TWI818101B - Spectrum analysis device and spectrum analysis method - Google Patents
Spectrum analysis device and spectrum analysis method Download PDFInfo
- Publication number
- TWI818101B TWI818101B TW108137990A TW108137990A TWI818101B TW I818101 B TWI818101 B TW I818101B TW 108137990 A TW108137990 A TW 108137990A TW 108137990 A TW108137990 A TW 108137990A TW I818101 B TWI818101 B TW I818101B
- Authority
- TW
- Taiwan
- Prior art keywords
- spectrum
- analysis
- neural network
- data
- input
- Prior art date
Links
- 238000010183 spectrum analysis Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000004458 analytical method Methods 0.000 claims abstract description 132
- 238000001228 spectrum Methods 0.000 claims abstract description 69
- 238000013528 artificial neural network Methods 0.000 claims abstract description 63
- 239000000203 mixture Substances 0.000 claims description 12
- 239000013558 reference substance Substances 0.000 claims description 9
- 230000000306 recurrent effect Effects 0.000 abstract description 2
- 238000001237 Raman spectrum Methods 0.000 description 29
- 238000010586 diagram Methods 0.000 description 16
- 230000000052 comparative effect Effects 0.000 description 15
- WHUUTDBJXJRKMK-UHFFFAOYSA-N Glutamic acid Natural products OC(=O)C(N)CCC(O)=O WHUUTDBJXJRKMK-UHFFFAOYSA-N 0.000 description 14
- 235000004279 alanine Nutrition 0.000 description 12
- QNAYBMKLOCPYGJ-REOHCLBHSA-N L-alanine Chemical compound C[C@H](N)C(O)=O QNAYBMKLOCPYGJ-REOHCLBHSA-N 0.000 description 11
- 235000001014 amino acid Nutrition 0.000 description 11
- 229940024606 amino acid Drugs 0.000 description 11
- 150000001413 amino acids Chemical class 0.000 description 11
- 239000000843 powder Substances 0.000 description 11
- 235000013922 glutamic acid Nutrition 0.000 description 10
- 239000004220 glutamic acid Substances 0.000 description 10
- ZDXPYRJPNDTMRX-UHFFFAOYSA-N glutamine Natural products OC(=O)C(N)CCC(N)=O ZDXPYRJPNDTMRX-UHFFFAOYSA-N 0.000 description 10
- CIWBSHSKHKDKBQ-JLAZNSOCSA-N Ascorbic acid Chemical compound OC[C@H](O)[C@H]1OC(=O)C(O)=C1O CIWBSHSKHKDKBQ-JLAZNSOCSA-N 0.000 description 8
- 238000000513 principal component analysis Methods 0.000 description 8
- 239000011159 matrix material Substances 0.000 description 6
- WHUUTDBJXJRKMK-VKHMYHEASA-N L-glutamic acid Chemical compound OC(=O)[C@@H](N)CCC(O)=O WHUUTDBJXJRKMK-VKHMYHEASA-N 0.000 description 5
- ZDXPYRJPNDTMRX-VKHMYHEASA-N L-glutamine Chemical compound OC(=O)[C@@H](N)CCC(N)=O ZDXPYRJPNDTMRX-VKHMYHEASA-N 0.000 description 5
- 238000006073 displacement reaction Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 5
- 238000000491 multivariate analysis Methods 0.000 description 5
- DCXYFEDJOCDNAF-REOHCLBHSA-N L-asparagine Chemical compound OC(=O)[C@@H](N)CC(N)=O DCXYFEDJOCDNAF-REOHCLBHSA-N 0.000 description 4
- DHMQDGOQFOQNFH-UHFFFAOYSA-N Glycine Chemical compound NCC(O)=O DHMQDGOQFOQNFH-UHFFFAOYSA-N 0.000 description 3
- 238000001069 Raman spectroscopy Methods 0.000 description 3
- 125000003295 alanine group Chemical group N[C@@H](C)C(=O)* 0.000 description 3
- 239000012491 analyte Substances 0.000 description 3
- 239000013076 target substance Substances 0.000 description 3
- 239000004475 Arginine Substances 0.000 description 2
- DCXYFEDJOCDNAF-UHFFFAOYSA-N Asparagine Natural products OC(=O)C(N)CC(N)=O DCXYFEDJOCDNAF-UHFFFAOYSA-N 0.000 description 2
- CKLJMWTZIZZHCS-REOHCLBHSA-N L-aspartic acid Chemical compound OC(=O)[C@@H](N)CC(O)=O CKLJMWTZIZZHCS-REOHCLBHSA-N 0.000 description 2
- AGPKZVBTJJNPAG-WHFBIAKZSA-N L-isoleucine Chemical compound CC[C@H](C)[C@H](N)C(O)=O AGPKZVBTJJNPAG-WHFBIAKZSA-N 0.000 description 2
- ROHFNLRQFUQHCH-YFKPBYRVSA-N L-leucine Chemical compound CC(C)C[C@H](N)C(O)=O ROHFNLRQFUQHCH-YFKPBYRVSA-N 0.000 description 2
- COLNVLDHVKWLRT-QMMMGPOBSA-N L-phenylalanine Chemical compound OC(=O)[C@@H](N)CC1=CC=CC=C1 COLNVLDHVKWLRT-QMMMGPOBSA-N 0.000 description 2
- OUYCCCASQSFEME-QMMMGPOBSA-N L-tyrosine Chemical compound OC(=O)[C@@H](N)CC1=CC=C(O)C=C1 OUYCCCASQSFEME-QMMMGPOBSA-N 0.000 description 2
- KDXKERNSBIXSRK-UHFFFAOYSA-N Lysine Natural products NCCCCC(N)C(O)=O KDXKERNSBIXSRK-UHFFFAOYSA-N 0.000 description 2
- KZSNJWFQEVHDMF-UHFFFAOYSA-N Valine Natural products CC(C)C(N)C(O)=O KZSNJWFQEVHDMF-UHFFFAOYSA-N 0.000 description 2
- ODKSFYDXXFIFQN-UHFFFAOYSA-N arginine Natural products OC(=O)C(N)CCCNC(N)=N ODKSFYDXXFIFQN-UHFFFAOYSA-N 0.000 description 2
- 235000009582 asparagine Nutrition 0.000 description 2
- 229960001230 asparagine Drugs 0.000 description 2
- 238000011088 calibration curve Methods 0.000 description 2
- 239000012925 reference material Substances 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- JXYWFNAQESKDNC-BTJKTKAUSA-N (z)-4-hydroxy-4-oxobut-2-enoate;2-[(4-methoxyphenyl)methyl-pyridin-2-ylamino]ethyl-dimethylazanium Chemical compound OC(=O)\C=C/C(O)=O.C1=CC(OC)=CC=C1CN(CCN(C)C)C1=CC=CC=N1 JXYWFNAQESKDNC-BTJKTKAUSA-N 0.000 description 1
- 239000004471 Glycine Substances 0.000 description 1
- FFEARJCKVFRZRR-BYPYZUCNSA-N L-methionine Chemical compound CSCC[C@H](N)C(O)=O FFEARJCKVFRZRR-BYPYZUCNSA-N 0.000 description 1
- QIVBCDIJIAJPQS-VIFPVBQESA-N L-tryptophane Chemical compound C1=CC=C2C(C[C@H](N)C(O)=O)=CNC2=C1 QIVBCDIJIAJPQS-VIFPVBQESA-N 0.000 description 1
- KZSNJWFQEVHDMF-BYPYZUCNSA-N L-valine Chemical compound CC(C)[C@H](N)C(O)=O KZSNJWFQEVHDMF-BYPYZUCNSA-N 0.000 description 1
- ROHFNLRQFUQHCH-UHFFFAOYSA-N Leucine Natural products CC(C)CC(N)C(O)=O ROHFNLRQFUQHCH-UHFFFAOYSA-N 0.000 description 1
- 239000004472 Lysine Substances 0.000 description 1
- ONIBWKKTOPOVIA-UHFFFAOYSA-N Proline Natural products OC(=O)C1CCCN1 ONIBWKKTOPOVIA-UHFFFAOYSA-N 0.000 description 1
- MTCFGRXMJLQNBG-UHFFFAOYSA-N Serine Natural products OCC(N)C(O)=O MTCFGRXMJLQNBG-UHFFFAOYSA-N 0.000 description 1
