WO2015165394A1 - 面向多行业检测的激光拉曼光谱智能化辨识方法及*** - Google Patents

面向多行业检测的激光拉曼光谱智能化辨识方法及*** Download PDF

Info

Publication number
WO2015165394A1
WO2015165394A1 PCT/CN2015/077755 CN2015077755W WO2015165394A1 WO 2015165394 A1 WO2015165394 A1 WO 2015165394A1 CN 2015077755 W CN2015077755 W CN 2015077755W WO 2015165394 A1 WO2015165394 A1 WO 2015165394A1
Authority
WO
WIPO (PCT)
Prior art keywords
spectral data
substance
detection
identification
module
Prior art date
Application number
PCT/CN2015/077755
Other languages
English (en)
French (fr)
Inventor
范广明
尧伟峰
仲雪
倪天瑞
马宁
王中卿
汪春风
李子剑
郭浔
刘春伟
汪泓
Original Assignee
欧普图斯(苏州)光学纳米科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 欧普图斯(苏州)光学纳米科技有限公司 filed Critical 欧普图斯(苏州)光学纳米科技有限公司
Publication of WO2015165394A1 publication Critical patent/WO2015165394A1/zh

Links

Images

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
    • G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
    • G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
    • G01N21/65—Raman scattering

Definitions

  • the invention relates to a laser Raman spectroscopy intelligent automatic identification method and system for multi-industry detection, belonging to the detection application fields of food, medicine, health care products and cosmetics.
  • Health foods are foods that claim to have specific health functions or are intended to supplement vitamins and minerals. However, health foods are not equivalent to medicines. Health foods must have the same safety as food. Long-term use does not cause harm to the human body. Drugs usually have certain side effects. Health foods maintain a balanced state of health by regulating body function, and drugs directly produce pharmacological effects against disease mechanisms. Health foods do not have a strict dosage, but drugs must be taken in strict accordance with the prescribed dosage.
  • PDE-5 inhibitors such as sildenafil are prescription drugs with clear indications, contraindications, and side effects. Some people cannot take it. If the patient does not know it, it may cause serious adverse reactions. It even leads to death. Therefore, these adulterated health foods seriously endanger public health, disrupt market order, and have serious consequences for society and consumers.
  • Illegal additives that may be added to health products include (but are not limited to Table 1):
  • Illegal additives that may be added to pharmaceuticals include (but are not limited to Table 2):
  • Illegal additives that may be added to cosmetics include (but are not limited to Table 3):
  • the prevention and combat of illegally added chemical adulterated health products relies heavily on the ability of analytical testing techniques, especially rapid detection techniques.
  • the rapid inspection technology is based on modern analytical technology and information technology. It has high technical content and can be used in simple laboratories, mobile laboratories or supervised sites and completed in a short period of time, achieving high confidence results. Therefore, the rapid inspection technology is an important means of technical supervision of the health food market. It can achieve targeted inspections, reduce law enforcement costs, increase the technical content of legal supervision, and play a strong technical support role for administrative supervision.
  • Raman spectroscopy has many advantages in detection: Raman spectroscopy obtains the fingerprint spectrum of matter molecules, which has extremely high specificity; Raman scattering has strong penetrating power and can be transparent through glass, plastic, etc. Packing or container, suitable for all kinds of non-destructive rapid detection; Raman spectroscopy is suitable for the detection of aqueous samples, can identify and characterize inorganic compounds; portable Raman spectrometer developed with the development of opto-mechatronics technology, in actual use Very convenient, suitable for testing vehicles and on-site rapid detection; nano-enhanced Raman technology can achieve rapid detection of micro-trace substances, enabling Raman spectroscopy technology to be able to quickly detect illegally added chemical drugs in health foods, and can achieve a variety of Simultaneous rapid detection of substances.
  • it is generally necessary to conduct artificial analysis and comparison of the spectra to obtain conclusions which not only requires the operator to have a high professional level, but also affects the detection efficiency and the
  • the present invention aims to provide a laser Raman spectroscopy intelligent automatic identification method and system for multi-industry detection.
  • the technical content of the present invention is:
  • a laser Raman spectroscopy intelligent identification method for multi-industry detection the steps of which are:
  • the detection software client detects and recognizes the spectral data, saves the result on the client, and transmits the detection result to the cloud for saving; if the cloud identification is selected, the detection software The client sends the spectral data to the cloud for detection and identification and saves the detection result; wherein the method for detecting and identifying the spectral data is:
  • the identification information and the selected extraction are taken out.
  • the threshold information of the Raman characteristic peak is compared. If there is a Raman characteristic peak that meets the condition, the presence of the substance is detected; if the Raman characteristic peak with significant enhancement effect is not selected from the spectral data, the identification method for the setting is set.
