CN106960091A - A kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values - Google Patents
A kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values Download PDFInfo
- Publication number
- CN106960091A CN106960091A CN201710167209.0A CN201710167209A CN106960091A CN 106960091 A CN106960091 A CN 106960091A CN 201710167209 A CN201710167209 A CN 201710167209A CN 106960091 A CN106960091 A CN 106960091A
- Authority
- CN
- China
- Prior art keywords
- sample
- values
- thz
- bpann
- forecast
- 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
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 92
- 235000013372 meat Nutrition 0.000 title claims abstract description 84
- 238000001514 detection method Methods 0.000 title claims abstract description 38
- 230000003287 optical effect Effects 0.000 title claims abstract description 6
- 238000000034 method Methods 0.000 claims abstract description 75
- 238000004611 spectroscopical analysis Methods 0.000 claims abstract description 42
- 210000005036 nerve Anatomy 0.000 claims abstract description 8
- 238000012628 principal component regression Methods 0.000 claims abstract description 6
- 238000002834 transmittance Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims abstract description 4
- 238000005520 cutting process Methods 0.000 claims abstract description 3
- 239000011159 matrix material Substances 0.000 claims description 46
- 238000012549 training Methods 0.000 claims description 25
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 230000003595 spectral effect Effects 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000005259 measurement Methods 0.000 claims description 13
- 238000002790 cross-validation Methods 0.000 claims description 10
- 238000004128 high performance liquid chromatography Methods 0.000 claims description 10
- 238000002310 reflectometry Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000014509 gene expression Effects 0.000 claims description 7
- 230000001419 dependent effect Effects 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 4
- 238000005102 attenuated total reflection Methods 0.000 claims description 3
- 230000005284 excitation Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 238000011835 investigation Methods 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 210000002808 connective tissue Anatomy 0.000 abstract description 3
- 238000002203 pretreatment Methods 0.000 abstract description 2
- 235000015277 pork Nutrition 0.000 description 21
- 239000000126 substance Substances 0.000 description 16
- ZKHQWZAMYRWXGA-KQYNXXCUSA-N Adenosine triphosphate Chemical compound C1=NC=2C(N)=NC=NC=2N1[C@@H]1O[C@H](COP(O)(=O)OP(O)(=O)OP(O)(O)=O)[C@@H](O)[C@H]1O ZKHQWZAMYRWXGA-KQYNXXCUSA-N 0.000 description 15
- 238000012360 testing method Methods 0.000 description 12
- 238000010521 absorption reaction Methods 0.000 description 10
- HEMHJVSKTPXQMS-UHFFFAOYSA-M Sodium hydroxide Chemical compound [OH-].[Na+] HEMHJVSKTPXQMS-UHFFFAOYSA-M 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 9
- XTWYTFMLZFPYCI-KQYNXXCUSA-N 5'-adenylphosphoric acid Chemical compound C1=NC=2C(N)=NC=NC=2N1[C@@H]1O[C@H](COP(O)(=O)OP(O)(O)=O)[C@@H](O)[C@H]1O XTWYTFMLZFPYCI-KQYNXXCUSA-N 0.000 description 8
- XTWYTFMLZFPYCI-UHFFFAOYSA-N Adenosine diphosphate Natural products C1=NC=2C(N)=NC=NC=2N1C1OC(COP(O)(=O)OP(O)(O)=O)C(O)C1O XTWYTFMLZFPYCI-UHFFFAOYSA-N 0.000 description 8
- GRSZFWQUAKGDAV-UHFFFAOYSA-N Inosinic acid Natural products OC1C(O)C(COP(O)(O)=O)OC1N1C(NC=NC2=O)=C2N=C1 GRSZFWQUAKGDAV-UHFFFAOYSA-N 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 235000013902 inosinic acid Nutrition 0.000 description 7
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 239000000463 material Substances 0.000 description 6
- VLTRZXGMWDSKGL-UHFFFAOYSA-N perchloric acid Chemical compound OCl(=O)(=O)=O VLTRZXGMWDSKGL-UHFFFAOYSA-N 0.000 description 6
- 239000000047 product Substances 0.000 description 6
- 239000000243 solution Substances 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 5
- 238000003860 storage Methods 0.000 description 5
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 4
- 235000020997 lean meat Nutrition 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 230000005855 radiation Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 3
- UDMBCSSLTHHNCD-KQYNXXCUSA-N adenosine 5'-monophosphate Chemical compound C1=NC=2C(N)=NC=NC=2N1[C@@H]1O[C@H](COP(O)(O)=O)[C@@H](O)[C@H]1O UDMBCSSLTHHNCD-KQYNXXCUSA-N 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 150000001875 compounds Chemical class 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000009795 derivation Methods 0.000 description 3
- 230000001066 destructive effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- FDGQSTZJBFJUBT-UHFFFAOYSA-N hypoxanthine Chemical group O=C1NC=NC2=C1NC=N2 FDGQSTZJBFJUBT-UHFFFAOYSA-N 0.000 description 3
- 230000010354 integration Effects 0.000 description 3
- 239000007788 liquid Substances 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 235000013622 meat product Nutrition 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 238000009659 non-destructive testing Methods 0.000 description 3
- 239000013307 optical fiber Substances 0.000 description 3
- 239000012071 phase Substances 0.000 description 3
- 238000005064 physico chemical analysis method Methods 0.000 description 3
- 108090000623 proteins and genes Proteins 0.000 description 3
- 102000004169 proteins and genes Human genes 0.000 description 3
- 229910021642 ultra pure water Inorganic materials 0.000 description 3
- 239000012498 ultrapure water Substances 0.000 description 3
- KDCGOANMDULRCW-UHFFFAOYSA-N 7H-purine Chemical compound N1=CNC2=NC=NC2=C1 KDCGOANMDULRCW-UHFFFAOYSA-N 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 2
- 150000001413 amino acids Chemical class 0.000 description 2
- 239000007853 buffer solution Substances 0.000 description 2
- 238000005119 centrifugation Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 239000000470 constituent Substances 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 238000013210 evaluation model Methods 0.000 description 2
- 239000000796 flavoring agent Substances 0.000 description 2
- 235000019634 flavors Nutrition 0.000 description 2
- 235000021393 food security Nutrition 0.000 description 2
- 238000011068 loading method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 210000003205 muscle Anatomy 0.000 description 2
- 229910052757 nitrogen Inorganic materials 0.000 description 2
- 229920001184 polypeptide Polymers 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 102000004196 processed proteins & peptides Human genes 0.000 description 2
- 108090000765 processed proteins & peptides Proteins 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241000894007 species Species 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 239000006228 supernatant Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- LWIHDJKSTIGBAC-UHFFFAOYSA-K tripotassium phosphate Chemical compound [K+].[K+].[K+].[O-]P([O-])([O-])=O LWIHDJKSTIGBAC-UHFFFAOYSA-K 0.000 description 2
- AUHDWARTFSKSAC-HEIFUQTGSA-N (2S,3R,4S,5R)-3,4-dihydroxy-5-(hydroxymethyl)-2-(6-oxo-1H-purin-9-yl)oxolane-2-carboxylic acid Chemical compound [C@]1([C@H](O)[C@H](O)[C@@H](CO)O1)(N1C=NC=2C(O)=NC=NC12)C(=O)O AUHDWARTFSKSAC-HEIFUQTGSA-N 0.000 description 1
- 101150029019 ATP6 gene Proteins 0.000 description 1
- 241000894006 Bacteria Species 0.000 description 1
- 230000005526 G1 to G0 transition Effects 0.000 description 1
- 241000287828 Gallus gallus Species 0.000 description 1
- UGQMRVRMYYASKQ-UHFFFAOYSA-N Hypoxanthine nucleoside Natural products OC1C(O)C(CO)OC1N1C(NC=NC2=O)=C2N=C1 UGQMRVRMYYASKQ-UHFFFAOYSA-N 0.000 description 1
- 241001494479 Pecora Species 0.000 description 1
- 241000218636 Thuja Species 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000000149 argon plasma sintering Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 238000001311 chemical methods and process Methods 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 229940126678 chinese medicines Drugs 0.000 description 1
- 238000004587 chromatography analysis Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 210000002615 epidermis Anatomy 0.000 description 1
- 238000010812 external standard method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000000706 filtrate Substances 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003455 independent Effects 0.000 description 1
- 238000002329 infrared spectrum Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 229940028843 inosinic acid Drugs 0.000 description 1
- 239000004245 inosinic acid Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 239000007791 liquid phase Substances 0.000 description 1
- 238000005374 membrane filtration Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 235000016709 nutrition Nutrition 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 230000037368 penetrate the skin Effects 0.000 description 1
- 229920000768 polyamine Polymers 0.000 description 1
- 229910000160 potassium phosphate Inorganic materials 0.000 description 1
- 235000011009 potassium phosphates Nutrition 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000012372 quality testing Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000009738 saturating Methods 0.000 description 1
- 239000013049 sediment Substances 0.000 description 1
- 210000000697 sensory organ Anatomy 0.000 description 1
- 210000003491 skin Anatomy 0.000 description 1
- 150000003384 small molecules Chemical class 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 235000012976 tarts Nutrition 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 229910000404 tripotassium phosphate Inorganic materials 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
-
- 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
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
- G01N21/3586—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation by Terahertz time domain spectroscopy [THz-TDS]
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
- G01N2021/3572—Preparation of samples, e.g. salt matrices
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biochemistry (AREA)
- Molecular Biology (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Toxicology (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
A kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values, including step:One) meat collection and processing:Fresh meat is uniformly cut into after sliced meat and refrigerated;During sampling cutting, the fat and connective tissue of meat are avoided;Two) the THz spectroscopic datas of meat are determined:Using THz decay total reflection ATR detection pattern quick detection samples;During spectra collection, keep the epidemic disaster of environment stable;Three) using the freshness K values of the Relationship Prediction model quick detection fresh meat between THz spectroscopic datas and fresh meat K values:The relational model is principal component regression PCR forecast models or is that reverse transmittance nerve network returns BPANN forecast models.By step 2) obtained THz spectroscopic datas substitute into Relationship Prediction model, then quickly calculate the K values of unknown sample.The method of the sample pre-treatments of the present invention, it is only necessary to cut sample and be put into instrument with to form sliced meat and measure, simpler than existing method and quick many.
