.
The structure of each spectrum detection unit is shown in fig. 1, the spectrum detection unit mainly comprises a plurality of narrowband optical filters 1 and a plurality of random optical filters 2, the narrowband optical filters 1 and the random optical filters 2 are arranged according to a certain rule, in the embodiment, a preferable arrangement mode is shown in fig. 1, but the arrangement mode is not limited to the embodiment, specifically, 4 random optical filters 2 are arranged in the center, a circle of 12 narrowband optical filters 1 are arranged at the periphery of the random optical filters, and the spectrum detection unit can be formed by etching the narrowband optical filters 1 and the random optical filters 2 on the surface of a detector by using a photoetching process during manufacturing.
After the spectrum detection units are manufactured, the spectrum detection units are scanned or repeatedly arranged or periodically copied to form a calculation enhancement type pixel spectral imaging chip with different scales, as shown in fig. 2. The manufacturing method can enable the computational enhancement type pixel spectral imaging chip to traverse the whole view field, and realize spectral image acquisition of the whole view field.
In the calculation enhanced pixel spectral imaging chip, the spectral detection unit consists of the narrow-band optical filter and the random optical filter, and the combination of the spectral structures is not limited to the mode in fig. 1, so that the calculation enhanced pixel spectral imaging chip can acquire high-precision narrow-band spectral information and high-signal-to-noise ratio broadband spectral information at the same time.
The invention discloses a spectrum reconstruction method based on a calculation enhanced pixel spectroscopic spectrum imaging chip, which mainly comprises two spectrum reconstruction modes, namely a calculation enhanced spectrum reconstruction mode and a calculation straight-through spectrum reconstruction mode, wherein the two spectrum reconstruction modes are respectively described below.
1. Computing an enhanced spectral reconstruction pattern
The process of spectrum acquisition and reconstruction in the calculation enhanced spectrum reconstruction mode is shown in fig. 3. The original spectrum S is encoded and collected by utilizing narrow-band filters (FP 1 to FP 12) and random filters (RF 1 to RF 4) on a computational enhancement type pixel spectroscopic imaging chip, and the light intensity i after being encoded is sensed by a light intensity detector 1 To i 16 Then for the light intensity i 1 To i 16 Respectively performing analog-digital conversion (ADC) to obtain corresponding light intensity digital quantity I 1 To I 16 The method comprises the steps of carrying out a first treatment on the surface of the Digital quantity I of light intensity encoded by narrowband filters (FP 1 to FP 12) and random filters (RF 1 to RF 4) 1 To I 16 Together with the input to the spectrum reconstruction network
Figure SMS_39
(the spectrum reconstruction network->
Figure SMS_40
The system consists of 5 layers of full-connection layers, namely an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer; the input size of the input layer is 16; the input size of the first hidden layer is 300, and the output size of the first hidden layer is 500; the input size of the second hidden layer is 500, and the output size of the second hidden layer is 800; the input size of the third hidden layer is 800, and the output size of the third hidden layer is 500; the input size of the output layer is 500, and the output size of the output layer is 100) is subjected to network training and optimization to obtain a reconstructed spectrum +.>
Figure SMS_41
Thereby realizing high-precision, wide-spectrum and high signal-to-noise ratio spectrum information reconstruction.
The specific operation steps of the spectrum reconstruction network training and optimizing are as follows:
digital quantity of light intensity I 1 To I 16 Together with the input to the spectrum reconstruction network
Figure SMS_42
The reconstructed spectrum +.>
Figure SMS_43
By optimizing the tuning of the spectral reconstruction network +.>
Figure SMS_44
Network parameters of (a) such that the spectrum is reconstructed +.>
Figure SMS_45
Continuously approximates the original spectrum +.>
Figure SMS_46
. Spectrum reconstruction network
Figure SMS_47
The network parameter optimization expression of (a) is as follows:
Figure SMS_48
in the method, in the process of the invention,
Figure SMS_49
for regularization factor, ++>
Figure SMS_50
For network parameters +.>