  • the spectral data is processed and extracted by the wavelet analysis method, and if it matches the characteristic peak of the substance, the presence of the substance is detected;
  • the identification method for the setting is a substance having a supervised learning method in pattern recognition, and the spectral data is classified according to the labeled sample data of each substance by using a supervised learning classifier to detect whether a corresponding substance exists;
  • For the set identification method is the material of the unsupervised learning method in pattern recognition, calculate the differential value of the sample data of each substance as the feature vector of the substance, calculate the differential value of the spectral data as the feature vector, and then calculate the two features.
  • the similarity of the vector if it is greater than the set threshold, is detected as the presence of the corresponding substance.
  • the spectral data is preprocessed by:
  • the identification information includes: a range of the spectrum of the characteristic peak, a peak intensity, and an area.
  • the detecting software client sets an inquiry interface, and the detecting software client module queries the cloud according to the permission of the login user and the query request, and returns corresponding query information.
  • the detection software client comprises: a spectrum processing module, a configuration management module, an encryption module, a substance category management module, a user management module, a report management module, a spectrum operation module, a spectrum display module, an SOP help module, and a detection The result shows the module.
  • the detected substances are classified by a fixed level, that is, the category structure is classified by means of a folder name plus a configuration file of the same name; or the classification is performed by a free hierarchy, that is, by using a database to check materials and substances.
  • the free combination category structure is classified; or the detection is classified according to the detection items purchased by the user, that is, the detection category structure is organized according to the detection items purchased by different users.
  • the detected substances are displayed in a hierarchical manner: the primary directory is displayed by a large rounded graphic icon plus a description word at the bottom, and the secondary characters are displayed by combining the backgrounds of the different colors with the Chinese characters on the background image.
  • the directory distinguishes whether the user has purchased the test item by graying the icon of the category item or describing the gray process.
  • sample of the substance to be tested is prepared in accordance with a standard operating procedure SOP.
  • a laser Raman spectroscopy intelligent identification system for multi-industry detection which is characterized by comprising a laser Raman spectrometer module, an industry detection software client, and a cloud; wherein
  • the laser Raman spectrometer module is configured to collect spectral data of a sample of a substance to be tested placed in a detection pool of a laser Raman spectrometer under client control, and send the data to an industry testing software client;
  • the industry testing software client is configured to detect and identify the received spectral data, and save the detection result to the cloud; or send the spectral data to the cloud for detection and identification;
  • the cloud is configured to perform detection, identification, storage, and detection result management services on the spectral data, and perform client software and user rights management services, software module upgrade services, and detection category update services;
  • the industry testing software client or the cloud has a Raman spectroscopy database of an industry material, and each substance is provided with an identification method; when detecting and identifying, the Raman characteristic peak is first extracted from the spectral data; if the spectrum is extracted from the spectrum The data selects the Raman characteristic peak with significant enhancement effect. For the material whose identification method is the characteristic peak identification method, the identification information and the selected Raman are taken out. The threshold information of the characteristic peak is compared, and if the condition is satisfied, the presence of the substance is detected; if the Raman characteristic peak with significant enhancement effect is not selected from the spectral data, the wavelet is used for the substance whose identification method is the characteristic identification peak method, and the wavelet is used.
  • the analysis method processes and extracts the characteristic peak of the spectral data, and if it matches the characteristic peak of the substance, the presence of the substance is detected; for the set identification method, the substance having the supervised learning method in the pattern recognition is marked according to each substance
  • the sample data is classified by the supervised learning classifier to detect whether there is a corresponding substance; for the substance whose identification method is the unsupervised learning method in the pattern recognition, the differential value of the sample data of each substance is calculated as the The feature vector of the substance, the differential value of the spectral data is calculated as a feature vector, and then the similarity of the two feature vectors is calculated. If it is greater than the set threshold, the corresponding substance is detected.
  • the detection software client includes a spectral data preprocessing module for processing the acquired spectral data: differentiating the collected spectral data to determine a hot pixel position in the spectral data, if there is a hot pixel
  • the neighboring point mean method is used to compensate the hot pixel points. For the occurrence of consecutive hot pixel points, the size of the hot pixel value is determined first from left to right, and then the mean value is calculated, then the heat is determined from right to left.
  • the size of the pixel value is then calculated by the mean value to obtain the spectral data after the hot pixel is removed; the spectral data of the hot pixel is filtered by the filter such as Boxcar; the smoothing is performed by using the cubic uniform rational B-spline curve After the spectral data is modeled, the spectral data in the modeled pixel mode is obtained; a number of standard materials are selected, and a fitting equation is established for each standard substance, and the spectral data in the pixel mode is converted into a wave number by fitting the equation.