Description
Technical field
The technical program belongs to fresh meat detection technique field, specifically a kind of quick nothing of THz spectrum of fresh meat freshness K values
Damage detection method.
Background technology
Meat is the intake source of the main protein of people, fat etc..The freshness of fresh meat is especially weighed in food security
Will.By taking pork as an example, it is the main species of China's meat products consumption, and China's pork consumption figure accounts for consumption of meat total amount
77%, how to control the meat quality in production and sales link, ensure food security, receive manufacturing enterprise, consumer and state
The highest attention of family's quality testing department.Include nutritional ingredient, local flavor, tenderness, guarantor with the quality including fresh meats such as pig, ox, sheep, the flesh of fish
The key elements such as aqueous and freshness, wherein, freshness is an important aspect.
Freshness, which is determined, generally has two methods, hedonic scoring system and physico-chemical analysis method.Hedonic scoring system's subjective factor is big,
Poor repeatability;Physico-chemical analysis method is accurately credible, and repeatability is high.The freshness index that physico-chemical analysis is determined includes yellowish pink, micro- life
Thing index, total volatile basic nitrogen index, K values, pH value and polyamine compound etc., wherein, the K values based on ATP decomposable processes
It is proved to be feasible and is more and more paid attention to.
Some materials in fresh meat go bad with the increase of holding time, in the meat after government official, atriphos (ATP)
Understand automatic classifying, process is:Atriphos (ATP) → adenosine diphosphate (ADP) (ADP) → AMP (AMP) → hypoxanthine core
Thuja acid (IMP) → inosinicacid (HxR) → hypoxanthine (Hx).
According to ATP decomposable processes, there is the following indicial equation for determining K values:
ATP, ADP, AMP, IMP, HxR, Hx in formula --- the content of each material in the sample
Fresh K values are lower for meat, otherwise K values are higher.K values can generally use high efficiency liquid phase chromatographic analysis method
(High Performance Liquid Chromatography, HPLC) is obtained.Although such a physico-chemical analysis method is accurate and can
Letter, but process is time-consuming, have to sample it is destructive (method for damaging detection physical and chemical parameter is that meat is chopped into muddy flesh, then
It is processed further chemical examination), it is impossible to the quick and popularity measurement request in production and intermediate links is met, Fast nondestructive evaluation is used
The K values that method determines pork just turn into an important research focus of research work.
At present, the lossless detection method of fresh meat freshness index has near-infrared spectrum analysis, Visible-to-Near InfaRed high-spectrum
As analysis, Electronic Nose Technology etc..Spectra methods is based on the chemical substance related to fresh meat freshness in visible and near-infrared ripple
Section has the principle of feature representation, it is established that the Mathematical Modeling of Non-Destructive Testing pork freshness, realizes the quick of freshness and ratio
It is accurate to determine.But, it is necessary to sample bin vacuumize process near infrared spectrometer experimentation, it is impossible to realize online fast
Although speed detection, optical fiber type spectrometer does not need vacuumize process, but needs the spectral information with optical fiber collecting sample, coupling
Optical fiber is tiny, and detection zone is limited, it is impossible to realize disposable comprehensive collection to sample surface.Visible-to-Near InfaRed high-spectrum
Although as analytical technology can realize comprehensive collection, before each on-test and in long-time process of the test, being required for
Spectra collection system is calibrated with standard reflectivity scaling board, to prevent temperature drift and the light source of camera sensor devices
Brightness change influence to detecting system, be unfavorable for on-line continuous detection.Electronic Nose Technology gathers meat using gas sensor
Corrupt smell, the freshness of meat is determined based on degree of spoilage is deeper, smell releasing degree is stronger principle, but the method is needed
Sample is placed in closed detection storehouse, and needs stable placement certain time, make it that smell distribution is equal in detection storehouse
Even, odorousness reaches the Monitoring lower-cut of sensor, and the stand-by period of detection is long.
The content of the invention
THz wave (Terahertz, THz) is located between millimeter wave and infrared ray, belongs to far infrared band, frequency exists
In the range of 0.1THz~10THz (this case selects frequency in the THz wave of 0.1THz~2THz scopes).From energy, THz
Ripple is between electronics and photon.The diversification characteristic of THz ripples make it that many chemical moleculars are shown than at it under THz wave bands
Not available molecular motion characteristic under its wave band.Ucleotides (such as ADP, ATP) and its related substances (such as IMP) belong to biological
Small molecule, terahertz wave band absorption mainly due to the rotation, vibration or molecule group of its molecule itself body vibration,
There is THz wave spectrum signature structure, there are multiple absworption peaks in THz wave bands.
Therefore, the present invention, by exploring the THz spectral characteristics of fresh meat freshness, sets up THz light using K values as research parameter
Relational model between modal data and fresh meat freshness K values, realizes the Fast nondestructive evaluation to fresh meat K values, proposes a kind of fresh meat
Freshness Fast nondestructive evaluation new method, concrete technical scheme is as follows:
A kind of THz spectrum method for quick of fresh meat freshness K values, it is characterized in that including step:
One) meat collection and processing:Fresh meat is uniformly cut into after sliced meat and refrigerated;During sampling cutting, the fat of meat is avoided
And connective tissue;
Two) the THz spectroscopic datas of meat are determined:Using THz decay total reflection Attenuated Total
Reflectance, ATR detection pattern quick detection sample;During spectra collection, keep the epidemic disaster of environment stable;
Three) using the freshness K values of the Relationship Prediction model quick detection fresh meat between THz spectroscopic datas and fresh meat K values:
The relational model is principal component regression PCR forecast models or is that reverse transmittance nerve network returns BPANN forecast models;
By step 2) obtained THz spectroscopic datas substitute into Relationship Prediction model, then quickly calculate the K values of unknown sample;
The step 2) in:
The THz spectrum of fresh meat sample are sampled using THz decay total reflection ATR detection patterns;To fresh meat sample
THz spectrum, which are sampled, to be obtained sample spectrum data and is handled;
All modeling samples are randomly divided into calibration set and forecast set;(sample number in sample number and forecast set in calibration set
Ratio be about 2:1) spectroscopic data of calibration set is used to set up forecast model;The spectroscopic data of forecast set is used for examining prediction mould
The accuracy of type.Xn×mIt is to refer in particular to the spectroscopic data in calibration set.
Sample spectrum data matrix isIn formula, n is calibration set sample number, and m is light
Wave point number is composed, x is the ATR reflectivity of sample;
First spectral information is compressed:Matrix Xn×mTwo matrix Us and P are decomposed into, U is expressed asn×a=Xn×mPm×a;
In formula, matrix U represents data xij(i=1,2 ..., n;J=1,2 ..., the m) vector position in new coordinate-system;Matrix P
Column vector represent the linear transform coefficient of new coordinate system and former coordinate system;U dimension a is less than X dimension m;Take matrix U
Each row data IiThe input of the forecast model of (i=1,2 ..., a) as step 3);
The step 3) in:
A, the PCR forecast models:
Multiple linear regression is carried out to matrix U, multiple linear equation is obtained:Y=UB+E, as PCR forecast models, are used for
Predict K values;In formula, Y is the actual measurement K values of sample set, and E is the random error matrix of variables of Normal Distribution;
The preparation method of the parameter B is:In the modeling process of PCR forecast models, multiple samples are randomly divided into many
Individual calibration set and multiple forecast sets;The spectroscopic data of calibration set is used to set up forecast model;The spectroscopic data of forecast set is used for examining
Test the accuracy of forecast model;The actual measurement K values of calibration set sample are constituted into matrix Y, non trivial solution is obtained:B=(U ' U)-1U ' Y,
In formula, U ' is U transposed matrix.Obtaining just can be above-mentioned more with the THz spectroscopic datas substitution of forecast set sample after parameter matrix B
First linear equation obtains the prediction K values of sample in forecast set.