Figure SMS_51
Reconstructing spectrum +.>
Figure SMS_52
Is a regular term.
Spectrum reconstruction network
Figure SMS_54
The network parameter optimization expression of (2) can be divided into two items, the former item +.>
Figure SMS_57
Reconstruction error term, representing reconstruction spectrum +.>
Figure SMS_59
Is +.>
Figure SMS_55
Fitting degree of (3); the latter item->
Figure SMS_56
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure SMS_58
Overfitting is generally carried out using +.>
Figure SMS_60
Regularizing the constraint such that ∈ ->
Figure SMS_53
The two terms are added to form a penalty term, which is optimized such that the penalty term value approaches zero.
The specific optimization process comprises the following steps: first, forward propagation results in a reconstructed spectrum
Figure SMS_61
Then calculates the loss term value and its corresponding network parameter +.>
Figure SMS_62
According to which the network parameters are updated by a back propagation algorithm +.>
Figure SMS_63
Repeating this process so that the loss term value drop converges and approaches zero, thereby realizing a spectrum reconstruction network +.>
Figure SMS_64
Is described.
2. Computing through spectral reconstruction patterns
The process of spectrum acquisition and reconstruction in the calculation-through spectrum reconstruction mode is shown in FIG. 4, and the original spectrum is obtained by using narrow-band filters (FP 1 to FP 12) and random filters (RF 1 to RF 4) on the calculation-enhanced pixel spectral imaging chipS carries out coding acquisition, and the light intensity i after the coding is sensed by a light intensity detector 1 To i 16 Then for the light intensity i 1 To i 16 Respectively performing analog-digital conversion (ADC) to obtain corresponding light intensity digital quantity I 1 To I 16 The method comprises the steps of carrying out a first treatment on the surface of the Light intensity digital quantity I after encoding narrow-band filters (FP 1 to FP 12) 1 To I 12 Input to spectral restoration function s1=g ([ I ] 1 ,I 12 ]) Wherein G is a Gaussian function to obtain a restored spectrum S 1 The method comprises the steps of carrying out a first treatment on the surface of the Light intensity digital quantity I after coding random filters (RF 1 to RF 4) 13 To I 16 Input to a spectral reconstruction network
Figure SMS_65
Is subjected to network training and optimization to obtain a reconstructed spectrum +.>
Figure SMS_66
Finally, a high-precision recovery spectrum and a high-signal-to-noise specific gravity spectrum are obtained.
The specific operation steps of the spectrum reconstruction network training and optimizing are as follows:
digital quantity of light intensity I 13 To I 16 Together with the input to the spectrum reconstruction network
Figure SMS_67
The reconstructed spectrum +.>
Figure SMS_68
By optimizing the tuning of the spectral reconstruction network +.>
Figure SMS_69
Network parameters of (a) such that the spectrum is reconstructed +.>
Figure SMS_70
Continuously approximates the original spectrum +.>
Figure SMS_71
. Spectrum reconstruction network
Figure SMS_72
Network parameter optimization expression of (a)The following are provided:
Figure SMS_73
in the method, in the process of the invention,
Figure SMS_74
for regularization factor, ++>
Figure SMS_75
For network parameters +.>
Figure SMS_76
Reconstructing spectrum +.>
Figure SMS_77
Is a regular term.
Spectrum reconstruction network
Figure SMS_78
The network parameter optimization expression of (2) can be divided into two items, the former item +.>
Figure SMS_82
Reconstruction error term, representing reconstruction spectrum +.>
Figure SMS_84
Is +.>
Figure SMS_80
Fitting degree of (3); the latter item->
Figure SMS_81
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure SMS_83
Overfitting is generally carried out using +.>
Figure SMS_85
Regularizing the constraint such that ∈ ->
Figure SMS_79
Two, twoThe term addition constitutes a penalty term, optimized such that the penalty term value approaches zero.
The specific optimization process comprises the following steps: first, forward propagation results in a reconstructed spectrum
Figure SMS_86
Then calculates the loss term value and its corresponding network parameter +.>
Figure SMS_87
According to which the network parameters are updated by a back propagation algorithm +.>
Figure SMS_88
Repeating this process so that the loss term value drop converges and approaches zero, thereby realizing a spectrum reconstruction network +.>
Figure SMS_89
Is described.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.