  • the detection software client includes: a client monitoring module, a client browsing module, a spectrum processing module, a configuration management module, an encryption module, a substance category management module, a user management module, a report management module, and a spectrum operation module.
  • the identification information includes: a range of the spectrum of the characteristic peak, a peak intensity, and an area.
  • the system consists of a pre-processing module, a laser Raman spectrometer module, a client module and a server (cloud) module; wherein the server (cloud) module and the client module are connected through a network, the laser Raman spectrometer module and the client module Through the data line wired or wireless network connection, the sample processed by the pre-processing module can be placed in the detection pool of the laser Raman spectrometer, and the client module controls the laser Raman spectrometer module to obtain the test by the laser Raman spectroscopy acquisition control module.
  • the original spectrum of the material is detected qualitatively or quantitatively by the laser Raman spectroscopy module and the laser Raman spectroscopy intelligent identification module.
  • the detection process of the client detection module of the system is: sample pre-processing ⁇ on-machine detection ⁇ automatic identification (identification has two ways: client identification and cloud identification respectively, and the result is returned to the detection client after the cloud identification is completed) ⁇ Generate and print detection reports; in the client and cloud identification module, a large number of intelligent algorithms are used to intelligently and automatically identify samples based on spectral data, including two main categories: spectral preprocessing algorithm and spectral identification algorithm. .
  • the former includes, for example, filters, derivative differentials, polynomial fitting, and wavelet fractionation. Analysis and other methods.
  • the latter includes: feature peak identification and pattern recognition methods, wherein the pattern recognition method includes a supervised classifier method and an unsupervised clustering method.
  • the main process of the client browsing module of the system is: login system ⁇ permission review ⁇ submit data view request ⁇ generate and print view detection report.
  • the data processing and result analysis process of the Raman spectrum of the present invention are automatically completed by a computer, and the determination result is visually displayed on the software interface. Users do not need to know the details of the Raman spectrum generation process and analysis process, just follow the simple steps of the inspection process wizard to get the test results and reports.
  • the reason why Raman spectroscopy automatic identification software can obtain the identification result quickly and accurately depends on the hardware acquisition module and the biological pre-processing method, and relies on the Raman spectroscopy automatic identification system detection process and data processing analysis method.
  • the system can detect food safety problems in the food industry, including: non-edible chemicals, abuse of food additives, adulterated foods, pesticides, veterinary drugs and hormone residues; this system can detect hypoglycemia, blood pressure, Adding prescription ingredients to soothe the nerves, anti-fatigue and weight loss products; this system can detect the addition of prescription ingredients in whitening, acne, anti-dandruff, hair dye and anti-aging products in the cosmetics industry; the detection time is less than 20 minutes, Achieved the requirements of rapid detection.
  • the user can remotely control the Raman spectrometer module through different wired devices (such as tablet, mobile phone, etc.) through wired or wireless network connection, and collect Raman spectral data, which is built in the system.
  • the detection result can be displayed in the detection terminal device.
  • the automatic identification software of the system uses a large number of intelligent algorithms, so that the analysis, processing and identification process of Raman spectroscopy are automatically completed by the computer, and the judgment result is visually displayed on the software interface, which greatly improves the detection. Efficiency provides a better user experience.
  • the server (cloud) module of the system can implement software upgrade and detection substance category update service through the data service interface, and the detection substance category can be expanded according to requirements. When the cloud adds a new detection substance category, the user can synchronously update the detection client. .
  • the invention has the advantages of high detection sensitivity, short time, low cost, no need for pre-treatment or simple pre-treatment, small size, light weight and portability, so it can be used as an effective means for on-site rapid detection and applied to various industries. Rapid testing meets the needs of daily supervision and rapid testing of food and drug supervision departments.
  • Figure 1 is a schematic view of the overall frame of the system of the present invention.
  • Figure 2 is a general flow chart of the method of the present invention
  • Figure 3 is a schematic structural view of the system of the present invention.
  • FIG. 4 is a logic flow diagram of a software for detecting a client module (industry detection software client) according to the present invention
  • FIG. 5 is a flowchart of processing core data of a client module (industry detection software client) according to the present invention
  • FIG. 6 is a structural diagram of a detection client system according to the present invention.
  • FIG. 7 is a structural diagram of a server (cloud) system according to the present invention.
  • FIG. 8 is a system architecture diagram of a browsing client (supervisor platform user terminal) according to the present invention.