- AYFVYJQAPQTCCC-UHFFFAOYSA-N Threonine Natural products CC(O)C(N)C(O)=O AYFVYJQAPQTCCC-UHFFFAOYSA-N 0.000 description 1
- 239000004473 Threonine Substances 0.000 description 1
- QIVBCDIJIAJPQS-UHFFFAOYSA-N Tryptophan Natural products C1=CC=C2C(CC(N)C(O)=O)=CNC2=C1 QIVBCDIJIAJPQS-UHFFFAOYSA-N 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 235000003704 aspartic acid Nutrition 0.000 description 1
- OQFSQFPPLPISGP-UHFFFAOYSA-N beta-carboxyaspartic acid Natural products OC(=O)C(N)C(C(O)=O)C(O)=O OQFSQFPPLPISGP-UHFFFAOYSA-N 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 235000018417 cysteine Nutrition 0.000 description 1
- XUJNEKJLAYXESH-UHFFFAOYSA-N cysteine Natural products SCC(N)C(O)=O XUJNEKJLAYXESH-UHFFFAOYSA-N 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 229960000310 isoleucine Drugs 0.000 description 1
- AGPKZVBTJJNPAG-UHFFFAOYSA-N isoleucine Natural products CCC(C)C(N)C(O)=O AGPKZVBTJJNPAG-UHFFFAOYSA-N 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 229930182817 methionine Natural products 0.000 description 1
- 238000010202 multivariate logistic regression analysis Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- COLNVLDHVKWLRT-UHFFFAOYSA-N phenylalanine Natural products OC(=O)C(N)CC1=CC=CC=C1 COLNVLDHVKWLRT-UHFFFAOYSA-N 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- OUYCCCASQSFEME-UHFFFAOYSA-N tyrosine Natural products OC(=O)C(N)CC1=CC=C(O)C=C1 OUYCCCASQSFEME-UHFFFAOYSA-N 0.000 description 1
- 239000004474 valine Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/27—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands using photo-electric detection ; circuits for computing concentration
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
本發明之頻譜分析裝置1為基於由包含複數個基準物中之任意1或2個以上之基準物之分析對象物產生之光之頻譜分析該分析對象物者，且具備處理部10、輸入部20、學習部30及分析部40。處理部10具有遞歸型神經網路(RNN)。輸入部20使基準物或分析對象物所產生之光之頻譜之資料逐一輸入至RNN之各胞。藉此，實現能夠進行高效率且高精度之頻譜分析之裝置及方法。The spectrum analysis device 1 of the present invention analyzes an analysis target including any one or two or more reference objects among a plurality of reference objects based on the spectrum of the light generated by the analysis target, and is provided with a processing unit 10 and an input unit. 20. Learning Department 30 and Analysis Department 40. The processing unit 10 has a recurrent neural network (RNN). The input unit 20 inputs the data of the light spectrum generated by the reference object or the analysis object to each cell of the RNN one by one. Thereby, a device and method capable of performing high-efficiency and high-precision spectrum analysis are realized.
Description
本揭示係關於一種基於由分析對象物產生之光之頻譜對該分析對象物進行分析之裝置及方法。The present disclosure relates to a device and a method for analyzing an analysis object based on the spectrum of light generated by the analysis object.
由分析對象物產生之光之頻譜具有與分析對象物所含之成分之種類或比例相應之形狀。因此，可基於由分析對象物產生之光之頻譜分析該分析對象物。於由分析對象物產生之光之頻譜中，包含根據對分析對象物之光照射而由該分析對象物產生之光(例如由反射光、透射光、散射光、螢光、非線性光學現象(例如拉曼散射等)產生之光)之頻譜，又，包含藉由分析對象物中之化學反應而產生之化學發光之頻譜。再者，於光之頻譜中，亦包含自透射光或反射光獲得之折射率或吸收係數之頻譜。此處所言之光並未限定於紫外光、可視光、紅外光，亦包含例如太赫茲波等。The spectrum of light generated by the object of analysis has a shape corresponding to the types or proportions of components contained in the object of analysis. Therefore, the analysis object can be analyzed based on the spectrum of the light generated by the analysis object. The spectrum of light generated by an analysis object includes light generated by the analysis object upon irradiation of light to the analysis object (for example, reflected light, transmitted light, scattered light, fluorescence, and nonlinear optical phenomena ( For example, the spectrum of light produced by Raman scattering, etc., and the spectrum of chemiluminescence produced by chemical reactions in the analyzed object. Furthermore, the spectrum of light also includes the spectrum of the refractive index or absorption coefficient obtained from transmitted light or reflected light. The light mentioned here is not limited to ultraviolet light, visible light, and infrared light, but also includes, for example, terahertz waves.
先前，可於進行此種頻譜分析時使用多變量解析。作為多變量解析，已知有使用主成分分析、分類器、回歸分析等，並將該等組合之解析技術。又，於專利文獻1中，有旨在使用深層神經網路(Deep Neural Network)進行頻譜分析之暗示。若使用深層神經網路，可進行高效率且高精度之圖像辨識等(參照非專利文獻1)，因而只要可使用深層神經網路進行頻譜分析，即可期待與使用多變量解析之情形相比能夠進行更高效率且更高精度之分析。 [先前技術文獻] [專利文獻]Previously, it was possible to use multivariate analysis when performing this kind of spectral analysis. As multivariate analysis, analysis techniques that use principal component analysis, classifiers, regression analysis, etc. and combine these are known. Furthermore, Patent Document 1 suggests spectrum analysis using a deep neural network. If deep neural networks are used, high-efficiency and high-precision image recognition can be performed (see non-patent document 1). Therefore, as long as deep neural networks can be used for spectrum analysis, it can be expected to be similar to the case of using multi-variable analysis. It enables more efficient and more precise analysis. [Prior technical literature] [Patent Document]
[專利文獻1]日本專利特開2017-90130號公報 [非專利文獻][Patent Document 1] Japanese Patent Application Publication No. 2017-90130 [Non-patent literature]
[非專利文獻1]O. Russakovsky等人, "ImageNet Large Scale Visual Recognition Challenge", Int. J. Comput. Vis. 115, pp.211-252 (2015)[Non-patent document 1] O. Russakovsky et al., "ImageNet Large Scale Visual Recognition Challenge", Int. J. Comput. Vis. 115, pp.211-252 (2015)
[發明所欲解決之問題][Problem to be solved by the invention]
然而，於專利文獻1中，無與使用深層神經網路進行頻譜分析時之具體步驟相關之任何記載。又，於非專利文獻1中，無旨在使用深層神經網路進行頻譜分析之暗示。However, in Patent Document 1, there is no description of specific steps when performing spectrum analysis using a deep neural network. In addition, Non-Patent Document 1 does not suggest spectrum analysis using a deep neural network.