B, the modeling procedure of the BPANN forecast models include:
1) multilayer reverse transmittance nerve network is built:
Multilayer reverse transmittance nerve network BPANN is considered as nonlinear function, BPANN inputs and predicted value are respectively that this is non-
The independent variable and dependent variable of linear function, BPANN express the Function Mapping relation from independent variable to dependent variable;
BPANN topological structure is made up of three-layer network, is input layer, hidden layer and output layer respectively, every layer including more
It is the unidirectional connection of Weighted Coefficients between individual neuron, the neuron of adjacent layer;
IiIt is the input value of input layer, i=1,2 ..., a;HjIt is the output valve of hidden layer, j=1,2 ..., b;Output layer is
One nodeValue for BPANN predicted value, i.e. fresh meat prediction K values;hwijFor the neuron and hidden layer of input layer
Neuron between connection weight, owjNeuron for hidden layer and the connection weight between the neuron of output layer;
The nodes of hidden layerIn formula:A is input layer number, and c is the constant in the range of 1~10,1
For output layer nodes;
2) to step 1) model be trained, step includes:
2.1) netinit:A, b are determined according to input dimension and output dimension;Initialize hwij、owj, hidden layer threshold value
thaj, output layer threshold value thb, weights and threshold value are random number of the scope -1~1;Given learning rate η, η=0.01;
2.2) hidden layer output is calculated:The output of hidden layerIn formula, function f (x) is hidden
Excitation function containing layer, the expression formula of the function is:
2.3) output layer output is calculated:BPANN prediction output
2.4) error calculation:BPANN predicated errorIn formula, y is the K values surveyed using HPLC methods;
2.5) right value update:hwij=hwij+ηHj(1-Hj)IiowjE, i=1,2 ..., a;J=1,2 ..., b;owj=owj
+ηHjE, j=1,2 ..., b;
2.6) threshold value updates:thaj=thaj+ηHj(1-Hj)owjE, j=1,2 ..., b;Thb=thb+e;
2.7) whether terminated according to iterations and final iteration error evaluation algorithm iteration, if being not over, return to step
It is rapid 2.2).
The modeling procedure of the BPANN models also includes step 3) BPANN models are optimized, obtain BPANN improvement
Model, method is that handle is optimized to BPANN forecast models as weak fallout predictor, obtains strong fallout predictor, step includes:
3.1) initialize:N is that the distribution of weights of training data in the sample number of calibration set, initialization calibration set is Dt(j)
(j=1,2 ..., n), determine the threshold value Phi of weak fallout predictor predicated error to=1/n, are determined according to the input and output dimension of sample
The weak fallout predictor structures of BPANN, initialization network parameter;
3.2) weak fallout predictor is trained:When training t-th of weak fallout predictor, weak fallout predictor is trained with training data, obtains weak pre-
Survey the prediction K values of device output
Then predicated error is calculatedT is the number of weak fallout predictor;
3.3) weight of weak fallout predictor is calculated
3.4) adjusting training data distribution weight and normalize:In formula,
3.5) by 3.2~3.4) after step cycle T times, obtain T weak fallout predictors, forecast model is ft.By the power of renewal
It is overlapping to add, obtain the forecast model of strong fallout predictor
The Φ values are using Φ values when staying the performance of a cross-validation method Selection Model preferably, i.e., in certain limit
Interior, with fixed step-length, when investigation Φ takes different values, RMSECV of the modeling procedure in the case where staying cross-validation method value takes
Φ when RMSECV value is minimum is used as optimal threshold;
Staying the calculating process of a cross-validation method is:In n sample of calibration set, each sample is separately as prediction
Collection, remaining n-1 sample is used as calibration set;Using calibration set as foundation, the pre- of this forecast set sample is obtained by modeling procedure
Measured value;By n circulation, the predicted value of all samples is obtained;
The performance indications RMSECV for staying a cross-validation method is obtained,
The step 2) in, spectrum frequency range is 0.1THz~2THz.
The step 2) in, first sample spectrum data are pre-processed, obtained data are pre-processed as matrix X's
Data;The method that sample spectrum data are pre-processed is:
First use first derivative FD pre-processed spectrum data:In formula, x is that sample is i in wave point number
The ATR reflectivity at place;As i=1, its single order differential value is……;As i=m, its single order differential value is
SG moving-polynomial smoother spectrum are used again:Data point adjacent in odd number spectroscopic data sequence is selected, by these numbers
Strong point constitutes a window, and the central point of window is carried out smoothly;In smoothing process, the data point in window is fitted more than one
Item formula, and smooth point is calculated according to the gained multinomial;Obtain it is smooth after data point expression formula be:In formula, the width of window is 2d+1, and s is constant, λjFor SG smoothing factor sequences;
The fitting of a polynomial uses least square method.
The method of the sample pre-treatments of the present invention, it is only necessary to cut sample to form sliced meat and be put into measurement in instrument, than existing
The simple and quick many of some methods.
Brief description of the drawings
Fig. 1 is ATR annex light path schematic diagrams;
Fig. 2-1 is the K value changes tendency charts that HPLC is measured;
Fig. 2-2 is 6 THz primary light spectrograms that a meat sample continuously measures 6 gained;
Fig. 2-3 is the spectrogram after being handled through SVSRS;
Fig. 3-1 is the original spectrum of THz spectrum of 80 samples;
Fig. 3-2 is the THz optic spectrum line figures after 15 first derivations and SG are smooth;
Fig. 4-1 is the estimated performance variation diagram of PCR models under different differential widths;
Fig. 4-2 is the K values prediction scatter diagram of PCR models;
Fig. 5-1 is the structure chart of BPANN models;
Fig. 5-2 is the algorithm flow chart of BPANN models;
Fig. 5-3 is the RMSECV and differential width and node in hidden layer of BPANN models graph of a relation (RMSECV minimum
Value is marked with red circle);
Fig. 5-4 is the K values prediction scatter diagram of BPANN models;
Fig. 6-1 is the algorithm flow chart of BPANN improved model models;
Fig. 6-2 is RMSECV and threshold value Phi graph of a relation (threshold value Phi=0.05~0.25) in BPANN improved model models;
Fig. 6-3 is RMSECV and threshold value Phi graph of a relation (threshold value Phi=0.13~0.14) in BPANN improved model algorithms;
Fig. 6-4 is the K values prediction scatter diagram of BPANN improved model models.
Embodiment
Research and development below to the technical program, experimentation are illustrated:
1st, THz spectrum self character:
One distinguishing feature of Terahertz Technology is security.Compared to X-ray have kiloelectron-volt photon energy, too
The photon energy of Hertzion radiation is less than the bond energy of various chemical bonds, therefore it will not cause harmful ionization reaction.Further, since
Water has strong absorption to THz wave, and terahertz emission can not penetrate the skin of human body.Even if therefore strong Terahertz spoke
Penetrate, the influence to human body also only rests on skin surface, the inside of human body can be penetrated into rather than microwave.THz securities have
Beneficial to designing the non-destructive detecting device of safe and convenient, and tester is protected from radiation hazradial bundle.
Another distinguishing feature of Terahertz Technology is spectrum resolvability.Compared with near-infrared and intermediate infrared radiation, place
It is relatively low in the photon energy of the THz radiation of far infrared band, but this wave band still contains abundant spectral information.Have
Machine molecule, due to its rotation and the transition of vibration (including collective vibration), strong absorption and dispersion is shown in this frequency range
Characteristic.
2nd, embodiment of the related molecular substance of fresh meat freshness K values in THz wave bands:
What fresh meat freshness --- K values were measured is the content of ucleotides (such as ADP, ATP) and its related substances (such as IMP)
Than, they belong to biological micromolecule, terahertz wave band absorption mainly due to its molecule itself rotation, vibration or molecule
The body vibration of group, its THz wave spectrum signature structure is more apparent.Purine polycrystal exists many in 0.2~2.5THz wave bands
Individual absworption peak.
3rd, the selection of THz detection modes:
Due to including more atom in biomolecule, and with highdensity set vibration mode;It is each in molecule
Effect between individual atom causes biomolecule vibration to have larger anharmonicity, therefore the tera-hertz spectra of biomolecule is past
It is past to be overlapped by inhomogeneous broadening.