  • Figure 9 shows the detection results of a brand of slimming capsule sibutramine
  • Figure 10 shows the results of phenolphthalein test of a brand of slimming capsules
  • Figure 11 shows the detection results of a brand of spirulina slimming capsule sibutramine
  • Figure 12 shows the results of phenolphthalein test of a brand of spirulina slimming capsule
  • Figure 13 is a test result of a brand of Jiali capsule sibutramine
  • Figure 14 shows the results of phenolphthalein test of a brand Jiali capsule
  • Figure 15 shows the results of sildenafil detection of a brand of Shenqi capsule
  • Figure 16 shows the results of sildenafil detection of a brand of Shenqi Capsule
  • Figure 17 shows the results of the detection of metformin hydrochloride in a brand of Sanqi Huangqi Capsule
  • Figure 18 is a test result of a brand of Sanqi Huangqi Capsule rosiglitazone maleate
  • Figure 19 shows the results of the detection of metformin hydrochloride in a brand of hypoglycemic health products
  • Figure 20 is a test result of a brand of hypoglycemic health product rosiglitazone maleate
  • Figure 21 shows the results of the detection of phenformin hydrochloride in a brand of sugar Lining tablets.
  • the laser Raman spectroscopy automatic identification system of the present invention is composed of four modules: 1) pre-processing module, 2) laser Raman spectrometer module, 3) client module (client detection module, client service) Module), 4) server (cloud) module, where:
  • Pre-processing module Prepare the sample of the substance to be tested according to the SOP (Standard Operation Procedure), which provides the necessary guarantee for the accurate and reliable result of the detection system.
  • SOP Standard Operation Procedure
  • Laser Raman spectrometer module The sample of the substance to be tested after being processed by the pre-processing module can be placed in the detection pool of the laser Raman spectrometer, and the collected by the cooperation of the client Raman spectroscopy acquisition control module Spectral data is sent to the industry inspection software client module.
  • the client module includes a client monitoring module and a client browsing module.
  • the client detection module is composed of a core data processing module such as a laser Raman spectroscopy acquisition control module, a laser Raman spectroscopy module, and a laser Raman spectroscopy intelligent identification module.
  • the client software architecture has good scalability and can meet the material testing requirements of multiple industries.
  • the material detection category libraries of different industries are independent of each other, which is convenient for detecting substance category expansion and online update.
  • the test results are accompanied by intuitive red and green indications.
  • the lights and eye-catching text information are automatically displayed, the operation steps are simple, and the user interface is human friendly.
  • the main functions of the client browsing module are as follows: various users browse and query all or part of the monitoring and testing information according to the authority, for example, monthly reports and annual reports related to statistical reports; qualification rate statistics and risk warning information related to security warnings Release, etc.; security dynamics related to information services, policies and regulations, quality standards, etc.
  • Server (cloud) module with good software module scalability, based on big data services, currently provides software module upgrade, detection category update service, user rights management service, detection result storage, cloud result identification service and cloud data storage service.
  • the detection process of the client detection module of the system is: sample pre-processing ⁇ on-machine detection ⁇ automatic identification (identification has two ways: client identification and cloud identification respectively, where the cloud identification ends and the result will be Return to the detection client ⁇ generate and print the detection report; if the user uses the client identification, after the detection is finished, the user can choose to upload the detection result to the database of the server (the cloud) for saving; the main part of the client browsing module of the system
  • the process is: login system ⁇ permission review ⁇ submit data view request ⁇ generate and print view detection report.
  • the preprocessing module of the system includes a nano enhancement module and an SOP operation management module.
  • the laser Raman spectrometer module comprises a laser, a spectrometer and a spectrometer connection module;
  • the client detection module comprises a laser Raman spectroscopy acquisition control module, a laser Raman spectroscopy module, a laser Raman spectroscopy module, a system configuration module, a substance category management module, Quantitative management module, spectral data classification database and encryption module.
  • the laser Raman spectroscopy identification module a large number of intelligent algorithms are used to automatically identify samples based on spectral data, including two main types of algorithms: spectral preprocessing algorithm and spectral identification algorithm.
  • the former includes methods such as: filter, derivative differential, polynomial fitting, and wavelet analysis.
  • the latter includes: feature peak identification and pattern recognition methods, wherein the pattern recognition method includes a supervised classifier method and an unsupervised clustering method.
  • the client browsing module includes a data request module, a data display module, and a report printing module.
  • the server (cloud) module includes: a detection category configuration management module and a corresponding database for managing substance categories and sample categories; a user management module and a corresponding database for managing user rights and user groups; The management module and the database are used for recording and managing the detection results; the cloud identification algorithm module mainly includes the spectrum identification algorithm and the machine learning algorithm.
  • the identification algorithm may be limited due to the limited training samples of the client and the limited computing support of the client processor. The results have a certain impact, so the user can select the cloud database for identification in the process of spectral data identification. At the same time, the newly identified results (substance) can be added to the cloud database to optimize the identification algorithm model through machine learning.
  • the processing flow of the detection client of the present invention is performed. After the user logs in to the client, the user can select the industry category and the substance detection category of the sample to be tested, select whether the sample to be tested is quantitative analysis or qualitative analysis, and detect the client according to the user. The choice is to configure the system to complete the detection of the sample and output the test results.