本發明之目的在於提供一種可進行高效率且高精度之頻譜分析之裝置及方法。 [解決問題之技術手段]The object of the present invention is to provide a device and method that can perform high-efficiency and high-precision spectrum analysis. [Technical means to solve problems]
本發明之實施形態為頻譜分析裝置。頻譜分析裝置係基於由包含複數個基準物中之任意1或2個以上之基準物之分析對象物產生之光之頻譜而分析該分析對象物者，且具備：(1)處理部，其具有由等效地鏈狀連接有複數個胞之模型表示之遞歸型神經網路；(2)輸入部，其使光之頻譜之資料逐一輸入至遞歸型神經網路之各胞；及(3)分析部，其於藉由輸入部使分析對象物所產生之光之頻譜之資料輸入至遞歸型神經網路時，基於自遞歸型神經網路輸出之資料對分析對象物進行分析。An embodiment of the present invention is a spectrum analysis device. The spectrum analysis device analyzes an analysis target object based on the spectrum of light generated by an analysis target object including any one or two or more reference objects among a plurality of reference objects, and is provided with: (1) a processing unit having A recursive neural network represented by a model that is equivalent to a plurality of cells connected in a chain; (2) an input part that inputs light spectrum data into each cell of the recursive neural network one by one; and (3) An analysis unit that analyzes the analysis target based on the data output from the recursive neural network when the data of the light spectrum generated by the analysis target is input to the recursive neural network through the input unit.
本發明之實施形態為頻譜分析方法。頻譜分析方法係基於由包含複數個基準物中之任意1或2個以上之基準物之分析對象物產生之光之頻譜分析該分析對象物之方法，且具備：(1)輸入步驟，其使光之頻譜之資料逐一輸入至由等效地鏈狀連接有複數個胞之模型表示之遞歸型神經網路之各胞；及(2)分析步驟，其於使分析對象物所產生之光之頻譜之資料於輸入步驟中輸入至遞歸型神經網路時，基於自遞歸型神經網路輸出之資料，對分析對象物進行分析。 [發明之效果]An embodiment of the present invention is a spectrum analysis method. The spectrum analysis method is a method of analyzing the analysis object based on the spectrum of the light generated by the analysis object including any one or more of a plurality of reference objects, and has: (1) an input step, which uses The data of the spectrum of light is input one by one into each cell of the recursive neural network represented by a model that is equivalent to a plurality of cells connected in a chain; and (2) the analysis step, which is to analyze the light generated by the object to be analyzed. When the spectrum data is input to the recursive neural network in the input step, the analysis object is analyzed based on the data output from the recursive neural network. [Effects of the invention]
根據本發明之實施形態，可進行高效率且高精度之頻譜分析。According to the embodiment of the present invention, high-efficiency and high-precision spectrum analysis can be performed.
以下，參照隨附圖式，詳細說明用以實施本發明之形態。另，圖式之說明中，對相同要件附加相同之符號，並省略重複之說明。本發明並非限定於該等例示者。Hereinafter, embodiments for implementing the present invention will be described in detail with reference to the accompanying drawings. In addition, in the description of the drawings, the same symbols are attached to the same elements, and repeated explanations are omitted. The present invention is not limited to these examples.
圖1係顯示頻譜分析裝置1之構成之圖。頻譜分析裝置1為基於由包含複數個基準物中任意1或2個以上之基準物之分析對象物產生之光之頻譜分析該分析對象物者，且具備處理部10、輸入部20、學習部30及分析部40。FIG. 1 is a diagram showing the structure of the spectrum analysis device 1. The spectrum analysis device 1 analyzes an analysis target object based on the spectrum of light generated by an analysis target object including any one or two or more reference objects among a plurality of reference objects, and includes a processing unit 10, an input unit 20, and a learning unit. 30 and Analysis Department 40.
處理部10具有作為深度神經網路之一種之遞歸型神經網路(RNN：Recurrent Neural Network)。RNN為使胞之輸出之一部分遞歸地輸入該胞者，且由等效地鏈狀連接有複數個胞之模型表示。The processing unit 10 has a Recurrent Neural Network (RNN) which is a type of deep neural network. RNN is a model that recursively inputs part of the output of a cell into the cell, and is represented by a model in which a plurality of cells are equivalently connected in a chain.
處理部10較佳具有RNN之一種即LSTM(Long Short-Term Memory：長短期記憶網路)網路。以下，將其簡稱為「LSTM」。又，處理部10亦較佳具有LSTM之一種即Stacked LSTM。處理部10亦可藉由CPU(Central Processing Unit：中央處理單元)進行RNN中之處理，但較佳藉由可進行更高速之處理之DSP(Digital Signal Processor：數位信號處理器)或GPU(Graphics Processing Unit：圖形處理單元)進行。The processing unit 10 preferably has an LSTM (Long Short-Term Memory) network, which is one type of RNN. Hereinafter, it will be referred to as "LSTM" for short. Furthermore, the processing unit 10 preferably has a Stacked LSTM, which is one type of LSTM. The processing unit 10 may also use a CPU (Central Processing Unit: Central Processing Unit) to perform processing in RNN, but it is preferably a DSP (Digital Signal Processor: Digital Signal Processor) or a GPU (Graphics Processing Unit) that can perform higher-speed processing. Processing Unit: graphics processing unit).
輸入部20使基準物或分析對象物所產生之光之頻譜之資料逐一輸入至RNN之各胞。頻譜為以波長、波數或頻率為參數之複數個資料之集合。輸入部20較佳以頻譜之峰值強度為特定值之方式將頻譜標準化，並使該標準化後之頻譜之資料輸入至RNN。The input unit 20 inputs the data of the light spectrum generated by the reference object or the analysis object to each cell of the RNN one by one. A spectrum is a collection of data with wavelength, wave number or frequency as parameters. The input unit 20 preferably normalizes the spectrum so that the peak intensity of the spectrum is a specific value, and inputs the data of the normalized spectrum into the RNN.
學習部30使複數個基準物各者所產生之光之頻譜之資料作為學習用資料藉由輸入部20輸入至RNN，使RNN學習。又，學習部30使包含複數個基準物中任意1或2個以上之基準物且混合比例已知之混合物所產生之光之頻譜之資料作為學習用資料藉由輸入部20輸入至RNN，並使用已知之混合比例使RNN學習。此種深度神經網路之學習稱為深度學習(Deep Learning)。The learning unit 30 inputs the data of the spectrum of light generated by each of the plurality of reference objects to the RNN through the input unit 20 as learning data, thereby causing the RNN to learn. In addition, the learning unit 30 inputs the data of the spectrum of light generated by a mixture containing any one or two or more reference objects among the plurality of reference objects with a known mixing ratio into the RNN through the input unit 20 as learning data, and uses The known mixing ratio enables RNN to learn. The learning of this kind of deep neural network is called deep learning.
分析部40於使分析對象物所產生之光之頻譜之資料藉由輸入部20輸入至RNN時，基於自RNN輸出之資料對分析對象物進行分析。分析部40基於自RNN輸出之資料將分析對象物分類為複數個基準物中之任一者。又，分析部40基於自RNN輸出之資料，求得分析對象物所含之基準物之混合比例。When the data of the light spectrum generated by the analysis target is input to the RNN through the input unit 20, the analysis unit 40 analyzes the analysis target based on the data output from the RNN. The analysis unit 40 classifies the analysis target object into any of a plurality of reference objects based on the data output from the RNN. Furthermore, the analysis unit 40 obtains the mixing ratio of the reference substance contained in the analysis target object based on the data output from the RNN.