This problem can be solved using two ways.One kind, suction is narrowed using sub-cooled mode (such as liquid nitrogen cooling)
The broadening at peak is received, each absworption peak is isolated, observes its Absorption Characteristics;It is another, THz spectral lines are obtained under normal temperature metering system,
A variety of pretreatment modes, such as standard normal conversion (Standard Normal Variation, SNV), polynary are used spectral line
Scatter correction (Multi-scatter Calibration, MSC), derivative (including first derivative and second dervative),
Savitzky-Golay (SG) moving-polynomial smoother, wavelet transformation etc..The high frequency reduced as far as possible in the original spectrum collected is random
The noise informations such as noise, baseline drift, light scattering, use answering for linearly or nonlinearly mathematics model analysis spectral line and measured matter
Miscellaneous relation, sets up reliable and stable THz spectrum analysis models.
Purpose in view of this programme is to realize lossless quick detection to tested fresh meat sample, should be surveyed under normal temperature mode
Amount, so selection latter approach.
4th, THz detects the selection of sample
Epidermis in fresh meat (such as pork) sample, muscle and fat have different refractive indexes and saturating under THz wave bands
Penetrate rate.By structural constituent different in the section of piece of meat --- fat meat and lean meat, it is positioned in THz spectroscopic analysis systems,
In the detection range of ATR reflective-modes (0.1 to 4THz), absorption coefficient of the absorption coefficient more than fat meat of lean meat, and with
The frequency increase of THz ripples, difference between the two is increasing.It can be seen that, the different tissues composition of meat, the SPECTRAL DIVERSITY of THz ripples
It is larger.
In general, consumer prefers to buy thinner meat when buying meat products, so the quality of lean meat can be determined
The quality of whole meat.The technical program selects the lean meat part in sample as the detection object of THz wave spectrums.
This method is further described with reference to the embodiment of fresh pork:
1st, material and method
1.1 test materials
Test material is the tenderloin of cold fresh pork, and every morning is bought from local supermarket, and laboratory is transported back with refrigerating box
Afterwards, pork is uniformly cut into 2.5cm × 2.5cm × 0.5cm sliced meat.The fat and connective tissue of pork are avoided during sampling,
To prevent interference of these compositions to THz testing results.Pork sample is wrapped with freshness protection package and is placed in 4 DEG C of refrigerator-freezers after numbering
Storage is to be measured.The continuous acquisition meat sample of 8 days, with the date successively for numbering, is followed successively by 7d~0d.In 0d afternoon, 8 meat are completed
The THz spectra collections and K values of sample are determined.Experiment 10 times is repeated, the spectroscopic data and physics and chemistry value of 80 meat samples is obtained.
1.2 THz spectra collections
The model of THz detection devices is TAS7500SP (ADVANTEST companies, Japan), is worked under room temperature environment, frequency
Rate resolution ratio is 7.6GHz, and detection frequency range is 0.1THz~4THz, totally 498 sample frequency points (wave point), the spectrum of sample
Line is obtained by 2048 automatically scannings and after being averaged.
THz has strong interaction with water, there was only hundreds of microns for the meat sample transmission depth rich in water, so transmission
Pattern is not suitable for meat products freshness Non-Destructive Testing, and reflective-mode is as back wave is only absorbed by the water and can not detected.Use THz
Decay total reflection (Attenuated Total Reflectance, ATR) detection pattern, can overcome what is be rich in sample to dissociate
The strong absorption of water and combination water to THz ripples so that the THz characteristics of sample surfaces micron order thickness chemical substance can reflect
In the spectrum that THz is all-trans ejected wave.The light path of ATR detection accessories is as shown in Figure 1.
Each sample removes packaging after refrigerator taking-up, the smooth ATR that is put into of sample is detected into window surface, as shown in Figure 1.
Two surfaces up and down of sample gather 3 THz spectroscopic datas respectively, and each sample results in 6 parts of THz spectroscopic datas, by 6 parts
Spectroscopic data arithmetic average, is used as the final THz spectroscopic datas of the sample.Because THz spectrometers are quicker to temperature and humidity
Sense, so during spectra collection, keeping experiment indoor temperature, humidity basically identical.
1.3 K values are determined
Gather and carried out K values measure after the THz spectroscopic datas of sample to same sample immediately.Meat is detected using HPLC methods
K values in product.The association product (ATP, ADP, AMP, IMP, HxR and Hx) decomposed using ATP in mobile phase is in stationary phase
Flow velocity it is different, this six kinds of chemical constituents are separated, the content of these components is measured respectively, detected sample is calculated according to formula 1
The K values of product.
The sample pretreatment process detected for K values is as follows:Sample is minced into meat mud, therefrom take (2.00 ±
0.05) g is put into 50mL centrifuge tubes, the 10% perchloric acid solution 20mL, vortex oscillation 1min, with 8000r/ added after cooling
Min centrifugation 10min, take out supernatant.The determinand brought up again again with 5% perchloric acid solution 20mL in sediment, with
8000r/min centrifugation 10min, merge supernatant.Extracting liquid pH value nearly 6.0, Ran Houzai is adjusted with 10mol/L NaOH solution
Continue regulation pH with 1.0mol/L NaOH solution and be settled to 50mL to 6.0~6.4, then with ultra-pure water.With 0.45 μm of micropore
Membrane filtration, filtrate is to be measured in being preserved at 4 DEG C.
HPLC conditions are as follows:Finnigan Surveyor liquid chromatographs (Sai Mofei companies, the U.S.), AQ-C18 chromatograms
Post (Sai Mofei companies, the U.S.), mobile phase is 0.05mol/L K3PO4Buffer solution (pH=6.5), buffer solution is matched somebody with somebody with ultra-pure water
System, the μ L of sample size 1 of sample, flow velocity 200 μ L/min, Detection wavelength 254nm.ATP decompose association product by quantified by external standard method,
Measurement range is in 0~0.5mmol/L.
Details as Follows for the reagent used in experiment:ATP related compounds (ATP, ADP, AMP, IMP, HxR and Hx) totally 6 kinds of standards
Product (purity >=99%, Sigma-Aldrich company, the U.S.);Potassium phosphate, NaOH, perchloric acid (analyze pure, Chinese medicines group
Chemical reagent Co., Ltd, China);Test water is ultra-pure water prepared by Millipore Academic.
2nd, result and analysis
2.1 K values are analyzed
The K values of 80 samples are obtained using HPLC methods, and have done statistical analysis, as a result as shown in Fig. 2-1.Different tests
The meat sample of batch, storage duration it is the same, K values as individual difference and it is different.By the average value of sample can be seen that with
The increase of storage number of days, freshness is constantly reduced, and the K values of sample are being stepped up, but incrementss are not fixed.Wherein,
5d~6d, 6d~7d incrementss it is relatively large, illustrate within the above-mentioned storage cycle, the quality comparison of meat sample is larger;Sense organ is commented
Valency also confirms that meat sample within 6d~7d storage cycles, touches and relatively sticks, has obvious tart flavour, therefore can speculate meat sample within this cycle
Freshness is remarkably decreased.The K values of all samples of this experiment cover the different freshness of pork, wide coverage so that THz light
The Mathematical Modeling of spectrum prediction K values has preferable robustness.
80 samples are randomly divided into 54 calibration sets and 26 forecast sets, number ratio substantially 2:1.Wherein calibration set is used
In the Mathematical Modeling for setting up THz Forecast of Spectra K values;Forecast set is used for the standard for the model prediction unknown sample K values that inspection institute sets up
True property.It can be seen from Table 1 that, the K value scopes of calibration set, forecast set and sample total collection are essentially identical, average value and standard
Difference is also without significant difference, therefore the sample decomposition of calibration set and forecast set is suitable.
The K Data-Statistics information of the calibration set of table 1 and forecast set
2.2 Pretreated spectras
Spectral quality evaluation method:The same surface of meat sample is continuously measured 6 times under identical testing conditions, obtains 6
THz spectrum, this 6 spectrum should be completely superposed in theory, but the influence of the noise and test error due to instrument, be continuously repeated
6 spectrum measured can not possibly be completely superposed, in order to evaluate the quality of spectrum, can calculate same sample surface continuous several times weight
The n of repetition measurement examinationsBar spectrum standard variance spectrum (Standard Variance Spectrum of Repeat Spectral,
SVSRS), SVSRS is smaller, illustrates that spectral quality is better.
X in formulaij--- sample ith measures the ATR reflectivity at wave point j,--- n at wave point jsSecondary measurement ATR
The average of reflectivity.
Fig. 2-2 is 6 THz spectrum of some meat sample surface in 0.1THz~4THz, and Fig. 2-3 is that the spectrum is corresponding
SVSRS, as can be seen from the figure substantially increases, poor repeatability in 2THz~4THz SVSRS values.Therefore selection 0.1THz~2THz
It is used as the spectrum frequency range of modeling.