  • the detection client includes:
  • Raman spectroscopy data acquisition means For the company's different types of Raman spectroscopy, we have a unified interface encapsulation of the communication connection function library in the client software module, so that we can collect spectral data according to a unified data format. That is, through the function of the package, the integration time, the number of scans, and the instrument model parameters are transmitted to obtain pixel position information and spectral intensity information of the spectrum.
  • the set identification information (such as the range of the spectrum of the characteristic peak, peak intensity and area) is used to determine whether or not the Raman characteristic peak to be inspected is contained.
  • the peak analysis technique using wavelet analysis can effectively solve the problem of overlapping Raman characteristic peaks.
  • Unsupervised model clustering method For the problem of no significant feature peaks and no labeling samples, we use hierarchical clustering and non-hierarchical clustering to cluster.
  • the unknown spectrum can be processed by machine learning: when new substance detection is performed, if the spectral data of the new substance cannot be found in the spectral data we have established, then the Raman spectrum of the substance can be directly added to the database. As the basis for the next decision.
  • the laser Raman spectroscopy acquisition control module is used to set the acquisition parameters of the laser Raman spectrometer, and then the function of the package is called in the laser Raman spectroscopy acquisition control module to read the pixel and intensity data of the spectrum, thereby obtaining the original spectrum and laser pulling
  • the spectroscopy analysis module optimizes the original spectrum, and then analyzes and processes the above steps through the laser Raman spectroscopy identification module.
  • the spectral data can be selected for client identification or cloud identification. If the client identification is selected, the pattern recognition method or the feature peak identification method can be used for processing, and finally the detection result can be obtained.
  • Laser Raman spectroscopy acquisition control module and laser Raman spectroscopy module
  • the spectral data in the original pixel mode usually inevitably contains thermal pixel interference. This phenomenon is a well-known phenomenon.
  • the reason for the generation of hot pixels is mainly due to the fact that the CCD photosensitive element in the laser Raman spectrometer contains dead pixels and dead spots. .
  • the data output from the Raman spectral acquisition control module is the original spectral data in the pixel mode.
  • the original spectral data in this pixel mode will be used as the input data of the subsequent Raman spectral analysis module, and the data of the module is analyzed by the Raman spectrum.
  • the processing method will further process the raw spectral data to obtain the ideal spectral data that we need to study and analyze.
  • the hot pixel of the Raman spectrum appears as one or two points that suddenly differ greatly from the absolute value of the adjacent Raman intensity in the spectral data. For such a point, we use differential means to find its position, using the mean value of the neighboring points. The method performs mean compensation on the hot pixel points.
  • the smoothing method is essentially the least square fitting of the window moving polynomial. First, the signal-to-noise ratio of the Raman spectrum is calculated, and then the window size is automatically adjusted according to the signal-to-noise ratio of the Raman spectrum. If the signal-to-noise ratio is small, the vice versa is also However, with such processing, the data can be better smoothed. For the processed data, we call it the spectral data in the pixel mode after smoothing by digital filtering.
  • control points used are not the original spectral data, but the smoothed spectral data, and then the purpose of such fitting is to make the first-order second-order differential more convenient.
  • the data after this data fitting process is called the spectral data in the modeled pixel mode.
  • the data processing object in the above 1.1-1.3 multiple steps is data fitting in the pixel mode.
  • the Raman spectroscopy data under the wave number is usually used as the communication and communication standard, and the fitting equation can be established for a specific standard substance (such as acetonitrile, toluene, benzonitrile, etc.), and the pixel obtained in the previous step is obtained.
  • the spectral data in the mode is converted into spectral data under the wave number, that is, the spectral intensity data in the pixel coordinates is converted into the spectral intensity data in the wave number coordinates by fitting the equation.
  • Our innovation lies in the selection of the Raman peak of the spectral data in the standard material pixel mode.
  • n points are the highest.
  • n-1 order polynomial we choose the expression of the well-known cubic polynomial to fit.
  • spectral data For the processed data, we call the spectral data under the wave number.
  • the Raman spectrum is more complicated, the Raman spectrum is still an information-rich "molecular fingerprint".
  • the spectral data processed by the spectral analysis module is very convenient for analyzing the information of the substance to be tested.
  • the mainstream spectral identification methods are feature peak identification methods and pattern recognition methods. The former requires the characteristic peaks of known substances, combined with the peak finding algorithm to determine whether there are specific material components.
  • a set of industrial material Raman spectroscopy database which includes identification methods and identification peaks containing corresponding substances (for example, the identification peaks of sibutramine in Figures 9, 11, and 13 are 818 cm -1 and 1086 cm -1 ( The identified peak position range is generally ⁇ 3cm -1 of the identified peak position.