包含LSTM及Stacked LSTM之RNN迄今亦用於時間序列資料及音頻資料等之處理。先前，對RNN之各胞輸入頻譜資料。例如，輸入資料為音頻資料時，對RNN之各胞輸入有某時刻之音頻頻譜(頻譜資料)。RNNs including LSTM and Stacked LSTM have so far been used to process time series data and audio data. Previously, spectral data was input to each cell of the RNN. For example, when the input data is audio data, the audio spectrum (spectrum data) at a certain time is input to each cell of the RNN.
與此相對，於本實施形態中，輸入部20使標量資料輸入至RNN之各胞。具體而言，使由分光器測定之光之頻譜包含N個資料D(1)～D(N)，且其中第n資料D(n)為第n通道之資料。此外，將RNN模型中鏈狀連接之複數個胞中之第n胞表示為C(n)。此時，輸入部20例如使第n資料D(n)輸入至第n胞C(n)。又，輸入部20亦可進行頻譜之資料之選出、微調、任意值之填補等。輸入部20較佳藉由數值解析領域中使用之插值方法(樣條插值、拉格朗日插值、阿克瑪(Akima)插值等)或圖像處理領域中使用之壓縮方法(小波(Wavelet)轉換、離散餘弦轉換等)，將學習用資料及分析對象資料設為彼此相同之個數。On the other hand, in this embodiment, the input unit 20 inputs scalar data to each cell of the RNN. Specifically, the light spectrum measured by the spectrometer includes N pieces of data D(1)˜D(N), and the n-th data D(n) is the data of the n-th channel. In addition, the nth cell among the plurality of chain-connected cells in the RNN model is represented as C(n). At this time, the input unit 20 inputs the n-th data D(n) to the n-th cell C(n), for example. In addition, the input unit 20 can also perform selection, fine-tuning, filling of arbitrary values, etc. of spectrum data. The input unit 20 preferably uses an interpolation method (spline interpolation, Lagrangian interpolation, Akima interpolation, etc.) used in the field of numerical analysis or a compression method (Wavelet) used in the image processing field. transformation, discrete cosine transform, etc.), set the learning data and analysis target data to the same number.
輸入部20一般可使第n資料D(n)輸入至第(n+δ)胞C(n+δ)。δ為0或正負之整數。δ亦可於依據學習部30之學習步驟及依據分析部40之分析步驟之兩者中設為特定值。δ亦可於學習步驟中設為特定值，於分析步驟中每次輸入分析對象資料時變化。The input unit 20 generally allows the n-th data D(n) to be input to the (n+δ)-th cell C(n+δ). δ is 0 or a positive or negative integer. δ may be set to a specific value in both the learning step of the learning unit 30 and the analysis step of the analysis unit 40 . δ can also be set to a specific value in the learning step and change each time the analysis target data is input in the analysis step.
頻譜分析裝置1亦可具備受理分析對象之頻譜之選擇、分析開始之指示及分析條件之選擇等之輸入裝置。輸入裝置為例如鍵盤或滑鼠等。又，頻譜分析裝置1亦可具備顯示分析結果等之顯示裝置。顯示裝置為例如液晶顯示器等。頻譜分析裝置1亦可具備記憶分析對象之頻譜及分析結果等之記憶裝置。頻譜分析裝置1可構成為包含電腦。The spectrum analysis device 1 may also be provided with an input device that accepts selection of a spectrum to be analyzed, an instruction to start analysis, selection of analysis conditions, and the like. The input device is, for example, a keyboard or a mouse. Furthermore, the spectrum analysis device 1 may be provided with a display device that displays analysis results and the like. The display device is, for example, a liquid crystal display. The spectrum analysis device 1 may also be equipped with a memory device that memorizes the spectrum of the analysis target, the analysis results, and the like. The spectrum analysis device 1 may be configured to include a computer.
使用此種頻譜分析裝置1之頻譜分析方法具備由輸入部20進行之輸入步驟、由學習部30進行之學習步驟、及由分析部40進行之分析步驟。即，於輸入步驟中，使基準物或分析對象物所產生之光之頻譜之資料逐一輸入至RNN之各胞。於學習步驟中，使複數個基準物各者所產生之光之頻譜之資料作為學習用資料輸入至RNN，使RNN學習。於分析步驟中，使分析對象物所產生之光之頻譜之資料作為分析對象資料輸入至RNN，且基於自RNN輸出之資料對分析對象物進行分析。The spectrum analysis method using this spectrum analysis device 1 includes an input step by the input unit 20 , a learning step by the learning unit 30 , and an analysis step by the analysis unit 40 . That is, in the input step, the data of the light spectrum generated by the reference object or the analysis object is input to each cell of the RNN one by one. In the learning step, the data of the spectrum of light generated by each of the plurality of reference objects is input to the RNN as learning data, so that the RNN learns. In the analysis step, the data of the light spectrum generated by the analysis object is input to the RNN as the analysis object data, and the analysis object is analyzed based on the data output from the RNN.
只要於學習步驟中先進行一次RNN之學習，以後便可重複進行分析步驟，因而不必於每次進行分析步驟時進行學習步驟。基於同樣之理由，若RNN學習完畢，則不需要學習部30。As long as the RNN is learned once in the learning step, the analysis step can be repeated later, so there is no need to perform the learning step every time the analysis step is performed. For the same reason, if the RNN learning is completed, the learning part 30 is not needed.
本實施形態中，因使頻譜之資料輸入至RNN而進行頻譜分析，故即使進行複雜分類之情形及進行大量頻譜分類之情形等，亦可穩定進行高效率且高精度之頻譜分析。又，本實施形態中，可使用RNN進行定量分析。In this embodiment, spectrum data is input to the RNN and spectrum analysis is performed. Therefore, efficient and high-precision spectrum analysis can be performed stably even when complex classification is performed or when a large number of spectrum classifications are performed. In addition, in this embodiment, RNN can be used for quantitative analysis.
接著，對第1～第4實施例進行說明。於各實施例中，使用以下20種氨基酸粉末作為基準物。使用包含該等20種氨基酸粉末中之任1種或2種氨基酸粉末者作為分析對象物。 丙胺酸(Ala)、精胺酸(Arg)、天門冬醯胺(Asn)、天門冬胺酸(Asp) 半胱胺酸(Cys)、穀氨醯胺(Gln)、穀胺酸(Glu)、甘胺酸(Gly) 組胺酸(His)、異亮胺酸(Ile)、亮胺酸(Leu)、賴胺酸(Lys) 蛋胺酸(Met)、苯丙胺酸(Phe)、脯胺酸(Pro)、絲胺酸(Ser) 蘇胺酸(Thr)、色胺酸(Trp)、酪胺酸(Tyr)、纈胺酸(Val)Next, the first to fourth embodiments will be described. In each example, the following 20 kinds of amino acid powders were used as benchmarks. A powder containing any one or two types of amino acid powders among these 20 types of amino acids was used as the analysis object. Alanine (Ala), arginine (Arg), asparagine (Asn), aspartic acid (Asp) Cysteine (Cys), Glutamine (Gln), Glutamic acid (Glu), Glycine (Gly) Histine (His), Isoleucine (Ile), Leucine (Leu), Lysine (Lys) Methionine (Met), Phenylalanine (Phe), Proline (Pro), Serine (Ser) Threonine (Thr), tryptophan (Trp), tyrosine (Tyr), valine (Val)
對基準物及分析對象物照射中心波長785 nm之雷射光，且將此時產生之拉曼散射光之強度按拉曼偏移量(波數)之各值測定，求得拉曼頻譜。對於各拉曼頻譜，以峰值強度為特定值之方式進行標準化，且使該標準化後之拉曼頻譜之資料輸入至RNN。向RNN輸入資料時，使拉曼頻譜之第n通道之資料即第n資料D(n)輸入至RNN之模型中鏈狀連接之複數個胞中之第n胞C(n)。The reference object and the object to be analyzed are irradiated with laser light with a central wavelength of 785 nm, and the intensity of the Raman scattered light generated at this time is measured according to each value of the Raman shift amount (wave number) to obtain the Raman spectrum. For each Raman spectrum, the peak intensity is normalized to a specific value, and the data of the normalized Raman spectrum is input to the RNN. When inputting data to the RNN, the data of the nth channel of the Raman spectrum, that is, the nth data D(n), is input to the nth cell C(n) among the plurality of chain-connected cells in the model of the RNN.