Fig. 3-1 is THz optic spectrum line of 80 samples in 0.1THz~2THz regions, it can be seen that different pork samples
Original spectrum intensity have very big difference.The difference of spectrum not only contains the difference of sample component, also including measurement error,
Baseline drift and ambient noise., must also be to light in addition to keeping experimental enviroment factor as far as possible unanimously in order to eliminate interference information
Modal data is pre-processed, to weaken or remove various disturbing factors.
The present invention uses first derivative (First Order Derivative, FD) pre-processed spectrum data, first derivative
The data deviation that needle position misalignment, drift and ambient interferences are caused can be reduced so that the closely related spectral characteristic with freshness
Become more notable.
The first derivative FD pre-processed spectrum data that the present invention is used:In formula, x is sample in ripple
The ATR reflectivity counted at for i;As i=1, its single order differential value is……;As i=m, its first differential
It is worth and is
Because derivative calculations can increase noise, therefore put down after derivative pretreatment using Savitzky-Golay (SG) multinomial
Sliding spectrum.During derivation, differential width selection is particularly significant:If differential width is too small, noise can be very big, and influence is built
The quality of model;If differential width is too big, smooth excessiveness can lose substantial amounts of detailed information.The selection of SG moving-polynomial smoothers is strange
Adjacent data point in several data sequences, a window is made up of these data points, and the central point of window is carried out smoothly.It is flat
In cunning, a multinomial is fitted to the data point in window, and smooth point is calculated according to gained multinomial.It is polynomial
Least square method is used in fitting.According to calculate can obtain it is smooth after data point expression formula be:
In formula, the width of window is 2d+1;S is constant, and its numerical value is relevant with window width;λjFor SG smoothing factor sequences
Row.
Fig. 3-2 is the first derivative spectrum figure obtained after 15 first derivations and SG are smooth.
2.3 pork K value prediction models
Changed by the content that can make the related ATP association products of K values in pork decay process, and these biomolecule
There is sensitive spectral response to THz ripples, THz spectrum can reflect the change of these biomolecule contents.Therefore THz spectrum
There is a kind of indirect relevance between data and the freshness of pork.This method uses principal component regression (Principal
Components Regression, PCR), nonlinear algorithm-error backward propagation method return (Back
Propagation Artificial Neural Network, BPANN) and BPANN improved models separately verify this association
Property.
Detection model evaluation method:
The index of evaluation model quality is to use calibration set coefficient correlation (RC), calibration set root-mean-square error (Root Mean
Squared Error of Calibration Set, RMSEC), forecast set coefficient correlation (RP) and forecast set root-mean-square error
(Root Mean Squared Error of Prediction Set, RMSEP).These parameters are defined as follows:
In formula--- the predicted value of i-th of sample in sample set (including calibration set and forecast set);
yi--- the measured value of i-th of sample in sample set (including calibration set and forecast set);
nC--- the sample number of calibration set;
nP--- the sample number of forecast set;
ym--- the average value of calibration set or forecast set sample.
Coefficient RCAnd RPIt is bigger, and/or RMSEC and RMSEP is smaller, then the predictive ability of model is better.
This example is calculated and built to spectroscopic data using Matlab R2009b (Mathworks companies, the U.S.) software
Mould.
2.3.1 principal component regression PCR forecast models
The THz spectroscopic datas of 80 samples can be represented with matrix form:
In formula, n is sample number, and m is spectrum wave point number, and x is the ATR reflectivity of sample;Xn×mSpectroscopic data refers to calibration set
In data.
Each spectrum in this experiment is all containing 250 wave points, and the wave point number of spectrum is much larger than sample number, if directly
For regression analysis, it may appear that over-fitting, the precision of prediction and stability of model are reduced.Exist simultaneously between the data of spectrum wave point
Higher synteny and correlation, can also make it that the result of regression model produces distortion.PCA can be passed through
(Principal Component Analysis, PCA) is compressed to spectral information, passes through the score of a few principal component
Approximately to reflect former spectroscopic data, the information redundancy and correlation of spectroscopic data point are eliminated.
PCA algorithms are a kind of conventional and effective information compressing method, the variable covariances square that it is constituted from multiple samples
Battle array is set out, and obtains the maximum principal component of variance instead of original to become using the method (i.e. the linear transformation of coordinate system) of feature decomposition
Amount, has obtained new variables --- principal component scores.It is separate between principal component, orthogonal, it can eliminate in initial data and deposit
Correlation and information redundancy.Therefore, PCA is obtained in the field such as multivariate statistical analysis and spectral information compression, extraction
It is widely applied.
With principal component analysis, former matrix of variables X can be decomposed into two matrix Us and P, be expressed as Un×a=Xn×mPm×a;
In formula, in formula, matrix U is score matrix (score matrix), represents data xij(i=1,2 ..., n;J=1,
2 ..., m) in new coordinate-system (principal component coordinate-system) vector position (i.e. coordinate value);Matrix P is loading matrix
(loading matrix), its column vector represents the linear transform coefficient of new coordinate system and former coordinate system;U dimension a is less than X
Dimension m;Therefore, the spectroscopic data space of higher-dimension is reduced the principal component feature space of the linear independence of low-dimensional.Take matrix U
Each row data Ii(i=1,2 ..., a) as the input of forecast model;
PCR forecast models:Multiple linear regression is carried out to matrix U, multiple linear equation is obtained:Y=UB+E, as PCR
Forecast model, for predicting K values;In formula, Y is the actual measurement K values of sample set, and E is the random error variable of Normal Distribution
Matrix;
The preparation method of the parameter B is:In the modeling process of PCR forecast models, multiple samples are randomly divided into many
Individual calibration set and multiple forecast sets;The spectroscopic data of calibration set is used to set up forecast model;The spectroscopic data of forecast set is used for examining
Test the accuracy of forecast model;The actual measurement K values of calibration set sample are constituted into matrix Y, non trivial solution is obtained:B=(U ' U)-1U ' Y,
In formula, U ' is U transposed matrix.Obtaining just can be above-mentioned more with the THz spectroscopic datas substitution of forecast set sample after parameter matrix B
First linear equation obtains the prediction K values of sample in forecast set.
In summary, PCR can be divided into two steps:1 determines number of principal components, i.e., dropped spectrum matrix X with PCA
Dimension;2 carry out linear regression analysis again for the X matrix of dimensionality reduction.
Pretreated spectrum matrix is carried out principal component decomposition by the present invention with PCA methods, takes accumulation contribution rate to arrive
95% principal set cooperation is the input of forecast model.The differential width of Pretreated spectra reaches highest phase according to forecast model
Close coefficients RPSelected during with minimum RMSEP.Model, which is calculated, to be shown when first-order difference width is 15, the master of accumulation contribution rate 95%
Optimal (the R of estimated performance that component collections number is 26, PCRp=0.63, RMSEP=16.78%), as shown in Fig. 4-1.Fig. 4-2 is
K value prediction scatter diagram of the PCR models under this Parameter Conditions.
2.3.2 error backward propagation method returns BPANN forecast models
BPANN, which is one, can disclose the complicated Power analysis instrument contacted between input and output signal.BPANN is that one kind is more
Layer feedforward neural network, this network is construed as a nonlinear function, and network inputs and predicted value are respectively the letter
Several independents variable and dependent variable.BPANN expresses the Function Mapping relation from independent variable to dependent variable.Used by the present invention
BPANN topological structure is as shown in fig. 5-1.
BPANN topological structure is made up of three-layer network, is input layer, hidden layer and output layer respectively, every layer including more
It is the unidirectional connection of Weighted Coefficients between individual neuron, the neuron of adjacent layer;
IiIt is the input value of input layer, i=1,2 ..., a;HjIt is the output valve of hidden layer, j=1,2 ..., b;Output layer is
One nodeValue for BPANN predicted value, i.e. fresh meat prediction K values;hwijFor the neuron and hidden layer of input layer
Neuron between connection weight, owjNeuron for hidden layer and the connection weight between the neuron of output layer;
Output layer is a node, Yl(l=1) it is that BPANN predicted value predicts K values;wijAnd wjkFor BPANN nerve
First connection weight.
In order to reduce the complexity of neutral net, using above-mentioned PCA methods by pretreated spectrum matrix carry out it is main into
Point decompose, take the input that principal set cooperation of the accumulation contribution rate to 95% is neutral net, the nodes of single hidden layer by
Empirical equation is drawn:
In formula:The nodes of hidden layerIn formula:A is input layer number, and c is in the range of 1~10
Constant, 1 is output layer nodes.The nodes scope selection of hidden layer is 4~17 accordingly.
Training network is first had to before BP neural network prediction, makes network that there is associative memory and predictive ability by training.
BPANN training process includes following steps:
1) netinit:A, b are determined according to input dimension and output dimension;Initialize hwij、owj, hidden layer threshold value
thaj, output layer threshold value thb, weights and threshold value are random number of the scope -1~1;Given learning rate η, η=0.01;
2) hidden layer output is calculated:The output of hidden layerIn formula, function f (x) is implicit
Layer excitation function, the expression formula of the function is:
3) output layer output is calculated:BPANN prediction output
4) error calculation:BPANN predicated errorIn formula, y is the K values surveyed using HPLC methods;
5) right value update:hwij=hwij+ηHj(1-Hj)IiowjE, i=1,2 ..., a;J=1,2 ..., b;
owj=owj+ηHjE, j=1,2 ..., b;
6) threshold value updates:thaj=thaj+ηHj(1-Hj)owjE, j=1,2 ..., b;Thb=thb+e;
7) whether terminated according to iterations and final iteration error evaluation algorithm iteration, if being not over, return to step
2)。
BPANN's is mainly characterized by signal propagated forward, error back propagation.In forward direction transmission, input signal is from defeated
Enter layer successively to handle through hidden layer, until output layer.One layer of neuron state under the influence of each layer of neuron state.If
Output layer cannot get desired output, then be transferred to backpropagation, network weight and threshold value be adjusted according to predicated error, so that BP is refreshing
Constantly desired output is approached through neural network forecast output.After training terminates, it is possible to use the neural network forecast K values after training.
It can be seen that, the K values prediction modeling based on BPANN includes BPANN structure, three steps of training and prediction.The algorithm flow
Represented with Fig. 5-2.
Train before neutral net, the spectrum matrix of input is normalized in the range of -1~1, actual measurement K values are normalized
To in the range of 0~1.Because the nodes of hidden layer can influence the performance of neural network prediction, the nodes of hidden layer
Using staying a cross-validation method to choose, that is, choose nodes during RMSECV values minimum.According to this principle, forecast model is investigated
In the range of pretreatment differential width 3~47, the nodes of hidden layer are in the range of 4~17, RMSECV numerical value change feelings
Condition, as shown in Fig. 5-3.As can be seen that pretreatment differential width is 13 and node in hidden layer is when being 9, RMSECV numerical value is minimum
Estimated performance for 18.10%, BPANN models is optimal.Fig. 5-4 is K value prediction scatterplot of the BPANN models under this Parameter Conditions
Figure.2.3.3BPANN improved model
The core concept of this innovatory algorithm is that different fallout predictors (weak fallout predictor) are trained for same calibration set, then
These weak fallout predictors are gathered, a stronger final fallout predictor (strong fallout predictor) is constituted.The characteristics of algorithm is weak pre-
The poor performance of device is surveyed, but the performance of strong fallout predictor is very excellent.Its reason is that it can reasonably divide training set, Yi Jihe
The weak fallout predictor of merging of reason is to form strong fallout predictor.It is embodied in following two aspect in place of rationally:Chosen using after weighting
Training data rather than the training sample that randomly selects, the focus of training is so concentrated on to the training data sample of the difficult prediction of comparison
In sheet;Weak fallout predictor is joined together using the voting mechanism of weighting, the weak fallout predictor for making prediction effect good has larger power
Weight, and the fallout predictor of prediction effect difference has less weight.This improved model assign BPANN as weak fallout predictor, repetition training
BPANN forecast samples are exported, and strong fallout predictor is constituted as the weak fallout predictors of multiple BPANNs of the innovatory algorithm by obtained by.
Specific innovatory algorithm is as follows:
1. initialization:N is that the distribution of weights of training data in the sample number of calibration set, initialization calibration set is Dt(j)=
(j=1,2 ..., n), determine the threshold value Phi of weak fallout predictor predicated error, BPANN are determined according to the input and output dimension of sample 1/n
Weak fallout predictor structure, initialization network parameter;
2. the weak fallout predictor of training:Train weak fallout predictor:When training t-th of weak fallout predictor, weak prediction is trained with training data
Device, obtains the prediction K values of weak fallout predictor output
Then predicated error is calculatedT is the number of weak fallout predictor;
3. calculate the weight w of weak fallout predictort:
4. test data weight is adjusted,
In formula, BtIt is normalization factor
5. after 2,3,4 step cycle T times, T weak fallout predictors are obtained, forecast model is ft.By the weighted superposition of renewal,
Obtain the forecast model of strong fallout predictor
Innovatory algorithm flow is as in Figure 6-1, it is seen that it is a kind of integration algorithm, is integrated and gathered by above-mentioned weighted strategy
The predictive ability of interior each weak fallout predictor and the raising for obtaining macro-forecast performance.The present invention uses BPANN innovatory algorithms, with above
Based on the BPANN model parameters of foundation, its estimated performance is lifted using innovatory algorithm, the BPANN for building K values improves prediction mould
Type.
The threshold value Phi of predicated error has significant impact to the performance of BPANN improved models, using staying a cross-validation method
Choose Φ during RMSECV values minimum:First, Φ is selected as 0.05~0.25 in larger scope using wider step-length 0.01,
The situation of RMSECV values is as in fig. 6-2.As can be seen that in the range of 0.13~0.14, the Performance comparision of model is good.Connect
, within this range, investigate 0.13~0.14 with less 0.001 step-length, it can be seen that in Fig. 6-3, when Φ takes 0.136,
RMSECV values are minimum, and BPANN improved models result in optimal estimated performance.This parameter is substituted into BPANN innovatory algorithms
Iteration, obtains 10 weak BPANN fallout predictors, and the performance of these fallout predictors is as shown in table 2.It can be seen that:The property of weak fallout predictor
Can be variant, wherein the RMSEC of weak fallout predictor 2 and 8 is smaller, estimated performance is preferable, its predicated error and εtSmaller, iteration is given birth to
Into weight wtLarger, represent the two weak fallout predictors in the strong fallout predictor after integration has larger specific gravity;On the contrary, weak prediction
The RMSEC of device 6 is larger, and estimated performance is worst, its predicated error and εtMaximum, the weight w that iteration is generatedtMinimum, represents
Integrate this weak fallout predictor contribution very little during strong fallout predictor;The estimated performance of above-mentioned 10 weak fallout predictors is all undesirable,
But work as and use weight wtWeighting is integrated after weak fallout predictor, and the strong fallout predictor obtained but has preferable performance, and its estimated performance is excellent
In each weak fallout predictor.The K values for obtaining BPANN improved models by this strong fallout predictor predict scatter diagram, as shown in Fig. 6-4.
The fallout predictor performance of the BPANN improved models of table 2
The comparison of 2.4THz spectral prediction models
Above-mentioned three kinds of forecast models use the K value prediction accuracies of forecast set data verification model, and method is:For prediction
Collect data, the pretreated forecast set sample spectrum data of first differential of learning from else's experience:Formula
In, n ' is forecast set sample number, and m is spectrum wave point number, and x is the ATR reflectivity of forecast set sample;
Forecast set spectral information is compressed using the linear transform coefficient matrix P in calibration set PCA algorithms:It is expressed as
Un′×a=Xn′×mPm×a, take each row data I of matrix Ui(i=1,2 ..., a) as the input of well-established forecast model, warp
The calculating for crossing forecast model has obtained model output, that is, predicts K values.By the calculating of formula (5) and formula (6), it can obtain
The predictive ability of model, is used as the index of evaluation model performance.
The system of table 3 by BPANN improved models and traditional BPANN and principal component regression (Principal
Components Regression, PCR) model is compared in the aspect of performance of THz Forecast of Spectra K values., it can be seen that light
It is suitable, difference that the differential width of modal data pretreatment, which selects 13 (BPANN, BPANN improved models) or 15 (PCR models),
Width is too small or crosses the estimated performance of big city's reduction model.It can also be seen that (BPANN, BPANN improve mould to nonlinear model
Type) estimated performance be substantially better than linear model (PCR models).Moreover, K value estimated performance of the BPANN improved models than BPANN
Have been improved.The above results can be explained by the following aspects:
The regression result of 3 three kinds of K value prediction models of table compares
1st, from the nonlinear trend of THz spectrum and pork K values, K values are determined by the ratio of ATP 6 kinds of related compounds contents
Fixed, content and the spectral line data in THz spectrum of these molecular substances are non-linear relations.Therefore the fitting of nonlinear model is imitated
Fruit can be better than linear model.
2nd, from the chemical mechanism that pork goes bad, the rotten of pork is a complicated chemical process, is caused several
In the presence of rotten bacterium, the protein in muscle is first hydrolyzed to polypeptide, then is hydrolyzed into amino acid, and is further broken into various having
Machine material.The research of prior art shows that the biomolecule such as protein, polypeptide and amino acid all has respective in THz wave bands
Characteristic absorption, its characteristic absorption mostlys come from the collective vibration mould of molecule, so THz spectrum can reflect that these change.But
It is that biomolecule specy in pork is various, each characteristic spectrum has plyability at normal temperatures, it is not only table to cause THz spectrum
Reach the change information of K values, but the integrated information of the complex chemical composition of pork change.Therefore, the pass of K values and THz spectrum
System should be complex nonlinear relation, and linear model can not explain this complicated regression relation.
3rd, from the theory structure of modeling algorithm, nonlinear model is more good at self study and self-adjusting than linear model,
BPANN network topology structure is more suitable for the complicated chemical composition of analysis.When running into the regression relation of complexity, prediction mould is built
Type can be more excellent.BPANN innovatory algorithms by the weak fallout predictor of reasonable integration in an iterative process, make the weak forecast models of BPANN by
Gradually it is promoted to strong forecast model.Therefore, BPANN improved models have been improved than the estimated performance of BPANN model.
The practical significance of 2.5 models
R in table 3PThe accuracy predicted with RMSEP index expressions model unknown sample K values.Unknown sample is that is, unknown
The sample of its K value, different from calibration set and the sample of forecast set, its K value is obtained by physico-chemical analysis HPLC methods.Pass through
First differential is pre-processed and PCA spectral information compression methods, and the THz spectroscopic datas of the unknown sample after processing are substituted into mould
Type, can quickly calculate the K values of unknown sample.This programme is used for the sample of calibration set to model, and the sample of forecast set is used for
Judgment models are good and bad, and the K values prediction of unknown sample is exactly the practical of model.The practical significance of model is exactly to pass through THz spectrum
The K values of quick detection unknown sample, rather than only with known sample (including calibration set and forecast set) modeling with training.
3rd, conclusion
This paper results of study show, be all-trans emission mode using ATR, cold fresh lean is obtained in the range of 0.1THz~2THz
The THz spectroscopic datas on surface, the Mathematical Modeling built after first differential and SG the disposal of gentle filter can Fast nondestructive evaluation pig
Meat K values, to evaluate pork freshness.Pass through the comparative studies table to three kinds of forecast model PCR, BPANN, BPANN improved models
Bright, non-linear BPANN and BPANN improved models can preferably be fitted THz spectroscopic datas and freshness K than linear PCR model
Complex nonlinear relation between value., will and BPANN improved models have certain advantage on processing complex relationship model
R has been arrived in BPANN estimated performance liftingPIt is 13.45% for 0.78, RMSEP.
Based on identical THz Cleaning Principles, THz spectra methods can also the other meats of Non-Destructive Testing (such as chicken, ox
Meat, flesh of fish etc.) K values.The present invention provides to be based further on the portable rapid non-destructive detecting device of the method exploitation design
Theoretical foundation.
Claims (4)
1. a kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values, it is characterized in that including step:
One) meat collection and processing:Fresh meat is uniformly cut into after sliced meat and refrigerated;During sampling cutting, the fat and knot of meat are avoided
Form tissue;
Two) the THz spectroscopic datas of meat are determined:Using THz decay total reflection Attenuated Total Reflectance,
ATR detection pattern quick detection samples;During spectra collection, keep the epidemic disaster of environment stable;
Three) using the freshness K values of the Relationship Prediction model quick detection fresh meat between THz spectroscopic datas and fresh meat K values:It is described
Relational model is principal component regression Principal Components Regression, PCR forecast models or is reverse pass
Broadcast neural net regression Back Propagation Artificial Neural Network, BPANN forecast models;
By step 2) obtained THz spectroscopic datas substitute into Relationship Prediction model, then quickly calculate the K values of unknown sample;
The step 2) in:
The THz spectrum of fresh meat sample are sampled using THz decay total reflection ATR detection patterns;To the THz light of fresh meat sample
Spectrum, which is sampled, to be obtained sample spectrum data and is handled;
All samples for modeling are randomly divided into calibration set and forecast set;The spectroscopic data of calibration set is used to set up prediction mould
Type;The spectroscopic data of forecast set is used for examining the accuracy of forecast model;
Sample spectrum data matrix isIn formula, n is calibration set sample number, and m is spectrum wave point
Number, x is the ATR reflectivity of sample;Xn×mRefer to the spectroscopic data in calibration set;
First spectral information is compressed:Matrix Xn×mTwo matrix Us and P are decomposed into, U is expressed asn×a=Xn×mPm×a;In formula,
Matrix U represents data xij(i=1,2 ..., n;J=1,2 ..., the m) vector position in new coordinate-system;Matrix P row
The linear transform coefficient of the new coordinate system of vector representation and former coordinate system;U dimension a is less than X dimension m;Take each row of matrix U
Data IiThe input of the forecast model of (i=1,2 ..., a) as step 3);
The step 3) in:
A, the PCR forecast models:
Multiple linear regression is carried out to matrix U, multiple linear equation is obtained:Y=UB+E, as PCR forecast models, for predicting
K values;In formula, Y is the actual measurement K values of sample set, and E is the random error matrix of variables of Normal Distribution;
The preparation method of the parameter B is:In the modeling process of PCR forecast models, multiple samples are randomly divided into multiple schools
Positive collection and multiple forecast sets;The spectroscopic data of calibration set is used to set up forecast model;The spectroscopic data of forecast set is used for examining pre-
Survey the accuracy of model;The actual measurement K values of calibration set sample are constituted into matrix Y, non trivial solution is obtained:B=(U ' U)-1In U ' Y, formula,
U ' is U transposed matrix;Obtain after parameter matrix B, then substitute into above-mentioned multiple linear side with the THz spectroscopic datas of forecast set sample
Journey obtains the prediction K values of sample in forecast set;
B, the modeling procedure of the BPANN forecast models include:
1) multilayer reverse transmittance nerve network is built:
Multilayer reverse transmittance nerve network BPANN is considered as nonlinear function, BPANN inputs and predicted value are respectively that this is non-linear
Argument of function and dependent variable, BPANN express the Function Mapping relation from independent variable to dependent variable;
BPANN topological structure is made up of three-layer network, is input layer, hidden layer and output layer respectively, and every layer includes multiple god
It is the unidirectional connection of Weighted Coefficients through member, between the neuron of adjacent layer;
IiIt is the input value of input layer, i=1,2 ..., a;HjIt is the output valve of hidden layer, j=1,2 ..., b;Output layer is one
Node Value for BPANN predicted value, i.e. fresh meat prediction K values;hwijThe nerve of neuron and hidden layer for input layer
Connection weight between member, owjNeuron for hidden layer and the connection weight between the neuron of output layer;
The nodes of hidden layerIn formula:A is input layer number, and c is the constant in the range of 1~10, and 1 is defeated
Go out node layer number;
2) to step 1) model be trained, step includes:
2.1) netinit:A, b are determined according to input dimension and output dimension;Initialize hwij、owj, hidden layer threshold value thaj、
Output layer threshold value thb, weights and threshold value are random number of the scope -1~1;Given learning rate η, η=0.01;
2.2) hidden layer output is calculated:The output of hidden layerIn formula, function f (x) is hidden layer
Excitation function, the expression formula of the function is:
2.3) output layer output is calculated:BPANN prediction output
2.4) error calculation:BPANN predicated errorIn formula, y is the K values surveyed using HPLC methods;
2.5) right value update:hwij=hwij+ηHj(1-Hj)IiowjE, i=1,2 ..., a;J=1,2 ..., b;owj=owj+η
HjE, j=1,2 ..., b;
2.6) threshold value updates:thaj=thaj+ηHj(1-Hj)owjE, j=1,2 ..., b;Thb=thb+e;
2.7) whether terminated according to iterations and final iteration error evaluation algorithm iteration, if being not over, return to step
2.2)。
2. the THz spectrum method for quick of fresh meat freshness K values according to claim 1, it is characterized in that the BPANN
The modeling procedure of model also includes step 3) BPANN models are optimized, BPANN improved models are obtained, method is right
BPANN forecast models are optimized as weak fallout predictor, obtain strong fallout predictor, and step includes:
3.1) initialize:N is that the distribution of weights of training data in the sample number of calibration set, initialization calibration set is Dt(j)=1/n
(j=1,2 ..., n), determine the threshold value Phi of weak fallout predictor predicated error, determine that BPANN is weak according to the input and output dimension of sample
Fallout predictor structure, initialization network parameter;
3.2) weak fallout predictor is trained:When training t-th of weak fallout predictor, weak fallout predictor is trained with training data, weak fallout predictor is obtained
The prediction K values of output
Then predicated error is calculatedT is the number of weak fallout predictor;
3.3) weight of weak fallout predictor is calculated
3.4) adjusting training data distribution weight and normalize:In formula,
3.5) by 3.2~3.4) after step cycle T times, obtain T weak fallout predictors, forecast model is ft.Folded by the weight of renewal
Plus, obtain the forecast model of strong fallout predictor
The Φ values be using Φ values when staying the performance of a cross-validation method Selection Model preferably, i.e., within the specific limits, with
Fixed step-length, when investigation Φ takes different values, RMSECV of the modeling procedure in the case where staying cross-validation method value takes RMSECV
Value for it is minimum when Φ be used as optimal threshold;
Staying the calculating process of a cross-validation method is:In n sample of calibration set, each sample separately as forecast set,
Remaining n-1 sample is used as calibration set;Using calibration set as foundation, the prediction of this forecast set sample is obtained by modeling procedure
Value;By n circulation, the predicted value of all samples is obtained;
The performance indications RMSECV for staying a cross-validation method is obtained,
3. the THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values according to claim 1, it is characterized in that described
Step 2) in, spectrum frequency range is 0.1THz~2THz.
4. the THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values according to claim 1, it is characterized in that step
Two) in, first sample spectrum data are pre-processed, the data of obtained data as matrix X are pre-processed;Sample spectrum data
The method pre-processed is:
First use first derivative FD pre-processed spectrum data:In formula, x is that sample is at i in wave point number
ATR reflectivity;As i=1, its single order differential value is……;As i=m, its single order differential value is
SG moving-polynomial smoother spectrum are used again:Data point adjacent in odd number spectroscopic data sequence is selected, by these data points
A window is constituted, the central point of window is carried out smoothly;It is multinomial to the data point fitting one in window in smoothing process
Formula, and smooth point is calculated according to the gained multinomial;Obtain it is smooth after data point expression formula be:In formula, the width of window is 2d+1, and s is constant, λjFor SG smoothing factor sequences;
The fitting of a polynomial uses least square method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710167209.0A CN106960091B (en) | 2017-03-20 | 2017-03-20 | THz spectrum rapid nondestructive testing method for fresh meat freshness K value |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710167209.0A CN106960091B (en) | 2017-03-20 | 2017-03-20 | THz spectrum rapid nondestructive testing method for fresh meat freshness K value |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106960091A true CN106960091A (en) | 2017-07-18 |
CN106960091B CN106960091B (en) | 2020-05-05 |
Family
ID=59470384
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710167209.0A Expired - Fee Related CN106960091B (en) | 2017-03-20 | 2017-03-20 | THz spectrum rapid nondestructive testing method for fresh meat freshness K value |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106960091B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019223364A1 (en) * | 2018-05-25 | 2019-11-28 | 深圳市太赫兹科技创新研究院有限公司 | Method for detecting quality of cordyceps sinensis based on electromagnetic wave in terahertz band |
CN113008172A (en) * | 2021-03-03 | 2021-06-22 | 北京理工大学 | Terahertz wave-based ice and snow track inspection device and method |
CN113218878A (en) * | 2020-01-21 | 2021-08-06 | 青岛海尔电冰箱有限公司 | Method for judging freshness of food materials of refrigerator, refrigerator and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102507459A (en) * | 2011-11-23 | 2012-06-20 | 中国农业大学 | Method and system for quick lossless evaluation on freshness of fresh beef |
CN103487382A (en) * | 2013-09-27 | 2014-01-01 | 浙江工商大学 | Method for judging tuna meat freshness by redness index |
CN104655761A (en) * | 2015-02-28 | 2015-05-27 | 华南理工大学 | Method for measuring fish freshness index value K in online way based on multi-spectral imaging |
CN105548029A (en) * | 2015-12-14 | 2016-05-04 | 北京农业质量标准与检测技术研究中心 | Meat product freshness detection method based on spectral imaging technology |
-
2017
- 2017-03-20 CN CN201710167209.0A patent/CN106960091B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102507459A (en) * | 2011-11-23 | 2012-06-20 | 中国农业大学 | Method and system for quick lossless evaluation on freshness of fresh beef |
CN103487382A (en) * | 2013-09-27 | 2014-01-01 | 浙江工商大学 | Method for judging tuna meat freshness by redness index |
CN104655761A (en) * | 2015-02-28 | 2015-05-27 | 华南理工大学 | Method for measuring fish freshness index value K in online way based on multi-spectral imaging |
CN105548029A (en) * | 2015-12-14 | 2016-05-04 | 北京农业质量标准与检测技术研究中心 | Meat product freshness detection method based on spectral imaging technology |
Non-Patent Citations (4)
Title |
---|
QIPING HUANG 等: "Non-destructively sensing pork’s freshness indicator using near infrared multispectral imaging technique", 《JOURNAL OF FOOD ENGINEERING》 * |
XIN B. HUANG 等: "The meat freshness detection based on terahertz wave", 《PROCEEDINGS SPIE 9795,SELECTED PAPERS OF THE PHOTOELECTRONIC TECHNOLOGY COMMITTEE CONFERENCES HELD JUNE–JULY 2015》 * |
张海云 等: "基于光谱技术和支持向量机的生鲜猪肉水分含量快速无损检测", 《光谱学与光谱分析》 * |
齐亮 等: "冷鲜猪肉的新鲜度无损检测技术现状及THz检测技术展望", 《食品与机械》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019223364A1 (en) * | 2018-05-25 | 2019-11-28 | 深圳市太赫兹科技创新研究院有限公司 | Method for detecting quality of cordyceps sinensis based on electromagnetic wave in terahertz band |
CN113218878A (en) * | 2020-01-21 | 2021-08-06 | 青岛海尔电冰箱有限公司 | Method for judging freshness of food materials of refrigerator, refrigerator and storage medium |
CN113008172A (en) * | 2021-03-03 | 2021-06-22 | 北京理工大学 | Terahertz wave-based ice and snow track inspection device and method |
Also Published As
Publication number | Publication date |
---|---|
CN106960091B (en) | 2020-05-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107014769B (en) | A kind of fresh meat K value Fast nondestructive evaluation model based on THz spectrum analysis | |
Tian et al. | Early detection of freezing damage in oranges by online Vis/NIR transmission coupled with diameter correction method and deep 1D-CNN | |
Luo et al. | Robustness improvement of NIR-based determination of soluble solids in apple fruit by local calibration | |
CN101430276B (en) | Wavelength variable optimization method in spectrum analysis | |
Malvandi et al. | Application of NIR spectroscopy and multivariate analysis for Non-destructive evaluation of apple moisture content during ultrasonic drying | |
Jiang et al. | Quantitative analysis of wheat maltose by combined terahertz spectroscopy and imaging based on Boosting ensemble learning | |
Huang | Chemometric methods in analytical spectroscopy technology | |
CN101413885A (en) | Near-infrared spectrum method for rapidly quantifying honey quality | |
CN101520412A (en) | Near infrared spectrum analyzing method based on isolated component analysis and genetic neural network | |
Che et al. | Application of visible/near‐infrared spectroscopy in the prediction of azodicarbonamide in wheat flour | |
CN105445217A (en) | Method for rapidly detecting content of aflatoxin in brown rice based on attenuated total reflection Fourier transform infrared spectrum technique | |
Malvandi et al. | Non-destructive measurement and real-time monitoring of apple hardness during ultrasonic contact drying via portable NIR spectroscopy and machine learning | |
Akbarzadeh et al. | Microwave spectroscopy based on the waveguide technique for the nondestructive freshness evaluation of egg | |
CN106960091A (en) | A kind of THz optical spectrum rapid nondestructive detection methods of fresh meat freshness K values | |
Wang et al. | A novel CC-tSNE-SVR model for rapid determination of diesel fuel quality by near infrared spectroscopy | |
Wang et al. | NIR based wireless sensing approach for fruit monitoring | |
Mishra et al. | Improved prediction of minced pork meat chemical properties with near-infrared spectroscopy by a fusion of scatter-correction techniques | |
CN105158201A (en) | Rapid detection method for content of aflatoxin in brown rice based on FT-NIR technology | |
CN107024445A (en) | The modeling method and detection method of the quick detection of Nitrate in Vegetable | |
Martins et al. | Unscrambling spectral interference and matrix effects in Vitis vinifera Vis-NIR spectroscopy: Towards analytical grade ‘in vivo’sugars and acids quantification | |
Kim et al. | Neural network based prediction of soluble solids concentrationin oriental melon using VIS/NIR spectroscopy | |
Jiang et al. | Comparison of wavelength selected methods for improving of prediction performance of PLS model to determine aflatoxin B1 (AFB1) in wheat samples during storage | |
Kemps et al. | The importance of choosing the right validation strategy in inverse modelling | |
Liu et al. | Rapid determination of acidity index of peanut during storage by a portable near-infrared spectroscopy system | |
CN103134770A (en) | Method for eliminating influence on infrared spectrum detection of soil total nitrogen content from moisture |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200505 |
|
CF01 | Termination of patent right due to non-payment of annual fee |