Claims (9)

1. The calculation enhancement type pixel spectroscopic spectrum imaging chip is characterized by comprising a plurality of spectrum detection units, wherein each spectrum detection unit consists of a plurality of narrow-band optical filters and a plurality of random optical filters, and the plurality of narrow-band optical filters and the plurality of random optical filters are etched on the surface of a detector by using a photoetching process to form the spectrum detection unit.
2. The computational enhancement pixel spectral imaging chip according to claim 1, wherein each spectral detection unit consists of 4 narrowband filters and 12 random filters, the 4 random filters are arranged in the center, and a circle of 12 narrowband filters are arranged at the periphery.
3. The spectrum reconstruction method based on the calculation enhanced pixel spectroscopic imaging chip is characterized by being realized by adopting the calculation enhanced pixel spectroscopic imaging chip as claimed in claim 1 or 2, and comprises the following steps of:
the original spectrum is subjected to narrow-band filter and random filter on the calculation enhanced pixel spectroscopic imaging chip
Figure QLYQS_1
Performing coding acquisition, sensing coded light intensity by a light intensity detector, performing analog-digital conversion on the light intensity to obtain corresponding light intensity digital quantity, inputting all the light intensity digital quantity into a spectrum reconstruction network for network training and optimization, and optimizing and adjusting network parameters of the spectrum reconstruction network to ensure that a reconstructed spectrum is #, is #, and the like>
Figure QLYQS_2
Continuously approximates the original spectrum +.>
Figure QLYQS_3
High-precision, wide-spectrum and high-signal-to-noise ratio spectrum information reconstruction is realized.
4. The spectrum reconstruction method based on the computational enhancement pixel spectroscopic imaging chip as set forth in claim 3, wherein the specific operation steps of the network training and optimization are as follows:
first forward propagating to obtain a reconstructed spectrum
Figure QLYQS_4
Calculating a loss term value and its relation to the network parameter +.>
Figure QLYQS_5
According to which the network parameters are updated by a back propagation algorithm +.>
Figure QLYQS_6
Repeating this process causes the term to be lostThe value dip converges and approaches zero, thereby achieving optimization of the spectral reconstruction network.
5. The spectrum reconstruction method based on the computational enhancement pixel spectroscopic imaging chip as set forth in claim 3, wherein the network parameter optimization expression of the spectrum reconstruction network is as follows:
Figure QLYQS_10
the method comprises the steps of carrying out a first treatment on the surface of the In (1) the->
Figure QLYQS_14
For regularization factor, ++>
Figure QLYQS_17
For network parameters +.>
Figure QLYQS_7
Reconstructing spectrum +.>
Figure QLYQS_11
Is a regular term; spectral reconstruction network->
Figure QLYQS_15
The network parameter optimization expression of (2) is divided into two items,/->
Figure QLYQS_18
Reconstruction error term, representing reconstruction spectrum +.>
Figure QLYQS_9
Is +.>
Figure QLYQS_12
Fitting degree of (3); />
Figure QLYQS_16
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure QLYQS_19
Overfitting with +.>
Figure QLYQS_8
Regularizing the constraint such that ∈ ->
Figure QLYQS_13
The two terms are added to form a penalty term, which is optimized such that the penalty term value approaches zero.
6. The spectrum reconstruction method based on the calculation enhanced pixel spectroscopic imaging chip is characterized by being realized by adopting the calculation enhanced pixel spectroscopic imaging chip as claimed in claim 1 or 2, and comprises the following steps of:
the original spectrum is subjected to narrow-band filter and random filter on the calculation enhanced pixel spectroscopic imaging chip
Figure QLYQS_20
Performing coding acquisition, namely sensing coded light intensity by a light intensity detector, and performing analog-digital conversion on the light intensity to obtain a corresponding light intensity digital quantity; inputting the light intensity digital quantity coded by the narrow-band filter into a spectrum restoration function to obtain a restored spectrum; inputting the light intensity digital quantity encoded by the random optical filter into a spectrum reconstruction network for network training and optimization, and optimizing and adjusting network parameters of the spectrum reconstruction network to reconstruct a spectrum +.>
Figure QLYQS_21
Continuously approximates the original spectrum +.>
Figure QLYQS_22
Finally, a high-precision recovery spectrum and a high-signal-to-noise specific gravity spectrum are obtained.
7. The computationally-enhanced pixel-based spectroscopic method according to claim 6The spectrum reconstruction method of the spectrum imaging chip is characterized in that the spectrum restoration function is S1=G ([ I) 1 ,I 12 ]) G is a Gaussian function.
8. The spectrum reconstruction method based on the computational enhancement pixel spectroscopic imaging chip as claimed in claim 6, wherein the specific operation steps of the network training and optimization are as follows:
first forward propagating to obtain a reconstructed spectrum
Figure QLYQS_23
Calculating a loss term value and its relation to the network parameter +.>
Figure QLYQS_24
According to which the network parameters are updated by a back propagation algorithm +.>
Figure QLYQS_25
Repeating this process causes the loss term drop to converge and approach zero, thereby achieving optimization of the spectral reconstruction network.
9. The spectrum reconstruction method based on the computational enhancement pixel spectroscopic imaging chip as set forth in claim 6, wherein the network parameter optimization expression of the spectrum reconstruction network is as follows:
Figure QLYQS_27
the method comprises the steps of carrying out a first treatment on the surface of the In (1) the->
Figure QLYQS_32
For regularization factor, ++>
Figure QLYQS_35
For network parameters +.>
Figure QLYQS_26
Reconstructing spectrum +.>
Figure QLYQS_30
Is a regular term; spectral reconstruction network->
Figure QLYQS_34
The network parameter optimization expression of (2) is divided into two items,/->
Figure QLYQS_37
Reconstruction error term, representing reconstruction spectrum +.>
Figure QLYQS_29
Is +.>
Figure QLYQS_33
Fitting degree of (3); />
Figure QLYQS_36
The regular term is a network parameter constraint term for preventing spectral reconstruction network->
Figure QLYQS_38
Overfitting with +.>
Figure QLYQS_28
Regularizing the constraint such that ∈ ->
Figure QLYQS_31
The two terms are added to form a penalty term, which is optimized such that the penalty term value approaches zero. />
CN202310456976.9A 2023-04-26 2023-04-26 Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip Active CN116164841B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310456976.9A CN116164841B (en) 2023-04-26 2023-04-26 Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310456976.9A CN116164841B (en) 2023-04-26 2023-04-26 Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip

Publications (2)

Publication Number Publication Date
CN116164841A true CN116164841A (en) 2023-05-26
CN116164841B CN116164841B (en) 2023-07-07

Family

ID=86418612

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310456976.9A Active CN116164841B (en) 2023-04-26 2023-04-26 Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip

Country Status (1)

Country Link
CN (1) CN116164841B (en)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110175971A (en) * 2019-05-27 2019-08-27 大连海事大学 A kind of deep learning image reconstructing method of multispectral single pixel imaging
CN110324546A (en) * 2018-03-28 2019-10-11 黑魔法设计私人有限公司 Image processing method and filter array
CN111999788A (en) * 2020-09-28 2020-11-27 浙江大学 Wide spectrum coding full color filter array
CN112881308A (en) * 2021-01-22 2021-06-01 浙江大学 Spectrum camera based on wide spectrum coding and deep learning
CN114912598A (en) * 2021-02-08 2022-08-16 清华大学 Optical artificial neural network intelligent chip and preparation method thereof
CN115183870A (en) * 2022-07-29 2022-10-14 同济大学 Broadband coding optical filter array and snapshot type hyperspectral camera
CN115452735A (en) * 2022-09-15 2022-12-09 国科大杭州高等研究院 Ultramicro area spectrum test system and method
CN115731191A (en) * 2022-11-22 2023-03-03 国科大杭州高等研究院 Narrow-band spectral imaging method based on neural network
CN115755257A (en) * 2022-11-10 2023-03-07 国科大杭州高等研究院 Wide-spectrum coding array optical filter and preparation method thereof
CN115855250A (en) * 2022-12-21 2023-03-28 中国科学院长春光学精密机械与物理研究所 Spectrum imaging system based on deep learning spectrum coding and decoding