  • the identification peaks of phenolphthalein in Figures 10, 12 and 14 are 822cm -1 , 1012cm -1 and 1150cm -1 .
  • FIG. 15 and Figure 16 The identification peaks of dinafur are 624cm -1 , 810cm -1 , 1232cm -1 and 1574cm -1 .
  • the identification peaks of metformin hydrochloride in Figure 17 and Figure 19 are 718cm -1 and 1440cm -1 .
  • Figure 18 and Figure 20 The identification peaks of rosiglitazone were 616 cm -1 , 734 cm -1 , 1176 cm -1 , 1250 cm -1 and 1322 cm -1 , and the identification peaks of phenformin hydrochloride in Fig. 21 were 986 cm -1 and 1192 cm -1 ). Interval range, threshold intensity and area. Through this database, the problem of material identification can be effectively solved.
  • the original spectrum is preprocessed (hot pixel removal, filter smoothing, data fitting, X-axis correction, baseline processing), according to the identification information set in the database (the range range, peak intensity and area of the spectrum of the characteristic peak) To compare one by one, determine whether there is a Raman characteristic peak to be investigated.
  • the wavelet analysis method The essence of the method is to extract the information of different frequency segments of the original signal and display it on the time axis, which can reflect the time domain characteristics of the signal as well as the frequency domain characteristics of the signal.
  • the wavelet analysis algorithm is introduced into the Raman spectral identification process, which can better solve the separation of the weak signal characteristic peaks in the Raman spectrum.
  • labeled samples refer to positive and negative sample data that have been classified in the experimental data
  • the main point of the classifier design is to select the classification model, establish the similarity evaluation index and select the feature range and feature vector.
  • the classifier includes K-nearest neighbor, perceptron, naive Bayes, support vector machine and other classification models.
  • the similarity evaluation indicators include cosine similarity, Euclidean distance, Mahalanobis distance and other indicators.
  • the clustering method includes hierarchical clustering and k-means clustering (non-hierarchical clustering).
  • the key of these algorithms is to select the similarity evaluation index, select the feature range and feature vector, and use the differential value to establish the feature vector for each sample.
  • the similarity evaluation index includes the cosine similarity, the Euclidean distance, the Mahalanobis distance and the like.
  • the architecture diagram of the detection client system of the present invention includes a spectrum processing module, a configuration management module, an encryption module, a substance category management module, a user management module, a report management module, a spectrum operation module, and a spectrum display module. , SOP help module, test result display module, etc.
  • the above modules work together to realize the automatic detection process of the substances to be inspected, and the characteristics of the software UI style are reflected in the following aspects:
  • Free-level detection category implementation through the database according to the material (the carrier where the substance to be tested), according to the material from the combination category structure.
  • ⁇ Sub-user management detection category implementation organize the detection category structure according to the inspection items purchased by different users.
  • the basic framework of the software has good scalability: through the server (cloud) software to establish the detection substance category, edit detection parameters and user purchase information, the client software later implements the client software configuration file through the network synchronization server (cloud) software. And automatic update of the database.
  • the detection substance category adopts the online update method: when the new detection category is released, it is only necessary to unify the detection module of the specific substance category in the server (cloud) software.
  • the user can log in to the client software by using the authorization account of the client software.
  • Data online comparison, understanding of their purchase information and the manufacturer's updated test substance category, can perform purchases and other operations on new substances.
  • One-button detection operation mode calibration, detection, and viewing operation help are all one-button operations.
  • Material detection progress expression dynamic picture, progress bar, description word above the progress bar, status prompt word below the progress bar, together constitute the progress prompt and expression.
  • the overall layout of the software interface uses the outer border and the middle rounded rectangle to represent the overall design style.
  • Customized software disk operation mode In order to facilitate the input operation on the tablet, the operation function of the custom soft keyboard is added.
  • the client software can add software tool modules as needed:
  • Quantitative curve analysis tool module Quantitative experiment can be carried out by collecting samples of different concentrations, and the tool software can make a quantitative curve of the detected substances.
  • ⁇ Spectral management tool module You can view and browse the currently acquired spectra and historical spectra, and perform related spectral operations.
  • ⁇ Report Management Tool Module Edit, modify, and preview test reports, and save test reports as PDF files.
  • FIG. 7 it is a server (cloud) system architecture diagram of the present invention. It includes a cloud database (including big data), a substance category management module, a user management module, a detection result module, a cloud identification module, and a cloud data storage module.
  • manufacturers can analyze and process data in the cloud database (including big data) through the server (cloud) software module, and then obtain valuable classified data information. Manufacturers can also configure, edit, and publish new detection categories through the server (cloud) software module.
  • the user can obtain the classified data required by the manufacturer cloud user by detecting the client software according to the requirement, and the user can also select whether to purchase a new detection substance category according to the demand.
  • FIG. 8 it is a system architecture diagram of a user browsing end of the present invention. Including quality management module (important goods, unqualified goods, quality inspection), various statistical charts, various types of ledger printing, document regulation interface, basic information management, report management module and user management module.
  • quality management module important goods, unqualified goods, quality inspection
  • various statistical charts various types of ledger printing
  • document regulation interface basic information management
  • report management module user management module
  • Embodiment 1 is a diagrammatic representation of Embodiment 1:
  • the test item is sildenafil, a common illegal chemical component, and it is judged whether it contains illegally added chemical components. Two of them are samples. Sildenafil was detected and the results are shown in Table 6:
  • test items are common illegal addition of chemical components such as metformin hydrochloride, phenformin hydrochloride, rosiglitazone maleate, and pioglitazone hydrochloride. Contains illegally added chemical ingredients, of which 3 samples were found to contain illegally added chemical components. The results are shown in Table 7:

Landscapes

  • Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)

Abstract

一种面向多行业检测的激光拉曼光谱智能化辨识方法及***，该方法为：1)将待检物质样品置于激光拉曼光谱仪的检测池中，采集的光谱数据发送至行业检测软件客户端；2)根据选择在客户端或云端辨识并保存检测结果；其中，对该光谱数据检测识别的方法为：建立一行业物质的拉曼光谱数据库，对该光谱数据进行拉曼特征峰提取；如果从该光谱数据选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征峰辨识方法的物质，将其辨识信息与所选取出的拉曼特征峰的阈值信息进行对比，检测是否存在此物质；否则利用小波分析方法对该光谱数据处理；对于辨识方法为模式识别方法的物质，利用分类器对该光谱数据进行分类检测是否存在对应的物质。

Description

面向多行业检测的激光拉曼光谱智能化辨识方法及*** 技术领域
本发明涉及一种面向多行业检测的激光拉曼光谱智能化自动辨识方法及***，属于食品、药品、保健品和化妆品等检测应用领域。
背景技术
自古以来，中国就有食疗保健的传统***衡健康状态，药品则是直接针对疾病机理产生药理作用。保健食品没有严格的服用剂量，但药品必须严格按照规定的剂量服用。
由于保健食品只是通过调节人体自身机能平衡而起作用，所以效果显现通常较为缓慢，但其又需具有特定保健功能，因此很容易成为非法添加药物的对象。不法分子常把可产生与处方药物具有类似感觉作用的化学药物违法添加到保健药品和食品中，从而产生立竿见影的功效来蒙骗使用者，以不法牟利。如在减肥类保健品中添加禁用药物***，在抗疲劳类男性功能保健品中添加处方药物西地那非等。多年临床研究结果显示，使用***可能增加受试者的严重心血管风险，包括心梗、心脏骤停、心血管死亡等，已有多例死亡报告，因此该药物已于2010年10月在包括中国、美国、欧盟等国家和地区停止生产、销售和使用。而像西地那非等PDE-5抑制剂属于处方药，有明确的适应症、禁忌症和副作用，某些人群不能服用，若患者在不知情的情况下摄入，容易引起严重的不良反应，甚至导致死亡。因此，这些掺假保健食品严重的危害了公众健康，扰乱了市场秩序，给社会及消费者带来严重后果。