於第1實施例中，使用LSTM作為RNN，將分析對象物分類為20種基準物中之任一者。In the first embodiment, LSTM is used as the RNN to classify the analysis object into any one of 20 types of reference objects.
學習步驟中，關於作為基準物之各氨基酸粉末，使用50個拉曼頻譜作為學習用資料。學習用資料之總數為1000(=50×20種)。作為學習用資料使用之拉曼頻譜為信號雜訊比(SN：signal-to-noise ratio)較高者。例如，若說明丙胺酸，則使僅含丙胺酸之氨基酸粉末之拉曼頻譜之資料輸入至LSTM，且將丙胺酸之學習用標號設為值1，且將其他氨基酸粉末之學習標號設為值0，使LSTM學習。In the learning step, 50 Raman spectra of each amino acid powder as a reference substance are used as learning data. The total number of learning materials is 1000 (=50×20 types). The Raman spectrum used as a learning material is one with a higher signal-to-noise ratio (SN: signal-to-noise ratio). For example, if alanine is specified, the data of the Raman spectrum of the amino acid powder containing only alanine is input to the LSTM, and the learning label of alanine is set to the value 1, and the learning label of other amino acid powders is set to the value 0, enables LSTM to learn.
於分析步驟中，關於作為分析對象物之各氨基酸粉末，使用40個拉曼頻譜作為分析對象資料。分析對象資料之總數為800(=40×20種)。例如若說明丙胺酸，則準備SN不同之4種僅含丙胺酸之氨基酸粉末之拉曼頻譜，且將各拉曼頻譜之資料輸入至LSTM，並使輸出標號自LSTM輸出。In the analysis step, 40 Raman spectra are used as analysis target data for each amino acid powder as the analysis target. The total number of analysis target data is 800 (=40×20 types). For example, if alanine is specified, the Raman spectra of four types of amino acid powders containing only alanine with different SNs are prepared, and the data of each Raman spectrum is input to the LSTM, and the output label is output from the LSTM.
圖2係顯示作為基準物之各氨基酸粉末之拉曼頻譜之例之圖。圖3係顯示作為分析對象物之丙胺酸之拉曼頻譜之例之圖。該圖3顯示有SN比不同之4種拉曼頻譜。FIG. 2 is a diagram showing an example of the Raman spectrum of each amino acid powder used as a reference material. FIG. 3 is a diagram showing an example of the Raman spectrum of alanine as an analysis target substance. Figure 3 shows four types of Raman spectra with different SN ratios.
圖4係顯示LSTM之模型之例之圖。該圖係顯示有使將分析對象物設為丙胺酸時測定之拉曼頻譜之第n資料D(n)輸入至LSTM之模型之第n胞C(n)。又，該圖係顯示有自LSTM輸出之輸出標號，其中丙胺酸(Ala)為0.85，精胺酸(Arg)為0.05，天門冬醯胺(Asn)為0.01。於該圖之例中，因自LSTM輸出之各氨基酸粉末之輸出標號中丙胺酸之輸出標號之值最大，故分析對象物分類為丙胺酸。Figure 4 is a diagram showing an example of an LSTM model. This figure shows the n-th cell C(n) of the model in which the n-th data D(n) of the Raman spectrum measured when the analyte object is alanine is input to the LSTM. In addition, this figure shows the output labels output from the LSTM, among which alanine (Ala) is 0.85, arginine (Arg) is 0.05, and asparagine (Asn) is 0.01. In the example shown in the figure, among the output labels of each amino acid powder output from LSTM, the output label of alanine has the largest value, so the analysis target substance is classified as alanine.
於用以與第1實施例比較之第1比較例中，如下所示，藉由多變量解析而將分析對象物分類為20種基準物中之任一者。即，基於對學習用資料實施主成分分析(PCA：Principal Component Analysis)之結果，藉由支持向量機(SVM：Support Vector Machine)構成20組圖案識別器。PCA之主成分數為18，PCA之貢獻率為0.79。對利用SVM之圖案識別器輸入分析對象資料。In the first comparative example for comparison with the first embodiment, as shown below, the analysis target object is classified into any of the 20 reference substances by multivariate analysis. That is, based on the results of PCA (Principal Component Analysis) performed on the learning data, 20 sets of pattern recognizers were constructed using support vector machines (SVM: Support Vector Machine). The principal component of PCA is 18, and the contribution rate of PCA is 0.79. Input the analysis target data to the pattern recognizer using SVM.
圖5係顯示對第1實施例之分類結果加以表示之混同矩陣之圖。圖6係顯示對第1比較例之分類結果加以表示之混同矩陣之圖。第1實施例之分類中，正答率為98.5%。第1比較例之分類中，正答率為98.125%。第1實施例與第1比較例獲得同程度之分類精度。FIG. 5 is a diagram showing a confusion matrix representing the classification results of the first embodiment. FIG. 6 is a diagram showing a confusion matrix representing the classification results of the first comparative example. In the classification of the first embodiment, the correct answer rate was 98.5%. In the classification of the first comparative example, the correct answer rate was 98.125%. The first embodiment and the first comparative example achieved the same level of classification accuracy.
於第2實施例中，作為RNN使用Stacked LSTM，求出分析對象物所含之基準物之混合比例。In the second embodiment, Stacked LSTM is used as the RNN to obtain the mixing ratio of the reference substance contained in the analysis target object.
於學習步驟中，使用包含穀氨醯胺(Gln)及穀胺酸(Glu)之兩者或一者且混合比例已知之混合物之拉曼頻譜，作為學習用資料。準備有將穀氨醯胺與穀胺酸之混合比例(mo1比)設為x：(1-x)，且x以0.1之增量設於0～1之範圍內，從而具有11種混合比例之混合物之拉曼頻譜。另，x=1時穀氨醯胺為100%，x=0時穀胺酸為100%，但為求簡便，亦可稱為混合物。關於各混合比例，使用50個拉曼頻譜作為學習用資料。學習用資料之總數為550(=50×11種)。將該等拉曼頻譜之資料輸入至Stacked LSTM，且將學習用標號設為與混合比例相應之值，且使Stacked LSTM學習。In the learning step, the Raman spectrum of a mixture containing both or one of glutamine (Gln) and glutamic acid (Glu) with a known mixing ratio is used as learning data. It is prepared that the mixing ratio of glutamine and glutamic acid (mo1 ratio) is set to x: (1-x), and x is set in the range of 0 to 1 in increments of 0.1, so that there are 11 mixing ratios Raman spectrum of the mixture. In addition, when x=1, glutamine is 100%, and when x=0, glutamic acid is 100%. However, for simplicity, it can also be called a mixture. Regarding each mixing ratio, 50 Raman spectra were used as learning materials. The total number of study materials is 550 (=50×11 types). Input the data of the Raman spectrum into the Stacked LSTM, set the learning label to a value corresponding to the mixing ratio, and let the Stacked LSTM learn.
於分析步驟中，使用包含穀氨醯胺及穀胺酸之兩者或一者之混合物之拉曼頻譜，作為分析對象資料。將穀氨醯胺及穀胺酸之混合比例同樣設為11種。關於各混合比例，使用25個拉曼頻譜作為分析對象資料。分析對象資料之總數為275(=25×11種)。In the analysis step, a Raman spectrum containing a mixture of both or one of glutamine and glutamic acid is used as the analysis target data. The mixing ratio of glutamine and glutamic acid is also set to 11 types. Regarding each mixing ratio, 25 Raman spectra were used as analysis target data. The total number of analysis target data is 275 (=25×11 types).
圖7係顯示穀氨醯胺與穀胺酸之混合物之拉曼頻譜之例之圖。該圖係顯示作為已準備之與各混合比例相關之學習資料之拉曼頻譜。Figure 7 is a graph showing an example of the Raman spectrum of a mixture of glutamine and glutamic acid. This figure shows the Raman spectrum prepared as study material related to each mixing ratio.
圖8係顯示Stacked LSTM之模型之例之圖。Stacked LSTM之模型為多段構成LSTM之模型之各胞者。該圖顯示有將穀氨醯胺(G1n)與穀胺酸(Glu)之混合比例為0.60：0.40之混合物作為分析對象物，且將該分析對象物之拉曼頻譜之第n資料D(n)輸入至Stacked LSTM之模型之第n胞C(n)之情況。又，該圖顯示有自Stacked LSTM輸出之輸出標號係穀氨醯胺(G1n)為0.65，穀胺酸(Glu)為0.35。於該圖之例中，求得分析對象物之混合比例為0.65：0.35。Figure 8 is a diagram showing an example of a Stacked LSTM model. The model of Stacked LSTM is composed of multiple segments that form the cells of the LSTM model. This figure shows that a mixture of glutamine (G1n) and glutamic acid (Glu) in a ratio of 0.60:0.40 is used as the analyte, and the nth data D(n) of the Raman spectrum of the analyte is ) input to the nth cell C(n) of the Stacked LSTM model. Furthermore, this figure shows that the output labels from the Stacked LSTM are 0.65 for glutamine (G1n) and 0.35 for glutamic acid (Glu). In the example shown in the figure, the mixing ratio of the analysis object is found to be 0.65:0.35.
於用以與第2實施例比較之第2比較例中，如下所示，藉由多變量解析求得分析對象物所含之基準物之混合比例。即，基於對學習用資料實施PCA之結果且藉由多元回歸法(MLR：Multivariate Linear Regression)作成檢量線，並使用該檢量線進行定量。PCA之主成分數為54，PCA之貢獻率為0.864。In the second comparative example for comparison with the second example, as shown below, the mixing ratio of the reference substance contained in the analysis target object was obtained by multivariate analysis. That is, based on the results of performing PCA on the learning data, a calibration curve is created by a multiple regression method (MLR: Multivariate Linear Regression), and the calibration curve is used for quantification. The principal component of PCA is 54, and the contribution rate of PCA is 0.864.
圖9係顯示第2實施例之定量結果之圖。圖10係顯示第2比較例之定量結果之圖。該等圖顯示有真混合比例(橫軸)與定量結果之混合比例(縱軸)之關係。於該等圖中，四方圖形顯示學習用資料之定量結果，且白圓之圖形顯示有分析對象資料之定量結果。又，該等之圖所示之斜率為1之直線為理論直線。若正確進行學習，則圖形聚集於理論直線上。以均方根誤差評估真混合比例與定量結果之混合比例之差時，於第2實施例中為0.623，於第2比較例中為0.696。第2實施例與第2比較例獲得同程度之定量精度。Fig. 9 is a graph showing the quantitative results of the second example. Fig. 10 is a graph showing the quantitative results of the second comparative example. The graphs show the relationship between the true mixing ratio (horizontal axis) and the quantitative result mixing ratio (vertical axis). In these figures, the square graph shows the quantitative results of the learning data, and the white circle graph shows the quantitative results of the analysis target data. In addition, the straight line with a slope of 1 shown in these figures is a theoretical straight line. If learning is done correctly, the graphs converge on the theoretical straight line. When the difference between the true mixing ratio and the quantitative result mixing ratio was evaluated based on the root mean square error, it was 0.623 in the second example and 0.696 in the second comparative example. The second example and the second comparative example achieved the same level of quantitative accuracy.
於第3實施例中，使用一般RNN，將分析對象物分類為20種基準物中之任一者。於第3實施例中，使用與第1實施例之情形相同之學習用資料及分析對象資料，與第1實施例之情形同樣進行學習步驟及分析步驟。圖11係顯示對第3實施例之分類結果加以表示之混同矩陣之圖。於第3實施例之分類中，正答率為99.125%，與第1實施例之分類之正答率(98.5%)程度相同。In the third embodiment, a general RNN is used to classify the analysis object into any one of 20 types of reference objects. In the third embodiment, the same learning data and analysis target data as in the first embodiment are used, and the learning step and the analysis step are performed in the same way as in the first embodiment. FIG. 11 is a diagram showing a confusion matrix representing the classification results of the third embodiment. In the classification of the third embodiment, the correct answer rate is 99.125%, which is the same as the correct answer rate (98.5%) of the classification in the first embodiment.
於第4實施例中，使用LSTM作為RNN，將分析對象物分類為20種基準物中之任一者。於第4實施例中，使用與第1實施例之情形相同之學習用資料及分析對象資料，與第1實施例之情形同樣進行學習步驟及分析步驟。其中，於第4實施例中，於學習步驟中將拉曼頻譜之第n資料D(n)輸入至LSTM之模型之第n胞C(n)，於分析步驟中，將拉曼頻譜之第n資料D(n)輸入至LSTM之模型之第(n+δ)胞C(n+δ)，並將位移量δ設為0或正之各整數值。In the fourth embodiment, LSTM is used as the RNN to classify the analysis target object into any of 20 types of reference objects. In the fourth embodiment, the same learning data and analysis target data as in the first embodiment are used, and the learning step and the analysis step are performed in the same way as in the first embodiment. Among them, in the fourth embodiment, in the learning step, the nth data D(n) of the Raman spectrum is input to the nth cell C(n) of the LSTM model, and in the analysis step, the nth data D(n) of the Raman spectrum is input to the nth cell C(n) of the LSTM model. n data D(n) is input into the (n+δ)th cell C(n+δ) of the LSTM model, and the displacement δ is set to 0 or any positive integer value.
於用以與第4實施例比較之第4比較例中，與第1比較例之情形同樣，藉由多變量解析將分析對象物分類為20種基準物中之任一種。其中，於第4比較例中，對於圖案識別器構成時，於分析時賦予與第4實施例同樣之位移量δ。In the fourth comparative example for comparison with the fourth embodiment, the analysis target object is classified into any one of the 20 reference objects by multivariate analysis in the same manner as in the first comparative example. In the fourth comparative example, when the pattern recognizer is configured, the same displacement amount δ as in the fourth embodiment is given during analysis.
關於此處使用之拉曼頻譜，於波數330 cm-1 附近，每通道之波數差為2.31 cm-1 ，於波數1900 cm-1 附近，每通道之波數差為1.24 cm-1 。Regarding the Raman spectrum used here, near the wave number 330 cm -1 , the wave number difference per channel is 2.31 cm -1 , and near the wave number 1900 cm -1 , the wave number difference per channel is 1.24 cm -1 .
於第4實施例及第4比較例之任一者中，正答率隨著位移量δ變大而變低。與第4比較例相比，第4實施例之正答率之降低程度小。於第4實施例中，若位移量δ在5以下，則獲得足夠高之正答率。這意味著即便於波長校準不完全之狀態下測定頻譜，亦可藉由利用RNN進行頻譜分析而獲得高正答率。In either of the fourth embodiment and the fourth comparative example, the correct answer rate becomes lower as the displacement amount δ becomes larger. Compared with the fourth comparative example, the decrease in the correct answer rate of the fourth embodiment is small. In the fourth embodiment, if the displacement amount δ is less than 5, a sufficiently high correct answer rate can be obtained. This means that even when the spectrum is measured with incomplete wavelength calibration, a high correct answer rate can be obtained by using RNN for spectrum analysis.
本發明之頻譜分析裝置及頻譜分析方法並非限定於上述實施形態及構成例者，可進行各種變化。The spectrum analysis device and spectrum analysis method of the present invention are not limited to the above-described embodiments and configuration examples, and can be modified in various ways.
上述實施形態之頻譜分析裝置係基於由包含複數個基準物中任意1或2個基準物之分析對象物產生之光之頻譜而分析該分析對象物者；且構成為具備：(1)處理部，其具有由等效地鏈狀連接有複數個胞之模型表示之遞歸型神經網路；(2)輸入部，其使光之頻譜之資料逐一輸入至遞歸型神經網路之各胞；及(3)分析部，其於藉由輸入部將分析對象物產生之光之頻譜之資料輸入至遞歸型神經網路時，基於自遞歸型神經網路輸出之資料對分析對象物進行分析。The spectrum analysis device of the above embodiment analyzes an analysis target object based on the spectrum of light generated by the analysis target object including any one or two of a plurality of reference objects; and is configured to include: (1) a processing unit , which has a recursive neural network represented by a model that is equivalent to a plurality of cells connected in a chain; (2) the input part, which inputs the light spectrum data to each cell of the recursive neural network one by one; and (3) An analysis unit that analyzes the analysis target based on the data output from the recursive neural network when the data on the spectrum of light generated by the analysis target is input to the recursive neural network through the input unit.
於上述構成之分析裝置中，亦可構成為處理部具有LSTM網路作為遞歸型神經網路。又，於上述構成之分析裝置中，亦可構成為輸入部以頻譜之峰值強度為特定值之方式使頻譜標準化，且使該標準化後之頻譜之資料輸入至遞歸型神經網路。In the analysis device configured as described above, the processing unit may be configured to include an LSTM network as a recursive neural network. Furthermore, in the analysis device configured as described above, the input unit may be configured to normalize the spectrum so that the peak intensity of the spectrum becomes a specific value, and input the data of the normalized spectrum to the recursive neural network.
上述構成之分析裝置亦可構成為進而具備：學習部，其使複數個基準物之各者所產生之光之頻譜之資料藉由輸入部輸入至遞歸型神經網路，且使遞歸型神經網路學習。該情形時，亦可構成為，學習部使包含複數個基準物中之任意1或2個以上之基準物且混合比例已知之混合物所產生之光之頻譜之資料藉由輸入部輸入至遞歸型神經網路，使用混合比例使遞歸型神經網路學習。The analysis device configured as described above may further include a learning unit that inputs the data of the spectrum of light generated by each of the plurality of reference objects to the recursive neural network through the input unit, and causes the recursive neural network to road to learning. In this case, the learning unit may also be configured such that the data of the spectrum of light generated by a mixture including any one or two or more reference objects among the plurality of reference objects and with a known mixing ratio is input to the recursive type through the input unit. Neural Networks, using mixing ratios to enable recursive neural network learning.
於上述構成之分析裝置中，亦可構成為分析部基於自遞歸型神經網路輸出之資料將分析對象物分類為複數個基準物中之任一者。又，於上述構成之分析裝置中，亦可構成為分析部基於自遞歸型神經網路輸出之資料，求得分析對象物所含之基準物之混合比例。In the analysis device configured as described above, the analysis unit may be configured to classify the analysis target object into any one of a plurality of reference objects based on the data output from the recursive neural network. Furthermore, in the analysis device configured as described above, the analysis unit may be configured to obtain the mixing ratio of the reference substance contained in the analysis target based on the data output from the recursive neural network.
上述實施形態之頻譜分析方法係基於由包含複數個基準物中之任意1或2個以上之基準物之分析對象物產生之光之頻譜分析該分析對象物之方法，且構成為具備：(1)輸入步驟，其使光之頻譜之資料逐一輸入至由等效地鏈狀連接有複數個胞之模型所表示之遞歸型神經網路之各胞；及(2)分析步驟，其於使分析對象物所產生之光之頻譜之資料於輸入步驟中輸入至遞歸型神經網路時，基於自遞歸型神經網路輸出之資料，對分析對象物進行分析。The spectrum analysis method of the above embodiment is a method of analyzing an analysis target object based on the spectrum of the light generated by the analysis target object including any one or two or more reference objects among a plurality of reference objects, and is configured to include: (1) ) the input step, which inputs the light spectrum data one by one into each cell of the recursive neural network represented by a model that is equivalent to a plurality of cells connected in a chain; and (2) the analysis step, which enables the analysis When the data of the light spectrum generated by the object is input to the recursive neural network in the input step, the analysis object is analyzed based on the data output from the recursive neural network.
於上述構成之分析方法中，亦可構成為使用LSTM網路作為遞歸型神經網路。又，於上述構成之分析方法中，亦可構成為於輸入步驟中，以頻譜之峰值強度成為特定值之方式將頻譜標準化，且使該標準化後之頻譜之資料輸入至遞歸型神經網路。In the analysis method configured above, it is also possible to use an LSTM network as a recursive neural network. Furthermore, in the above-described analysis method, the input step may be such that the spectrum is normalized so that the peak intensity of the spectrum becomes a specific value, and the data of the normalized spectrum is input to the recursive neural network.
上述構成之分析方法亦可構成為進而具備：學習步驟，其使複數個基準物之各者所產生之光之頻譜之資料於輸入步驟中輸入至遞歸型神經網路，使遞歸型神經網路學習。該情形時，亦可構成為，於學習步驟中，使包含複數個基準物中之任意1或2個以上之基準物且混合比例已知之混合物所產生之光之頻譜之資料於輸入步驟中輸入至遞歸型神經網路，使用混合比例使遞歸型神經網路學習。The above-described analysis method may also be configured to further include: a learning step, which inputs the data of the spectrum of light generated by each of the plurality of reference objects to the recursive neural network in the input step, so that the recursive neural network learn. In this case, it may also be configured such that, in the learning step, the data of the light spectrum generated by a mixture including any one or two or more reference objects among the plurality of reference objects and with a known mixing ratio is input in the input step. To recursive neural networks, use mixing ratios to make recursive neural networks learn.
於上述構成之分析方法中，亦可構成為於分析步驟中，基於自遞歸型神經網路輸出之資料，將分析對象物分類為複數個基準物中之任一者。又，於上述構成之分析方法中，亦可構成為於分析步驟中，基於自遞歸型神經網路輸出之資料，求得分析對象物所含之基準物之混合比例。 [產業上之可利用性]The above-described analysis method may also be configured such that, in the analysis step, the analysis target object is classified into any of a plurality of reference objects based on the data output from the recursive neural network. Furthermore, in the above-described analysis method, the analysis step may be such that based on the data output from the recursive neural network, the mixing ratio of the reference substance contained in the analysis target object is obtained. [Industrial availability]
本發明可作為能夠進行高效率且高精度之頻譜分析之裝置及方法加以利用。The present invention can be utilized as a device and method capable of performing high-efficiency and high-precision spectrum analysis.
1:頻譜分析裝置 10:處理部 20:輸入部 30:學習部 40:分析部 C:胞 D:資料 δ:位移量 1: Spectrum analysis device 10:Processing Department 20:Input part 30:Learning Department 40:Analysis Department C: cell D:data δ: displacement
圖1係顯示頻譜分析裝置之構成之圖。 圖2係顯示作為基準物之各氨基酸粉末之拉曼頻譜之例之圖。 圖3係顯示作為分析對象物之丙胺酸之拉曼頻譜之例之圖。 圖4係顯示LSTM之模型之例之圖。 圖5係顯示表示第1實施例之分類結果之混同矩陣之圖。 圖6係顯示表示第1比較例之分類結果之混同矩陣之圖。 圖7係顯示穀氨醯胺與穀胺酸之混合物之拉曼頻譜之例之圖。 圖8係顯示Stacked LSTM(堆疊式長短期記憶網路)之模型之例之圖。 圖9係顯示第2實施例之定量結果之圖。 圖10係顯示第2比較例之定量結果之圖。 圖11係顯示表示第3實施例之分類結果之混同矩陣之圖。Figure 1 is a diagram showing the structure of a spectrum analysis device. FIG. 2 is a diagram showing an example of the Raman spectrum of each amino acid powder used as a reference material. FIG. 3 is a diagram showing an example of the Raman spectrum of alanine as an analysis target substance. Figure 4 is a diagram showing an example of an LSTM model. FIG. 5 is a diagram showing a confusion matrix showing the classification results of the first embodiment. FIG. 6 is a diagram showing a confusion matrix showing the classification results of the first comparative example. Figure 7 is a graph showing an example of the Raman spectrum of a mixture of glutamine and glutamic acid. Figure 8 is a diagram showing an example of a Stacked LSTM (stacked long short-term memory network) model. Fig. 9 is a graph showing the quantitative results of the second example. Fig. 10 is a graph showing the quantitative results of the second comparative example. FIG. 11 is a diagram showing a confusion matrix showing the classification results of the third embodiment.
1:頻譜分析裝置 1: Spectrum analysis device
10:處理部 10:Processing Department
20:輸入部 20:Input part
30:學習部 30:Learning Department
40:分析部 40:Analysis Department
Claims (14)
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2018206465A JP7239300B2 (en) | 2018-11-01 | 2018-11-01 | Spectrum analysis device and spectrum analysis method |
| JP2018-206465 | 2018-11-01 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW202035972A TW202035972A (en) | 2020-10-01 |
| TWI818101B true TWI818101B (en) | 2023-10-11 |
Family
ID=70463993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW108137990A TWI818101B (en) | 2018-11-01 | 2019-10-22 | Spectrum analysis device and spectrum analysis method |
Country Status (3)
| Country | Link |
|---|---|
| JP (1) | JP7239300B2 (en) |
| TW (1) | TWI818101B (en) |
| WO (1) | WO2020090336A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220283154A1 (en) | 2021-03-05 | 2022-09-08 | Sysmex Corporation | Measurement sample preparation method, analysis method, reagent, and reagent kit |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201017560A (en) * | 2008-10-28 | 2010-05-01 | Tsint | Fruit classification method using neural network |
| WO2017106638A1 (en) * | 2015-12-16 | 2017-06-22 | Gritstone Oncology, Inc. | Neoantigen identification, manufacture, and use |
| JP2018018354A (en) * | 2016-07-28 | 2018-02-01 | 高砂香料工業株式会社 | Quality prediction method for food and drink using deep learning, and food and drink |
| JP2018101317A (en) * | 2016-12-21 | 2018-06-28 | ホーチキ株式会社 | Abnormality monitoring system |
| JP2018147235A (en) * | 2017-03-06 | 2018-09-20 | ヤフー株式会社 | Information processing system, information processing method, and program |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6390054B2 (en) * | 2015-11-05 | 2018-09-19 | 株式会社Ｚｍｐ | Monitoring system |
-
2018
- 2018-11-01 JP JP2018206465A patent/JP7239300B2/en active Active
-
2019
- 2019-10-02 WO PCT/JP2019/039003 patent/WO2020090336A1/en active Application Filing
- 2019-10-22 TW TW108137990A patent/TWI818101B/en active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201017560A (en) * | 2008-10-28 | 2010-05-01 | Tsint | Fruit classification method using neural network |
| WO2017106638A1 (en) * | 2015-12-16 | 2017-06-22 | Gritstone Oncology, Inc. | Neoantigen identification, manufacture, and use |
| JP2018018354A (en) * | 2016-07-28 | 2018-02-01 | 高砂香料工業株式会社 | Quality prediction method for food and drink using deep learning, and food and drink |
| JP2018101317A (en) * | 2016-12-21 | 2018-06-28 | ホーチキ株式会社 | Abnormality monitoring system |
| JP2018147235A (en) * | 2017-03-06 | 2018-09-20 | ヤフー株式会社 | Information processing system, information processing method, and program |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2020071166A (en) | 2020-05-07 |
| WO2020090336A1 (en) | 2020-05-07 |
| TW202035972A (en) | 2020-10-01 |
| JP7239300B2 (en) | 2023-03-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| ElMasry et al. | Principles of hyperspectral imaging technology | |
| Vandenabeele | Practical Raman spectroscopy: an introduction | |
| Rosi et al. | Recent trends in the application of Fourier Transform Infrared (FT-IR) spectroscopy in Heritage Science: From micro-to non-invasive FT-IR | |
| Agelet et al. | A tutorial on near infrared spectroscopy and its calibration | |
| US20140267684A1 (en) | System and method for detecting contamination in food using hyperspectral imaging | |
| TWI790296B (en) | Spectrum analysis device and spectrum analysis method | |
| Shao et al. | Identification of pesticide varieties by detecting characteristics of Chlorella pyrenoidosa using Visible/Near infrared hyperspectral imaging and Raman microspectroscopy technology | |
| Sheehy et al. | Open-sourced Raman spectroscopy data processing package implementing a baseline removal algorithm validated from multiple datasets acquired in human tissue and biofluids | |
| US20240344985A1 (en) | Dye image acquisition method, dye image acquisition device, and dye image acquisition program | |
| TWI818101B (en) | Spectrum analysis device and spectrum analysis method | |
| Catelli et al. | A new miniaturised short-wave infrared (SWIR) spectrometer for on-site cultural heritage investigations | |
| CN106841167A (en) | The lossless detection method of garden stuff pesticide residue | |
| Wang et al. | Quantitative analysis of BPO additive in flour via Raman hyperspectral imaging technology | |
| Vignati et al. | Hyperspectral Imaging for Fresh-Cut Fruit and Vegetable Quality Assessment: Basic Concepts and Applications | |
| Jha | Near infrared spectroscopy | |
| Quintero et al. | A novel data processing technique for expert resonant nano-pillars transducers: A case study measuring ethanol in water and wine liquid matrices | |
| US12066433B2 (en) | Method for analyzing test substance, analyzer, training method, analyzer system, and analysis program | |
| Udayanga et al. | Dual mode multispectral imaging system for food and agricultural product quality estimation | |
| CN114441474B (en) | Near-infrared spectrometer and control method and system thereof | |
| Shiddiq et al. | Wavelength selection of multispectral imaging for oil palm fresh fruit ripeness classification | |
| Russeil et al. | Multi-View Symbolic Regression | |
| Checa et al. | Preliminary study of the relation between the content of cadmium and the hyperspectral signature of organic cocoa beans | |
| Burger | Hyperspectral NIR image analysis | |
| Ergin | Enhanced Data Reduction, Segmentation, and Spatial Multiplexing Methods for Hyperspectral Imaging | |
| US20240210426A1 (en) | Portable multimodal optical sensing system |