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110324546A (en) * 2018-03-28 2019-10-11 黑魔法设计私人有限公司 Image processing method and filter array
CN110175971A (en) * 2019-05-27 2019-08-27 大连海事大学 A kind of deep learning image reconstructing method of multispectral single pixel imaging
CN111999788A (en) * 2020-09-28 2020-11-27 浙江大学 Wide spectrum coding full color filter array
CN112881308A (en) * 2021-01-22 2021-06-01 浙江大学 Spectrum camera based on wide spectrum coding and deep learning
CN114912598A (en) * 2021-02-08 2022-08-16 清华大学 Optical artificial neural network intelligent chip and preparation method thereof
CN115183870A (en) * 2022-07-29 2022-10-14 同济大学 Broadband coding optical filter array and snapshot type hyperspectral camera
CN115452735A (en) * 2022-09-15 2022-12-09 国科大杭州高等研究院 Ultramicro area spectrum test system and method
CN115755257A (en) * 2022-11-10 2023-03-07 国科大杭州高等研究院 Wide-spectrum coding array optical filter and preparation method thereof
CN115731191A (en) * 2022-11-22 2023-03-03 国科大杭州高等研究院 Narrow-band spectral imaging method based on neural network
CN115855250A (en) * 2022-12-21 2023-03-28 中国科学院长春光学精密机械与物理研究所 Spectrum imaging system based on deep learning spectrum coding and decoding

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
肖树林等: "像元映射变分辨率光谱成像重构", 《中国光学》, vol. 15, no. 05 *

Also Published As

Publication number Publication date
CN116164841B (en) 2023-07-07

Similar Documents

Publication Publication Date Title
CN111626317B (en) Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network
Piskunov et al. New algorithms for reducing cross-dispersed echelle spectra
CN112881308B (en) Spectrum camera based on wide spectrum coding and deep learning
Ma et al. Extensions of compressed imaging: flying sensor, coded mask, and fast decoding
CN114998213B (en) Runoff weight model-based lake and reservoir flood retention amount calculation method
CN116164841B (en) Spectrum reconstruction method based on calculation enhanced pixel spectroscopic imaging chip
CN114419463B (en) Cloud platform-based global solar photovoltaic panel remote sensing automatic identification method
CN117571128B (en) High-resolution polarized spectrum image imaging method and system
CN115183870A (en) Broadband coding optical filter array and snapshot type hyperspectral camera
Zhou et al. RGB-to-HSV: A frequency-spectrum unfolding network for spectral super-resolution of RGB videos
Tamilselvan et al. Survey and analysis of various image fusion techniques for clinical CT and MRI images
CN113702666A (en) Signal joint noise reduction method for fiber optic gyroscope inertial measurement unit
Molnar et al. Spectral deconvolution with deep learning: removing the effects of spectral PSF broadening
CN116433548A (en) Hyperspectral and panchromatic image fusion method based on multistage information extraction
CN108764097A (en) A kind of high-spectrum remote sensing target identification method based on segmentation rarefaction representation
CN114418911B (en) Method for reducing scale and improving water body definition through statistical regression of remote sensing images
CN104182953A (en) Normalized multi-end-member decomposition hyperspectral reconstruction method in pixel unmixing inverse process
Guo et al. An improved generative adversarial network for remote sensing image super‐resolution
CN116242484B (en) Spectrum imaging chip and spectrum reconstruction algorithm collaborative design method
EP4270941A1 (en) Imaging device and optical element
US20240171870A1 (en) Filter array, photodetector, and photodetection system
Zhu et al. Spectrum reconstruction of multispectral light field imager based on adaptive sparse representation
US20240121488A1 (en) Imaging device and optical element
CN108111759A (en) Towards the emulation design method of area array CCD opto-electronic conversion
Lang et al. Ionospheric ionogram denoising based on robust principal component analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant