CN117030628A - High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract - Google Patents
High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract Download PDFInfo
- Publication number
- CN117030628A CN117030628A CN202310768547.5A CN202310768547A CN117030628A CN 117030628 A CN117030628 A CN 117030628A CN 202310768547 A CN202310768547 A CN 202310768547A CN 117030628 A CN117030628 A CN 117030628A
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- quality
- extract
- traditional chinese
- chinese medicine
- 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.)
- Pending
Links
- 239000000284 extract Substances 0.000 title claims abstract description 164
- 238000001228 spectrum Methods 0.000 title claims abstract description 76
- 239000003814 drug Substances 0.000 title claims abstract description 68
- 238000004458 analytical method Methods 0.000 title claims abstract description 37
- 238000012937 correction Methods 0.000 claims abstract description 97
- 238000000034 method Methods 0.000 claims abstract description 59
- 238000000605 extraction Methods 0.000 claims abstract description 34
- 238000010801 machine learning Methods 0.000 claims abstract description 33
- 238000004519 manufacturing process Methods 0.000 claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 7
- 230000003595 spectral effect Effects 0.000 claims description 22
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 238000002203 pretreatment Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 13
- 238000009499 grossing Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008030 elimination Effects 0.000 claims description 4
- 238000003379 elimination reaction Methods 0.000 claims description 4
- 238000002310 reflectometry Methods 0.000 claims description 4
- 230000002068 genetic effect Effects 0.000 claims description 3
- -1 polytetrafluoroethylene Polymers 0.000 claims description 3
- 229920001343 polytetrafluoroethylene Polymers 0.000 claims description 3
- 239000004810 polytetrafluoroethylene Substances 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 2
- 230000000007 visual effect Effects 0.000 claims description 2
- 230000002087 whitening effect Effects 0.000 claims description 2
- 238000004891 communication Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 53
- DOUMFZQKYFQNTF-WUTVXBCWSA-N (R)-rosmarinic acid Chemical compound C([C@H](C(=O)O)OC(=O)\C=C\C=1C=C(O)C(O)=CC=1)C1=CC=C(O)C(O)=C1 DOUMFZQKYFQNTF-WUTVXBCWSA-N 0.000 description 28
- 230000000052 comparative effect Effects 0.000 description 25
- 238000011156 evaluation Methods 0.000 description 23
- 239000001606 7-[(2S,3R,4S,5S,6R)-4,5-dihydroxy-6-(hydroxymethyl)-3-[(2S,3R,4R,5R,6S)-3,4,5-trihydroxy-6-methyloxan-2-yl]oxyoxan-2-yl]oxy-5-hydroxy-2-(4-hydroxyphenyl)chroman-4-one Substances 0.000 description 14
- ZZAFFYPNLYCDEP-HNNXBMFYSA-N Rosmarinsaeure Natural products OC(=O)[C@H](Cc1cccc(O)c1O)OC(=O)C=Cc2ccc(O)c(O)c2 ZZAFFYPNLYCDEP-HNNXBMFYSA-N 0.000 description 14
- DFPMSGMNTNDNHN-ZPHOTFPESA-N naringin Chemical compound O[C@@H]1[C@H](O)[C@@H](O)[C@H](C)O[C@H]1O[C@H]1[C@H](OC=2C=C3O[C@@H](CC(=O)C3=C(O)C=2)C=2C=CC(O)=CC=2)O[C@H](CO)[C@@H](O)[C@@H]1O DFPMSGMNTNDNHN-ZPHOTFPESA-N 0.000 description 14
- 229940052490 naringin Drugs 0.000 description 14
- 229930019673 naringin Natural products 0.000 description 14
- ARGKVCXINMKCAZ-UZRWAPQLSA-N neohesperidin Chemical compound C1=C(O)C(OC)=CC=C1[C@H]1OC2=CC(O[C@H]3[C@@H]([C@@H](O)[C@H](O)[C@@H](CO)O3)O[C@H]3[C@@H]([C@H](O)[C@@H](O)[C@H](C)O3)O)=CC(O)=C2C(=O)C1 ARGKVCXINMKCAZ-UZRWAPQLSA-N 0.000 description 14
- ARGKVCXINMKCAZ-UHFFFAOYSA-N neohesperidine Natural products C1=C(O)C(OC)=CC=C1C1OC2=CC(OC3C(C(O)C(O)C(CO)O3)OC3C(C(O)C(O)C(C)O3)O)=CC(O)=C2C(=O)C1 ARGKVCXINMKCAZ-UHFFFAOYSA-N 0.000 description 14
- DOUMFZQKYFQNTF-MRXNPFEDSA-N rosemarinic acid Natural products C([C@H](C(=O)O)OC(=O)C=CC=1C=C(O)C(O)=CC=1)C1=CC=C(O)C(O)=C1 DOUMFZQKYFQNTF-MRXNPFEDSA-N 0.000 description 14
- TVHVQJFBWRLYOD-UHFFFAOYSA-N rosmarinic acid Natural products OC(=O)C(Cc1ccc(O)c(O)c1)OC(=Cc2ccc(O)c(O)c2)C=O TVHVQJFBWRLYOD-UHFFFAOYSA-N 0.000 description 14
- 239000000463 material Substances 0.000 description 7
- 240000004980 Rheum officinale Species 0.000 description 6
- 235000008081 Rheum officinale Nutrition 0.000 description 6
- 238000000701 chemical imaging Methods 0.000 description 6
- 238000002360 preparation method Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000004128 high performance liquid chromatography Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 4
- 239000000469 ethanolic extract Substances 0.000 description 4
- 238000002329 infrared spectrum Methods 0.000 description 4
- 239000002131 composite material Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 239000000047 product Substances 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- OCKGFTQIICXDQW-ZEQRLZLVSA-N 5-[(1r)-1-hydroxy-2-[4-[(2r)-2-hydroxy-2-(4-methyl-1-oxo-3h-2-benzofuran-5-yl)ethyl]piperazin-1-yl]ethyl]-4-methyl-3h-2-benzofuran-1-one Chemical compound C1=C2C(=O)OCC2=C(C)C([C@@H](O)CN2CCN(CC2)C[C@H](O)C2=CC=C3C(=O)OCC3=C2C)=C1 OCKGFTQIICXDQW-ZEQRLZLVSA-N 0.000 description 2
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 2
- 241000405414 Rehmannia Species 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 239000000706 filtrate Substances 0.000 description 2
- 239000008187 granular material Substances 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 239000000843 powder Substances 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 238000000611 regression analysis Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 244000307697 Agrimonia eupatoria Species 0.000 description 1
- 108010011485 Aspartame Proteins 0.000 description 1
- VWDXGKUTGQJJHJ-UHFFFAOYSA-N Catenarin Natural products C1=C(O)C=C2C(=O)C3=C(O)C(C)=CC(O)=C3C(=O)C2=C1O VWDXGKUTGQJJHJ-UHFFFAOYSA-N 0.000 description 1
- 229920000742 Cotton Polymers 0.000 description 1
- FBPFZTCFMRRESA-KVTDHHQDSA-N D-Mannitol Chemical compound OC[C@@H](O)[C@@H](O)[C@H](O)[C@H](O)CO FBPFZTCFMRRESA-KVTDHHQDSA-N 0.000 description 1
- 244000286838 Eclipta prostrata Species 0.000 description 1
- 239000010282 Emodin Substances 0.000 description 1
- RBLJKYCRSCQLRP-UHFFFAOYSA-N Emodin-dianthron Natural products O=C1C2=CC(C)=CC(O)=C2C(=O)C2=C1CC(=O)C=C2O RBLJKYCRSCQLRP-UHFFFAOYSA-N 0.000 description 1
- YOOXNSPYGCZLAX-UHFFFAOYSA-N Helminthosporin Natural products C1=CC(O)=C2C(=O)C3=CC(C)=CC(O)=C3C(=O)C2=C1O YOOXNSPYGCZLAX-UHFFFAOYSA-N 0.000 description 1
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 238000004566 IR spectroscopy Methods 0.000 description 1
- 229930195725 Mannitol Natural products 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 244000236658 Paeonia lactiflora Species 0.000 description 1
- 235000008598 Paeonia lactiflora Nutrition 0.000 description 1
- 235000002789 Panax ginseng Nutrition 0.000 description 1
- 238000001237 Raman spectrum Methods 0.000 description 1
- 244000153955 Reynoutria sachalinensis Species 0.000 description 1
- 235000003202 Reynoutria sachalinensis Nutrition 0.000 description 1
- NTGIIKCGBNGQAR-UHFFFAOYSA-N Rheoemodin Natural products C1=C(O)C=C2C(=O)C3=CC(O)=CC(O)=C3C(=O)C2=C1O NTGIIKCGBNGQAR-UHFFFAOYSA-N 0.000 description 1
- 241000913745 Spatholobus Species 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 239000000605 aspartame Substances 0.000 description 1
- 235000010357 aspartame Nutrition 0.000 description 1
- IAOZJIPTCAWIRG-QWRGUYRKSA-N aspartame Chemical compound OC(=O)C[C@H](N)C(=O)N[C@H](C(=O)OC)CC1=CC=CC=C1 IAOZJIPTCAWIRG-QWRGUYRKSA-N 0.000 description 1
- 229960003438 aspartame Drugs 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 235000000125 common agrimony Nutrition 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 206010012601 diabetes mellitus Diseases 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000000857 drug effect Effects 0.000 description 1
- RHMXXJGYXNZAPX-UHFFFAOYSA-N emodin Chemical compound C1=C(O)C=C2C(=O)C3=CC(C)=CC(O)=C3C(=O)C2=C1O RHMXXJGYXNZAPX-UHFFFAOYSA-N 0.000 description 1
- VASFLQKDXBAWEL-UHFFFAOYSA-N emodin Natural products OC1=C(OC2=C(C=CC(=C2C1=O)O)O)C1=CC=C(C=C1)O VASFLQKDXBAWEL-UHFFFAOYSA-N 0.000 description 1
- 238000002481 ethanol extraction Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000002262 irrigation Effects 0.000 description 1
- 238000003973 irrigation Methods 0.000 description 1
- 238000002386 leaching Methods 0.000 description 1
- 235000010355 mannitol Nutrition 0.000 description 1
- 239000000594 mannitol Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910052757 nitrogen Inorganic materials 0.000 description 1
- 235000016709 nutrition Nutrition 0.000 description 1
- 230000035764 nutrition Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- PKUBGLYEOAJPEG-UHFFFAOYSA-N physcion Natural products C1=C(C)C=C2C(=O)C3=CC(C)=CC(O)=C3C(=O)C2=C1O PKUBGLYEOAJPEG-UHFFFAOYSA-N 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000010298 pulverizing process Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 206010043554 thrombocytopenia Diseases 0.000 description 1
- 229940126680 traditional chinese medicines Drugs 0.000 description 1
- 238000000108 ultra-filtration Methods 0.000 description 1
- 238000001291 vacuum drying Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/58—Extraction of image or video features relating to hyperspectral data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Databases & Information Systems (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention belongs to the field of traditional Chinese medicine quality analysis, and provides a hyperspectral intelligent analysis method for quality of traditional Chinese medicine extract; the method comprises the steps of performing hyperspectral detection, correction, mask making and average spectrum calculation on an extract sample, performing pretreatment on the average spectrum or extraction of average spectrum characteristics to obtain a treated spectrum, and constructing an extract quality index quantitative correction model on the treated spectrum based on a machine learning algorithm so as to analyze the extract sample; compared with the prior art, the analysis method provided by the invention can be used for rapidly, nondestructively and non-contact analysis of the quality index of the Chinese medicinal extract, and provides basis and guidance for quality detection, analysis and control of the production and manufacturing process of the Chinese medicinal extract.
Description
Technical Field
The invention belongs to the field of traditional Chinese medicine quality analysis, and relates to a hyperspectral intelligent analysis method for quality of traditional Chinese medicine extractum.
Background
The Chinese medicine is used as the Chinese medicine component of national culture, is the pharmaceutical knowledge gradually accumulated by the laborers in the past and disease fight, and is one of the most influential subjects in the world of China. The traditional Chinese medicinal materials are often accompanied with the defects of low effective components, large dosage, slow drug effect and low absorption and utilization rate, so that in the application of the traditional Chinese medicinal materials, a considerable proportion of medicinal materials need to be prepared into a Chinese medicinal extract form for utilization.
In the field of traditional Chinese medicine extract production, concentration is a key process link for connecting extraction and preparation, and the quality of the concentrated extract directly influences the stability of the subsequent process and the quality of a finished product. The detection of the traditional Chinese medicine extract is mainly focused on the detection of the components of the traditional Chinese medicine extract and the detection of the relative density of the traditional Chinese medicine extract, and the traditional Chinese medicine extract or the leaching liquid is detected by adopting a traditional Chinese medicine extract densitometer, a traditional Chinese medicine extract relative density measuring instrument or a liquid chromatograph and other instruments and means, so that the problems of various detection equipment, complex sample treatment and long detection time exist.
Currently, a method for measuring the relative density of a traditional Chinese medicine extracting solution by adopting infrared spectrometry is available, and CN102106907B discloses a quality control method for an ethanol extraction process of traditional Chinese medicine rheum officinale, which is characterized by comprising the following steps: step 1, determining the relative density and emodin content of a qualified rheum officinale ethanol extract, and collecting a near infrared spectrum; step 2, respectively establishing a corresponding model between near infrared spectrum and relative density of the ethanol extract of rheum officinale and rheum officinale content by adopting chemometric software; step 3, measuring the near infrared spectrum of a rheum officinale ethanol extract sample to be measured; and 4, respectively obtaining the relative density and the rhein content of the rheum officinale ethanol extract sample to be detected by using a corresponding model through chemometric software.
However, the traditional Chinese medicine extract generally has the characteristics of high viscosity, high sugar content, poor fluidity, bubbles, particles and the like, and the detection of the extract by a conventional process rapid analyzer such as a near infrared spectrum or a Raman spectrum has a certain difficulty.
The hyperspectral imaging technology can obtain the image and spectral information of the sample in a non-contact mode, has unique advantages in the detection of non-uniform samples, and is widely applied to the fields of remote sensing, agriculture, food, medicine and the like in recent years. At present, the hyperspectral imaging technology is used for carrying out the identification of the origin of the traditional Chinese medicine, the nondestructive detection of the quality of the traditional Chinese medicine product and the like, and the hyperspectral imaging technology is not used for the rapid analysis of the quality of the traditional Chinese medicine extract.
CN113989525a discloses a hyperspectral traditional Chinese medicine identification method of a self-adaptive random block convolution kernel network, based on an optimal clustering framework, an optimal band subset of a hyperspectral image of traditional Chinese medicine is obtained, and then an optimal characteristic band is effectively selected from the optimal band subset by adopting a cluster sequencing method; using a random projection method to take a random block extracted from the hyperspectral image of the traditional Chinese medicine as a convolution kernel; then modifying the convolution kernel by using a pixel self-adaption method, and extracting features based on the Chinese medicinal material feature wave band image; thirdly, extracting the characteristics of the traditional Chinese medicinal materials by using a layered network, and constructing a traditional Chinese medicinal material hyperspectral training set and a testing set by combining the hyperspectral optimal wave band image data of the traditional Chinese medicinal materials; finally, training the training set by using the SVM to obtain a classification detection model, detecting the traditional Chinese medicine test set based on the model, greatly improving the identification and classification precision of the traditional Chinese medicine, solving the identification difficult problems of variety and complex components of the traditional Chinese medicine, and being applicable to rapid nondestructive identification of various traditional Chinese medicines.
At present, hyperspectrum is mainly used for quantitative and qualitative detection of chemical components of various substances, however, the hyperspectral has limited detection application in the aspect of physical indexes such as relative density and the like, and is especially used in the field of traditional Chinese medicine extractum; for the Chinese medicinal extract, the relative density is closely related to the preparation of the subsequent preparation.
For example, the granule is of a type requiring high relative density of Chinese medicinal extract, and CN115192653a discloses a preparation method of instant type fuxuening sugar-free granule, which comprises the following steps: the preparation method comprises the steps of putting 513.3g of giant knotweed, 320.8g of stir-fried white peony root, 641.8g of hairyvein agrimony, 513.3g of rehmannia root, 513.3g of prepared rehmannia root, 641.8g of suberect spatholobus stem, 192.5g of eclipta and 256.8g of radix pseudostellariae into a multifunctional extraction tank, and adding water for extraction to obtain an extraction solution; carrying out ultrafiltration on the extract by adopting a ceramic composite membrane to obtain filtrate; concentrating the filtrate under reduced pressure to obtain soft extract with relative density of 1.35, and vacuum drying at 85deg.C to obtain dry extract; pulverizing the dry extract to 80 mesh to obtain extract powder; fifthly, adding mannitol accounting for 14 percent and aspartame accounting for 1.0 percent of the total prescription mass into the extract powder, and granulating by adopting a dry method to prepare particles with the granularity of 20 meshes; sixthly, subpackaging the prepared particles into aluminized composite film bags according to the specification of 8g of each bag, and sealing to obtain the aluminum-plated composite film bag. The invention has low cost and high quality, can be used for patients suffering from diabetes and thrombocytopenia at the same time, and further expands the clinical application range; the application provides higher requirements for the range of the relative density of the traditional Chinese medicine extract, and shows that the rapid nondestructive detection of the relative density of the traditional Chinese medicine extract has wide application prospect. It is known that accurate control of the relative density of the Chinese medicinal extract is one of the key points for ensuring the quality of the prepared preparation.
Therefore, how to provide an effective, rapid, nondestructive method for detecting hyperspectral traditional Chinese medicine extract containing physical and chemical indexes of traditional Chinese medicine extract is one of the key problems of research by those skilled in the art.
Disclosure of Invention
Noun interpretation
ROI (region of interest) is the region of interest described in the present invention.
Aiming at the problems existing in the prior art, the invention provides a hyperspectral intelligent analysis method for the quality of traditional Chinese medicine extract.
According to the invention, the quality standard data of the extract sample is obtained by detecting the traditional Chinese medicine extract sample by utilizing a high performance liquid chromatography and a specific gravity bottle method, then the extract sample is subjected to hyperspectral detection, an interested spectrum region is obtained by information acquisition, black and white correction and sample mask manufacturing, the steps of calculating an average spectrum value, carrying out average spectrum pretreatment and extracting spectral characteristics on the interested spectrum region are further carried out to obtain a processed spectrum, the detection parameters are modeled according to the processed spectrum based on a machine learning algorithm, and modeling is applied to hyperspectral detection, so that the quality of the traditional Chinese medicine extract is rapidly analyzed.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a hyperspectral intelligent analysis method for the quality of Chinese medicinal extract comprises the following steps:
(1) Collecting an extract sample and measuring quality standard data of the extract sample;
(2) Collecting hyperspectral data of the extract sample in the step (1) to obtain a hyperspectral image of the extract sample;
(3) Performing black-and-white correction on the hyperspectral image obtained in the step (2) to obtain a corrected image;
(4) Manufacturing a sample mask by using the correction image obtained in the step (3), wherein the area covered by the sample mask is the region of interest;
(5) Calculating the average spectrum of the region of interest obtained in the step (4);
(6) Preprocessing the average spectrum in the step (5) or extracting spectral characteristics to obtain a processed spectrum;
(7) Constructing a quality index quantitative correction model based on a machine learning algorithm according to the spectrum processed in the step (6);
(8) And (3) detecting the traditional Chinese medicine extract samples according to the quality index quantitative correction model in the step (7) to obtain the visual results of the samples.
Preferably, the method for determining accurate data of the extract sample quality in the step (1) is as follows: measuring the component content of each extract sample by using high performance liquid chromatography; the relative density of each extract sample was determined using the relative density flask method.
Preferably, the wavelength range of the hyperspectral image in step (2) is 900-1800nm in the near infrared band and 400-1000nm in the ultraviolet/visible band.
Preferably, the method of black-and-white correction in step (3) is as follows: and calculating the hyperspectral image by using a correction formula.
Further preferably, the correction formula used for the black-and-white correction is:
I c to correct the image, I is the original hyperspectral image of the collected extract sample, I w Is a polytetrafluoroethylene plate total reflection whiteboard image, I d The full black image collected by the mirror cover is covered.
Preferably, the method for manufacturing the sample mask in the step (4) includes a step of whitening the sample mask and a step of removing noise.
Further preferably, the black-and-white processing is to select a band with the largest difference between the hyperspectral spectrum of the extract sample and the spectrum of the sample plate, extract an image of the band in the hyperspectral data, and convert the image into a black-and-white image after the binarization processing.
Still more preferably, the noise removing step is to perform an open operation and a close operation on the black-white image to find a connected region in the image, set a region threshold to remove the noise in the image, and the reserved pixel point coordinates are the sample mask.
Preferably, the calculation formula of the average spectrum in the step (5) is:
wherein,and m is the number of wave bands contained in the region of interest, which is the average spectrum of the current sample.
Further preferably, theThe calculation formula of (2) is as follows:
wherein,for the current sample flat in band wAverage reflectivity; n is the number of pixel points in the interested area of the current sample, and is determined by the mask coverage area of each extract sample.
Preferably, the method of preprocessing in step (6) comprises one or a combination of several processing methods of Multiple Scatter Correction (MSC), orthonormal transformation (SNV), first derivative (D1).
Further preferably, the method of pre-treatment further comprises a combination of Savitsky-Golay smoothing and multivariate scatter correction, i.e. a multivariate scatter correction treatment based on SG smoothing (sg+msc); or a combination of Savitsky-Golay smoothing and orthonormal transform, i.e., SG smoothed orthonormal transform processing (sg+snv).
Preferably, the method of spectral feature extraction in step (6) includes minimum angle regression (LARS), no information variable elimination (UVE), genetic Algorithm (GA), principal Component Analysis (PCA) and self-encoder (AE).
Preferably, the machine learning algorithm described in step (7) includes Partial Least Squares (PLS), support Vector Machines (SVMs), convolutional Neural Networks (CNNs), and LightGBM.
Further preferably, the machine learning algorithm is Convolutional Neural Network (CNN) and LightGBM.
Most preferably, the convolutional neural network algorithm model consists of 3 convolutional layers, 1 expansion layer and 2 full-connection layers, wherein the convolutional layer activation function adopts LeakyReLu, the full-connection layer activation function adopts Linear, the number of nodes of each convolutional layer is optimized at [16,32,64] by adopting a grid search method, and the convolutional kernel size of each convolutional layer is optimized between 1 and 5; the LightGBM optimizes the number of blades (leave) between [15,31,63], optimizes the learning rate (learning rate) between [0.01,0.1,0.2], optimizes the maximum depth (max_depth) between [3,4,5], and optimizes the minimum data amount (min_data_per_leaf) of each blade between [10,20,30] to construct a model.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, the hyperspectral image of the traditional Chinese medicine extract sample is subjected to correction, masking, calculation and extraction treatment, and a quantitative correction model of the quality index of the traditional Chinese medicine extract is constructed based on a machine learning method, so that the quality index of the traditional Chinese medicine extract can be rapidly, nondestructively and non-contactlessly analyzed according to the model by rapidly combining the hyperspectral image of the traditional Chinese medicine extract sample, and basis and guidance are provided for the application in the quality detection, analysis and control fields of the production and manufacturing process of the traditional Chinese medicine extract.
Drawings
FIG. 1 is a schematic diagram of hyperspectral imaging and pretreatment of a Chinese medicinal extract sample, wherein the schematic diagram comprises (a) a hyperspectral image of the Chinese medicinal extract sample, (b) a region of interest (ROI) to be selected, (c) a mask, (d) a region of interest (ROI), and (e) the region of interest after abnormal points are removed;
FIG. 2 is an original average spectrum of the samples of the Weifu spring extract 1 in example 1 and a spectrum after pretreatment;
FIG. 3 is an extraction chart of average spectral characteristics of samples of the Weifu chun extract 1 in example 1 provided by the present invention;
fig. 4 is a graph showing the results of visualization of the quality index of the samples of the gastric-recovering spring extract 1 in example 1 provided by the present invention, (a) naringin, (b) neohesperidin, (c) rosmarinic acid, and (d) relative density.
Detailed Description
It is to be noted that the raw materials used in the present invention are all common commercial products, and the sources thereof are not particularly limited.
The following raw material sources are exemplary illustrations:
the source of the hyperspectral image in the examples described below is a pentabell optical HSI-NIR hyperspectral imaging system.
The extract 1 sample Weifu chun extract is provided by Hangzhou Hu Qingyu Tang pharmaceutical Co.Ltd; the extract 2 sample is a self-made red ginseng extract in a laboratory.
The accuracy of the analysis and detection result of the quality index quantitative correction model constructed by the invention is evaluated by utilizing the decision coefficient and the root mean square error.
Determining coefficientsThe larger the explanatory modeling independent variable is, the higher the explanation degree of the independent variable is, and the variation caused by the independent variable isThe high percentage of motion to total variation indicates that the smaller the degree of dispersion of model data relative to the true data mean, the more accurate the model data is in [0,1]Within the range of>The larger the regression model, the better the predictive performance of the regression model is explained; the root mean square error, RMSE, which is the square root of the average of the sum of squares of the differences between the predicted and actual values, the smaller the RMSE, the better the predictive performance of the regression model; but->It is difficult to measure the detection result in absolute value with RMSE, < + >>In the present invention +.>Above 0.8, rmse less than 0.02 is used as a criterion for good relative density prediction of the Chinese medicinal extract.
Because of the characteristics of large viscosity, high sugar content, poor air permeability and the like of the traditional Chinese medicine extract, the evaluation of the accuracy of the detection of the components of the traditional Chinese medicine extract is inconsistent with the relative density of the traditional Chinese medicine extract, dihide et al indicate that the determination coefficient is The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluationThe information quantity and the authenticity are more improved; the authors also mention "even though regression analysis (R 2 RMSE, etc.) have been used in a number of machine learning studies, but there is no consensus on evaluating a single, uniform standard indicator of the outcome of the regression itself. In the Chinese medicinal extract component, we consider the quality index quantitative correction model constructed by SG, D2 and WT-based spectrum pretreatment method or CARS and SPA spectrum feature extraction method as evaluation reference, and consider the quality index calibration constructed by the inventionThe volume correction model has good detection capability of the components of the Chinese medicinal extract. Indeed, as known from the research work Nitrogen nutrition diagnosis for cotton under mulched drip irrigation using unmanned aerial vehicle multispectral images of Pei et al, in some situations, the machine learning model determines the coefficient R 2 Above 0.6, good detection can be considered.
Example 1A hyperspectral Intelligent analysis method for quality of Chinese medicinal extract
The traditional Chinese medicine extract sample adopted in the embodiment is extract 1, and the invention provides a high-spectrum intelligent analysis method for quality of traditional Chinese medicine extract, which comprises the following specific steps:
(1) The extractum was collected for a total of 218 samples as a modeled sample set. Measuring naringin, rosmarinic acid and neohesperidin content of each extract sample by high performance liquid chromatography; and measuring the relative density of each extract sample by using a relative density bottle method, and taking the relative density as a reference value of the quality index of the sample set.
(2) Collecting hyperspectral images of 512 wavebands in the 898-1751nm range from the modeling sample set collected in the step (1) by using a hyperspectral instrument, wherein parameters of the hyperspectral instrument are as follows: the movement speed of the sample platform was 1.6 mm.s -1 The camera exposure time was 25ms; due to instrument noise and the like, the 421-band hyperspectral information of 900-1601nm is reserved for subsequent analysis.
(3) Performing black-and-white correction processing on the hyperspectral image acquired in the step (2), wherein the correction is according to a formulaIs carried out, wherein I c Is a hyperspectral image after black and white correction; i is the original hyperspectral image, I d To collect the full black image of the field of view after covering the lens cover, I w To acquire the image of the polytetrafluoroethylene plate total reflection whiteboard.
(4) Selecting the hyperspectral image I after black and white correction in the step (3) c The 250 th band of (a) is the band for manufacturing the mask, the reflectivity interval is set to be 0.05-0.5, and the binarization processing and the opening and closing operation are carried out on the hyperspectral image after the black-white correction, specifically, twoThe segmentation threshold value of the valued processing is set to be 12/255-127/255, the pixel point threshold value of the region is set to be 1000 through the opening and closing operation, the reserved pixel point coordinates are the sample mask, and the region covered by the mask is the region of interest.
(5) Calculating an average spectrum in a region of interest (ROI) in step (4), the calculation formula being as follows:
wherein,for the average reflectivity of the current sample in the band w (w is 421 bands reserved in step (1)), n is the number of pixels of the region of interest of the current sample (determined by the mask coverage area of each sample),/>The average spectrum of each extract sample is calculated by taking the average spectrum of the current sample (the spectrum band number m is 421).
(6) Carrying out spectrum pretreatment or spectrum feature extraction on the average spectrum obtained in each step (5), and respectively carrying out spectrum pretreatment or spectrum feature extraction by adopting one of 9 average spectrum pretreatment modes or one of 7 average spectrum feature extraction methods, wherein the spectrum pretreatment modes are respectively pretreatment-free (original average spectrum), multi-element scattering correction (MSC) treatment, standard orthogonal transformation (SNV) treatment, first derivative (D1) treatment, multi-element scattering correction (SG+MSC) treatment based on SG smoothing and standard orthogonal transformation (SG+SNV) based on SG smoothing; the spectrum characteristic extraction method is respectively minimum angle regression (LARS), non-information variable elimination (UVE) and Genetic Algorithm (GA); and obtaining an original average spectrogram, a pretreated spectrogram and a spectrogram characteristic extraction image.
(7) Dividing the modeling sample set in the step (1) into a training set and a detection set according to a ratio of 3:1, and evaluating Root Mean Square Error (RMSE) and a decision coefficient (R) of a reference detection set by model performance 2 ) Results;the training set in the modeling sample set comprises 164 samples, and the detection set comprises 54 samples; according to the reference value of the quality index of the training set in the sample set in the step (1), the pretreated spectrogram obtained in the step (6) is based on a LightGBM machine learning algorithm, and a grid search method is adopted in [15,31,63]]Optimizing the number of leaves (leave) in between, in [0.01,0.1,0.2]]Optimizing learning rate between [3,4,5]]The maximum depth (max_depth) is optimized between [10,20,30]Optimizing the minimum data quantity (min_data_per_leaf) of each blade, constructing a quality index quantitative correction model of the traditional Chinese medicine extract, and evaluating the quality index quantitative correction model by using a detection set to obtain a quality index quantitative correction model evaluation result.
(8) Detecting a traditional Chinese medicine extract sample according to the quality index quantitative correction model obtained in the step (7), substituting the spectrum of each pixel point in the sample into the model, calculating the detection value of the spectrum to obtain a gray level map, converting different gray levels in the obtained gray level map into different color components according to a JET mapping relation, and further converting the gray level map into a pseudo-color map to obtain the visualized result of each sample.
A hyperspectral imaging and mask processing schematic diagram of the Chinese medicinal extract samples of the random six samples obtained in the steps (2) - (4) is shown in figure 1, wherein (a) the hyperspectral image of the Chinese medicinal extract sample, (b) the region of the ROI to be selected, (c) the mask, (d) the region of the ROI, and (e) the region of the ROI after removing the abnormal points.
The original average spectrum obtained in step (6) and the spectrum after pretreatment are shown in fig. 2.
The spectrum characteristic extraction chart obtained in the step (6) is shown in fig. 3.
The quality index visualization result obtained in the step (8) is shown in fig. 4, wherein (a) naringin, (b) neohesperidin, (c) rosmarinic acid and (d) the relative density are shown.
The evaluation results of the quantitative correction model of the quality index obtained in the step (7) are shown in tables 1 to 4.
It should be noted that, in the embodiment, the visualized result obtained in the step (8) is actually color, the distribution and content and relative density of each component in the Chinese medicinal extract sample can be determined according to the distribution and type of the color and according to the composition or the comparison relation between the content and color on the right side of the picture, and taking fig. 4d as an example, the color and distribution of the visualized result image in fig. 4d can be compared with the color-relative density value comparison relation between 1.20-1.36 on the right side, so as to determine the relative density value and distribution condition through the visualized result.
Example 2A hyperspectral Intelligent analysis method for quality of Chinese medicinal extract
This embodiment differs from embodiment 1 in that: the types of the adopted extractum are different, and the extractum 2 is selected to construct a quality index quantitative correction model; in the step (1), the content of naringin, rosmarinic acid and neohesperidin in each extract sample is measured without using a high performance liquid chromatography; in the step (7), only the accuracy of the quantitative correction model of the quality index in the aspect of detecting the relative density of the traditional Chinese medicine extract is evaluated; and step (8) is not included.
The evaluation results of the quantitative correction model of the quality index obtained in the step (7) are shown in tables 1-2.
Example 3A method for high-spectrum intelligent analysis of Chinese medicinal extract quality
This embodiment differs from embodiment 1 in that: in the step (1), the content of naringin, rosmarinic acid and neohesperidin in each extract sample is measured without using a high performance liquid chromatography; in the step (6), spectral feature extraction is not required; in the step (7), a quality index quantitative correction model based on a LightGBM machine learning algorithm is not adopted, an algorithm used for constructing the quality index quantitative correction model is a CNN machine learning algorithm, the CNN machine learning algorithm model is composed of 3 convolution layers, 1 expansion layer and 2 full connection layers, a convolution layer activation function is selected from LeakyReLu, a full connection layer activation function is Linear, the number of nodes of each convolution layer is optimized at [16,32,64] by adopting a grid search method, and the convolution kernel size of each convolution layer is optimized between 1 and 5; in the step (7), only the accuracy of the quantitative correction model of the quality index in the aspect of detecting the relative density of the traditional Chinese medicine extract is evaluated; and step (8) is not included.
The evaluation results of the quantitative correction model for quality index obtained in step (7) in example 3 are shown in Table 5.
Comparative example 1A hyperspectral intelligent analysis method for quality of Chinese medicinal extract
The difference between this comparative example and example 1 is that: in the step (6), spectrum pretreatment is not carried out, and only a UVE algorithm is adopted for extracting spectrum characteristics; in the step (7), based on the LightGBM machine learning algorithm, the mode of constructing the model is different: the method comprises the following steps: optimizing the number of blades (leaves) between [7,15,31], optimizing the learning rate (learning rate) between [0.01,0.1], optimizing the maximum depth (max_depth) between [3,4,5], optimizing the minimum data quantity (min_data_per_leaf) of each blade between [5,10,20], and constructing to obtain an extract quality index quantitative correction model; and step (8) is not included.
The evaluation results of the quantitative correction model for quality index obtained in step (7) in comparative example 1 are shown in tables 2 and 4.
Comparative example 2A hyperspectral intelligent analysis method for quality of Chinese medicinal extract
The difference between this comparative example and example 3 is that: in the step (6), only MSC algorithm is adopted to carry out average spectrum pretreatment; in the step (7), the mode of constructing the CNN machine learning algorithm model is different: the method comprises the following steps: the CNN machine learning algorithm model consists of 4 convolution layers, 2 expansion layers and 3 full-connection layers, wherein ReLu is selected as a convolution layer activation function, linear is adopted as a full-connection layer activation function, the number of nodes of each convolution layer is optimized at [16,32,64] by adopting a grid search method, and the convolution kernel size of each convolution layer is optimized between 1 and 5.
The evaluation results of the quantitative correction model for quality index obtained in step (7) in comparative example 2 are shown in table 5.
Comparative example 3A high-quality high-spectrum intelligent analysis method for Chinese medicinal extract
The difference between this comparative example and example 1 is that: the average spectrum pretreatment method and the average spectrum characteristic extraction method adopted in the step (6) are different: the method comprises the following steps: adopting one of an average spectrum preprocessing method SG (SG) smoothing process, a second derivative (D2) process and Wavelet Transformation (WT) or adopting an average spectrum characteristic extraction method to compete for one of self-adaptive re-weighted sampling (CARS) and a continuous projection algorithm (SPA); and step (8) is not included.
The evaluation results of the quantitative correction model for quality index obtained in step (7) in comparative example 3 are shown in tables 1 to 4.
Comparative example 4A high-quality high-spectrum intelligent analysis method for Chinese medicinal extract
The difference between this comparative example and example 3 is that: the average spectrum pretreatment method adopted in the step (6) is different; the method comprises the following steps: one of the average spectrum preprocessing method SG (SG) smoothing processing, the second derivative (D2) processing and the Wavelet Transformation (WT) is adopted respectively.
The evaluation results of the quantitative correction model for quality index obtained in step (7) in comparative example 4 are shown in table 5.
Results and analysis:
the evaluation results of the quantitative correction models for the quality indexes of the above examples 1 to 3 and comparative examples 1 to 4 are as follows:
TABLE 1 comparison of evaluation results of relative Density detection accuracy of extractum based on model constructed by different spectral pretreatment methods
As can be seen from analysis of the data in Table 1, examples 1 and 2 construct a spectrum pretreatment method based on SNV, MSC, D, SG+MSC and SG+SNV and further construct 5 quality index quantitative correction models based on the LightGBM machine learning algorithm respectively, the accuracy of the relative density of the extract detected by these models is evaluated, and the evaluation result shows that the quality index quantitative correction model constructed by the traditional Chinese medicine extract quality hyperspectral intelligent analysis method of the invention has a determination coefficient for the accuracy of the relative density detection of extract 1 and extract 2 extract samplesAre all larger than 0.823, and Root Mean Square Errors (RMSE) are all smaller than 0.009; therefore, the quality index quantitative correction model constructed by the invention shows good accuracy of detecting the relative density of the extract.
And 3 quality index quantification based on the LightGBM machine learning algorithm constructed based on the spectrum pretreatment method (SG, D2 and WT) used in comparative example 3 of the inventionCorrection model for determining coefficient of accuracy of relative density detection of extract 10.623, 0.684 and 0.629, respectively, which determine the coefficient of accuracy of relative density detection of extract 2>0.679, 0.710 and 0.601 respectively; the Root Mean Square Error (RMSE) of the relative density detection accuracy of the extract 1 is as follows: 0.012, 0.016 and 0.018; the Root Mean Square Error (RMSE) of the relative density detection accuracy of the extract 2 is as follows: 0.010, and 0.013; determining coefficient->The Root Mean Square Error (RMSE) is smaller than 0.710, and the RMSE is larger than 0.010, and a quality index quantitative correction model constructed by adopting the non-invention optimized spectrum pretreatment method shows the accuracy of detecting the relative density of the extractum which is worse than that of the invention; the requirements of the invention on the quantitative correction model of the quality index are not satisfied.
TABLE 2 comparison of evaluation results of relative density detection accuracy of extractum based on model built by different spectral feature extraction methods
LightGBM' is an algorithm model constructed according to comparative example 1
From analysis of the data in Table 2, it can be seen that examples 1 and 2 construct 3 quality index quantitative correction models based on LARS, UVE and GA spectral feature extraction methods and further based on the LightGBM machine learning algorithm, respectively, and evaluate the accuracy of detecting extract relative density by these models, and the evaluation result shows that the quality index quantitative correction model constructed by the traditional Chinese medicine extract quality hyperspectral intelligent analysis method of the invention is used for determining the accuracy of detecting the relative density of extract 1 and extract 2 samplesConstant coefficientAre all greater than 0.826, and the root mean square error RMSE is all less than 0.008; therefore, the quality index quantitative correction model constructed by the invention shows good accuracy of detecting the relative density of the extract.
In comparative example 1, although the UVE was also used as the spectral feature extraction method and the quantitative correction model of the quality index was constructed based on the LightGBM machine learning method, the manner of constructing the quantitative correction model of the quality index based on the LightGBM machine method was different; the quantitative correction model of quality index obtained in comparative example 1 was evaluated, and the evaluation result showed that the quantitative correction model of quality index, lightGBM', obtained by the method for constructing quantitative correction model of quality index according to the present invention, was used for determining the coefficient of accuracy of relative density detection of extract 1 and extract 2 samplesThe root mean square error is 0.013, and compared with the quality index quantitative correction model LightGBM constructed by the same characteristic extraction method, the accuracy of the relative density detection of the extract sample is obviously reduced.
The 2 quality index quantitative correction models based on the LightGBM machine learning algorithm constructed based on the spectral feature extraction method (CARS and SPA) used in comparative example 3 of the invention determine coefficients for the relative density detection accuracy of extract 10.095 and 0.429 respectively, which determine the coefficient of relative density detection accuracy for extract 2 +.>0.045 and 0.492, respectively, in that order; the Root Mean Square Error (RMSE) of the relative density detection accuracy of the extract 1 is as follows: 0.174 and 0.043; the Root Mean Square Error (RMSE) of the relative density detection accuracy of the extract 2 is as follows: 0.379 and 0.056; block for solving the problem ofConstant (F)>The root mean square errors RMSE are all smaller than 0.492, and the RMSE is all larger than 0.043, and the quality index quantitative correction model constructed by adopting the non-invention optimized spectrum characteristic extraction method shows the accuracy of detecting the relative density of the extractum which is worse than that of the invention; the requirements of the invention on the quantitative correction model of the quality index are not satisfied.
TABLE 3 comparison of evaluation results of detection accuracy of extract components based on model constructed by different spectral pretreatment methods
From analysis of the data in Table 3, it can be seen that example 1 constructs 5 quality index quantitative correction models based on SNV, MSC, D1, SG+MSC and SG+SNV, and further based on the LightGBM machine learning algorithm, respectively, and evaluates the accuracy of these models in detecting extract component content (naringin, neohesperidin and rosmarinic acid), and the evaluation result shows that the quality index quantitative correction model constructed by the traditional Chinese medicine extract quality hyperspectral intelligent analysis method of the invention has a determination coefficient for the accuracy of detection of naringin, neohesperidin and rosmarinic acid of extract 1 sampleBoth greater than 0.708 and both less than 0.900 root mean square error RMSE.
The 3 quality index quantitative correction models constructed by the SG, D2 and WT-based spectrum pretreatment method constructed in comparative example 3 and further based on the LightGBM machine learning algorithm respectively, due to the adoption of the spectrum pretreatment method which is not preferred in the invention, the constructed quality index quantitative correction model has the determination coefficients for the detection accuracy of naringin, neohesperidin and rosmarinic acid of the extract 1 sampleAll less than 0.710, root mean square error RMSE all greater than 0.079; the requirements of the invention on the quantitative correction model of the quality index are not satisfied.
TABLE 4 comparison of extract component detection accuracy assessment results based on different spectral feature extraction methods
LightGBM' is an algorithm model constructed according to comparative example 1
As can be seen from analysis of the data in Table 4, example 1 constructed 3 quantitative correction models of quality index based on LARS, UVE and GA spectral feature extraction method and further based on LightGBM machine learning algorithm, respectively, and the accuracy of these models in detecting extract component content (naringin, neohesperidin and rosmarinic acid) was evaluated, and the evaluation result shows that the quantitative correction model of quality index constructed by the method of high spectral intelligent analysis of Chinese medicinal extract quality of the present invention, for the determination coefficients of naringin, neohesperidin and rosmarinic acid detection accuracy of extract 1 sampleAre all greater than 0.742, and the root mean square error RMSE is all less than 1.350; therefore, the quality index quantitative correction model constructed by the invention shows good detection accuracy of extract naringin, neohesperidin and rosmarinic acid.
In comparative example 1, although the UVE was also used as the spectral feature extraction method and the quantitative correction model of the quality index was constructed based on the LightGBM machine learning method, the manner of constructing the quantitative correction model of the quality index based on the LightGBM machine method was different; the quantitative correction model of quality index obtained in comparative example 1 was evaluated, and the evaluation result showed that the quantitative correction model of quality index, i.e., lightGBM', obtained by the method for constructing quantitative correction model of quality index provided by the invention, had the determination coefficients for the accuracy of naringin, neohesperidin and rosmarinic acid detection of extract 1 sampleThe root mean square error is 1.576, 1.182 and 0.207 for 0.398, 0.566 and 0.612, and compared with the quality index quantitative correction model LightGBM constructed by the same characteristic extraction method, the accuracy of detecting naringin, neohesperidin and rosmarinic acid in extract samples is obviously reduced.
The 3 quality index quantitative correction models constructed by the spectral feature extraction method based on CARS and SPA and further based on the LightGBM machine learning algorithm and constructed by the spectral feature extraction method not preferred in the invention are used, and the determination coefficients of the quality index quantitative correction model constructed by the spectral feature extraction method based on the CARS and SPA for the detection accuracy of naringin, neohesperidin and rosmarinic acid of the extract 1 sample are determinedAll less than 0.710, root mean square error RMSE all greater than 0.079; the requirements of the invention on the quantitative correction model of the quality index are not satisfied.
TABLE 5 comparison of evaluation results of relative density detection accuracy of extractum based on model built by different spectral pretreatment methods
CNN' is the algorithm model constructed in comparative example 2
From analysis of the data in Table 5, it can be seen that example 3 constructs 3 quality index quantitative correction models based on the LARS, UVE and GA spectral feature extraction methods and further based on the CNN machine learning algorithm, and evaluates the accuracy of detecting the relative density of the extract by these models, and the evaluation result shows that the quality index quantitative correction model constructed by the traditional Chinese medicine extract quality hyperspectral intelligent analysis method of the invention has a determination coefficient for the accuracy of detecting the relative density of extract 1Are all greater than 0.816 and,the Root Mean Square Error (RMSE) is smaller than 0.015; therefore, the quality index quantitative correction model constructed by the invention shows good accuracy of detecting the relative density of the extract.
In comparative example 2, although MSC is also adopted as a spectrum preprocessing method and a quality index quantitative correction model is built based on a CNN machine learning method, the mode of building the quality index quantitative correction model based on the CNN machine method is different; the determination coefficient of the established quality index quantitative correction model CNN' for the relative density detection accuracy of the extract 1 extract sample0.732, root mean square error RMSE of 0.009; compared with the quality index quantitative correction model constructed by the CNN machine learning algorithm, the accuracy of detection is obviously reduced.
The 3 quality index quantitative correction models constructed by the SG, D2 and WT-based spectrum pretreatment method constructed in comparative example 4 and further based on CNN machine learning algorithm respectively, because the spectrum pretreatment method not preferred in the invention is adopted, the constructed quality index quantitative correction model has coefficient determined by the accuracy of relative density detection of extract 1 extract sampleThe Root Mean Square Errors (RMSE) are all smaller than 0.672, and the Root Mean Square Errors (RMSE) are all larger than 0.017; the requirements of the invention on the quantitative correction model of the quality index are not satisfied.
In summary, in combination with tables 1 to 5, it can be considered that the quality index quantitative correction model constructed by combining the LightBGM or CNN machine learning algorithm model adopted in embodiments 1 to 4 of the present invention with different spectrum pretreatment methods and feature extraction methods has better accuracy for detecting the relative density of the Chinese medicinal extract or the components of the Chinese medicinal extract.
Finally, it should be noted that the above description is only for illustrating the technical solution of the present invention, and not for limiting the scope of the present invention, and that the simple modification and equivalent substitution of the technical solution of the present invention can be made by those skilled in the art without departing from the spirit and scope of the technical solution of the present invention.
Claims (15)
1. A hyperspectral intelligent analysis method for the quality of Chinese medicinal extract is characterized in that: the method comprises the following steps:
(1) Collecting an extract sample and measuring quality standard data of the extract sample;
(2) Collecting hyperspectral data of the extract sample in the step (1) to obtain a hyperspectral image of the extract sample;
(3) Performing black-and-white correction on the hyperspectral image obtained in the step (2) to obtain a corrected image;
(4) Manufacturing a sample mask by using the correction image obtained in the step (3), wherein the area covered by the sample mask is the region of interest;
(5) Calculating the average spectrum of the region of interest obtained in the step (4);
(6) Performing spectrum pretreatment or spectrum characteristic extraction on the average spectrum in the step (5) to obtain a treated spectrum;
(7) Constructing a quality index quantitative correction model based on a machine learning algorithm according to the spectrum processed in the step (6);
(8) And (3) detecting the extract sample according to the quality index quantitative correction model in the step (7) to obtain a visual result of each sample.
2. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the wavelength range of the hyperspectral image in the step (2) is 900-1800nm in the near infrared band and 400-1000nm in the ultraviolet/visible light band.
3. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the correction formula used for the black-and-white correction in the step (3) is as follows:wherein I is c To correct the image, I is the original hyperspectral image of the collected extract sample, I w Totally reflecting white board picture of polytetrafluoroethylene boardImage, I d The full black image collected by the mirror cover is covered.
4. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the method for manufacturing the sample mask in the step (4) comprises the steps of whitening and eliminating noise.
5. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 4, which is characterized in that: the black-and-white processing is to select a wave band with the largest difference between the hyperspectral of the extract sample and the spectrum of the sample plate, extract an image of the wave band in hyperspectral data, and convert the image into a black-and-white image after binarization processing.
6. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 4, which is characterized in that: the noise elimination is to perform open operation and close operation on the black-and-white image to find out a communication area in the image, set an area threshold value to eliminate the noise in the image, and the reserved pixel point coordinates are the sample mask.
7. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the calculation formula of the average spectrum in the step (5) is as follows:wherein (1)>And m is the number of wave bands contained in the region of interest, which is the average spectrum of the current sample.
8. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 7, which is characterized in that: the saidThe calculation formula of (2) is：/>Wherein (1)>The average reflectivity of the current sample in the wave band w is obtained; n is the number of pixel points in the interested area of the current sample, and is determined by the mask coverage area of each extract sample.
9. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the pretreatment method in the step (6) comprises one or a combination of several treatment methods of Savitsky-Golay smoothing, multi-component scattering correction, orthonormal transformation, first derivative, second derivative and wavelet transformation.
10. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 9, which is characterized in that: the preprocessing method is a multi-element scattering correction processing based on SG smoothing or a standard orthogonal transformation processing based on SG smoothing.
11. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the method for extracting the spectral features in the step (6) comprises competition self-adaptive re-weighted sampling, minimum angle regression, continuous projection algorithm, non-information variable elimination, genetic algorithm, principal component analysis and self-encoder.
12. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 1, which is characterized in that: the machine learning algorithm described in step (7) includes partial least squares, support vector machines, convolutional neural networks, and LightGBM.
13. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 12, which is characterized in that: the machine learning algorithm is a convolutional neural network or a LightGBM.
14. The method for intelligently analyzing the quality hyperspectral of the traditional Chinese medicine extract according to claim 13, which is characterized in that: the convolutional neural network algorithm model consists of 3 convolutional layers, 1 expansion layer and 2 full-connection layers, wherein the convolutional layer activation function adopts LeakyReLu, the full-connection layer activation function adopts Linear, the number of nodes of each convolutional layer is optimized at [16,32,64] by adopting a grid search method, and the convolutional kernel size of each convolutional layer is optimized between 1 and 5; the LightGBM optimizes the number of blades (leave) between [15,31,63], optimizes the learning rate (learning rate) between [0.01,0.1,0.2], optimizes the maximum depth (max_depth) between [3,4,5], and optimizes the minimum data amount (min_data_per_leaf) of each blade between [10,20,30] to construct a model.
15. The use of the hyperspectral intelligent analysis method for the quality of Chinese medicinal extracts according to any one of claims 1 to 14 in the quality analysis of Chinese medicinal extracts.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310768547.5A CN117030628A (en) | 2023-06-27 | 2023-06-27 | High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310768547.5A CN117030628A (en) | 2023-06-27 | 2023-06-27 | High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117030628A true CN117030628A (en) | 2023-11-10 |
Family
ID=88640154
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310768547.5A Pending CN117030628A (en) | 2023-06-27 | 2023-06-27 | High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117030628A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118370788A (en) * | 2024-04-19 | 2024-07-23 | 周至年青保制药有限公司 | Preparation process of traditional Chinese medicine composition for treating urinary incontinence |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109493287A (en) * | 2018-10-10 | 2019-03-19 | 浙江大学 | A kind of quantitative spectra data analysis processing method based on deep learning |
| CN111968080A (en) * | 2020-07-21 | 2020-11-20 | 山东农业大学 | Hyperspectrum and deep learning-based method for detecting internal and external quality of Feicheng peaches |
| CN115909057A (en) * | 2022-11-10 | 2023-04-04 | 厦门中药厂有限公司 | Intelligent calculus bovis identification method based on hyperspectral imaging and application |
| CN115937670A (en) * | 2022-11-10 | 2023-04-07 | 厦门中药厂有限公司 | Intelligent musk identification method based on hyperspectral imaging and application |
| CN116000895A (en) * | 2023-03-28 | 2023-04-25 | 浙江大学 | Quality detection robot and method for traditional Chinese medicine pharmacy process based on deep learning |
-
2023
- 2023-06-27 CN CN202310768547.5A patent/CN117030628A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109493287A (en) * | 2018-10-10 | 2019-03-19 | 浙江大学 | A kind of quantitative spectra data analysis processing method based on deep learning |
| CN111968080A (en) * | 2020-07-21 | 2020-11-20 | 山东农业大学 | Hyperspectrum and deep learning-based method for detecting internal and external quality of Feicheng peaches |
| CN115909057A (en) * | 2022-11-10 | 2023-04-04 | 厦门中药厂有限公司 | Intelligent calculus bovis identification method based on hyperspectral imaging and application |
| CN115937670A (en) * | 2022-11-10 | 2023-04-07 | 厦门中药厂有限公司 | Intelligent musk identification method based on hyperspectral imaging and application |
| CN116000895A (en) * | 2023-03-28 | 2023-04-25 | 浙江大学 | Quality detection robot and method for traditional Chinese medicine pharmacy process based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 仲怿;茹晨雷;张伯礼;程翼宇;: "基于知识图谱的中药制药过程质量控制方法学研究", 中国中药杂志, no. 24, 16 January 2020 (2020-01-16) * |
| 王彩霞;王松磊;贺晓光;董欢;: "高光谱技术融合图像信息的牛肉品种识别方法研究", 光谱学与光谱分析, no. 03, 15 March 2020 (2020-03-15) * |
| 禹文杰;王彩霞;乔芦;贺晓光;何智武;王松磊;: "高光谱成像技术对牛肉水分含量及分布的快速检测", 光电子・激光, no. 03, 15 March 2020 (2020-03-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118370788A (en) * | 2024-04-19 | 2024-07-23 | 周至年青保制药有限公司 | Preparation process of traditional Chinese medicine composition for treating urinary incontinence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Weng et al. | Hyperspectral imaging for accurate determination of rice variety using a deep learning network with multi-feature fusion | |
| Huang et al. | Rapid and nondestructive prediction of amylose and amylopectin contents in sorghum based on hyperspectral imaging | |
| CN101915744B (en) | Near infrared spectrum nondestructive testing method and device for material component content | |
| Qiao et al. | Estimating maize LAI by exploring deep features of vegetation index map from UAV multispectral images | |
| Fan et al. | Nitrogen balance index prediction of winter wheat by canopy hyperspectral transformation and machine learning | |
| CN117030628A (en) | High-spectrum intelligent analysis method for quality of traditional Chinese medicine extract | |
| Lu et al. | Nondestructive testing of pear based on Fourier near-infrared spectroscopy | |
| CN113008805A (en) | Radix angelicae decoction piece quality prediction method based on hyperspectral imaging depth analysis | |
| Jiang et al. | Rapid nondestructive detecting of wheat varieties and mixing ratio by combining hyperspectral imaging and ensemble learning | |
| Gomes et al. | Prediction of sugar content in port wine vintage grapes using machine learning and hyperspectral imaging | |
| Zhong et al. | An online, non-destructive method for simultaneously detecting chemical, biological, and physical properties of herbal injections using hyperspectral imaging with artificial intelligence | |
| CN115937670A (en) | Intelligent musk identification method based on hyperspectral imaging and application | |
| CN111398211A (en) | Information distinguishing and processing method of rhizoma atractylodis granules | |
| CN116840110A (en) | Quality detection method for Guanxinning based on hyperspectral imaging technology and application | |
| Zhang et al. | Machine learning applications for multi-source data of edible crops: A review of current trends and future prospects | |
| Kang et al. | Nondestructive testing model of mango dry matter based on fluorescence hyperspectral imaging technology | |
| Wang et al. | Estimating cotton leaf nitrogen by combining the bands sensitive to nitrogen concentration and oxidase activities using hyperspectral imaging | |
| Wang et al. | The qualitative and quantitative analysis of industrial paraffin contamination levels in rice using spectral pretreatment combined with machine learning models | |
| Zhang et al. | Recognition of the rhizome of red ginseng based on spectral-image dual-scale digital information combined with intelligent algorithms | |
| Ren et al. | Description of tea quality using deep learning and multi-sensor feature fusion | |
| Tang et al. | Research Review on Quality Detection of Fresh Tea Leaves Based on Spectral Technology | |
| Zhang et al. | The rapid detection of the tannin content of grains based on hyperspectral imaging technology and chemometrics | |
| CN114527089A (en) | Rapid nondestructive testing method for soluble solid content in dried red dates | |
| Peng et al. | Determination of malathion content in sorghum grains using hyperspectral imaging technology combined with stacked machine learning models | |
| CN113917003A (en) | Method for detecting content of imperatorin in angelica dahurica decoction pieces |
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 |