在保健品中可能添加的非法添加物包括(但不限于表1)：
表1 保健品中可能添加的非法添加物
Figure PCTCN2015077755-appb-000001
Figure PCTCN2015077755-appb-000002
在药品中可能添加的非法添加物包括(但不限于表2)：
表2 药品中可能添加的非法添加物
Figure PCTCN2015077755-appb-000003
在化妆品中可能添加的非法添加物包括(但不限于表3)：
表3 化妆品中可能添加的非法添加物
Figure PCTCN2015077755-appb-000004
Figure PCTCN2015077755-appb-000005
对非法添加化学药品掺假保健品的预防与打击在很大程度上依赖于分析检测技术，尤其是快速检测技术的能力。快检技术建立在现代分析技术和信息技术的基础之上，技术含量较高，能在简易实验室、移动实验室或监督现场使用操作且在较短的时间完成，获得高置信度的结果。因此，快检技术是对保健食品市场进行技术监管的重要手段，可实现针对性地进行抽查检验，降低执法成本，增加依法监管的技术含量，对行政监管起到强有力的技术支撑作用。
近年来，拉曼光谱法在药品、保健品分析中的应用越来越多(参考文献：滕敏，陈俊科，孙予等.注射用针剂药物的拉曼散射光谱研究[J].光散射学报，2010，22(4)：555-557；周群，蔡少青，王建华等.拉曼光谱法快速鉴别黄芩中药材[J].光散射学报，2002，14(3)：166-168；王玉，李忠红，张正行等.拉曼光谱在药物分析中的应用[J].药学学报，2004，39(9)：764-768；曲晓波，赵雨，宋岩等.人参皂苷Rg3的拉曼光谱研究[J].光谱学与光谱分析，2008，28(3)：0569-0571；张进治，汪瑷，陈惠等.吴茱萸生物总碱的TLC-SERS研究[J].光谱学与光谱分析，2008，27(5)：944；张雁，尹利辉，金少鸿.表面增强拉曼光谱法检测微量添加物质的研究[J].中国药事，2012，26(4)：335-339)，中国药典2010版根据这一发展，在附录中新增拉曼光谱法指导原则，进一步促进这一方法在药品、保健品检验中的应用(参考：陈安宇，焦义，刘春伟等.采用纳米增强拉曼光谱检测技术对牛奶中三聚氰胺的检测[J].中国卫生检验杂志，2009，19(8)：1710-1712)。拉曼光谱法在检测方面独具诸多优势：拉曼光谱法获得的是物质分子的指纹光谱，具有极高的特异性；拉曼散射的穿透力强，可以透过玻璃、塑料等透明的包装或容器，适合各种无损快速检测；拉曼光谱法适于水溶液样品检测，可实现无机化合物的鉴定与表征；随着光机电一体化技术的发展而开发的便携式拉曼光谱仪，在实际使用中非常方便，适合检测车及现场快速检测；纳米增强拉曼技术可实现微痕量物质的快速检测，使拉曼光谱技术能够胜任保健食品中非法添加化学药物的快速检 测，并可实现多种物质的同时快速检测。但是在传统的结果分析中，一般需要对光谱进行人为的专业性分析和比对才能得到结论，这样不但要求操作人员具有较高的专业水平，还影响了检测效率和检测可重复性。
发明内容
针对现有技术中存在的技术问题，本发明的目的在于提供一种面向多行业检测的激光拉曼光谱智能化自动辨识方法及***。
本发明的技术内容为：
一种面向多行业检测的激光拉曼光谱智能化辨识方法，其步骤为：
1)将待检物质样品置于激光拉曼光谱仪的检测池中进行光谱数据采集，然后将采集的光谱数据发送至行业检测软件客户端；
2)选择客户端辨识或云端辨识；若选择客户端辨识，检测软件客户端对该光谱数据进行检测识别，在客户端保存结果，同时将检测结果传送到云端保存；若选择云端辨识，检测软件客户端将该光谱数据发送到云端进行检测识别并保存检测结果；其中，对该光谱数据进行检测识别的方法为：
21)建立一行业物质的拉曼光谱数据库，其中每一物质设有一辨识方法；
22)对该光谱数据进行拉曼特征峰提取；如果从该光谱数据选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征峰辨识方法的物质，将其辨识信息与所选取出的拉曼特征峰的阈值信息进行对比，如果存在符合条件的拉曼特征峰，则检测为存在此物质；如果从该光谱数据未选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征辨识峰方法的物质，利用小波分析方法对该光谱数据处理并提取特征峰，如果与该物质的特征峰匹配，则检测为存在此物质；
23)对于设置的辨识方法为模式识别中有监督学习方法的物质，根据每一物质已标注样本数据利用有监督学习分类器对该光谱数据进行分类，检测是否存在对应的物质；
24)对于设置的辨识方法为模式识别中无监督学习方法的物质，计算每一物质的样本数据的微分值作为该物质的特征向量，计算该光谱数据的微分值作为特征向量，然后计算两特征向量的相似度，如果大于设定阈值，则检测为存在对应的物质。
进一步的，该光谱数据进行检测识别之前，对该光谱数据进行预处理，其方法为：
1)对采集的光谱数据进行微分，确定光谱数据中的热像素点位置，如果光谱数据中存在热像素则采用临近点均值方法对热像素点进行均值补偿；对于光谱数据中出现连续多个热像素点，首先对光谱数据从左向右判定一次热像素值的大小，然后做均值计算，再对光谱数据从右向左判定一次热像素值的大小，然后做均值计算，得到热像素移除后的光谱数据；
2)对热像素移除后的光谱数据采用Boxcar等滤波器进行滤波平滑处理；
3)采用三次均匀有理B样条曲线对滤波平滑后的光谱数据进行建模，得到建模后的像素模式下的光谱数据；
4)选取若干标准物质，并对每一标准物质建立一拟合方程，通过拟合方程将像素模式下的光谱数据转换为波数模式下的光谱数据；
5)采用极值算法找到波数模式下的光谱数据的光谱基点位置，然后将所有基点做成基线，以基线对应的光谱强度为参考“0”值，移除步骤4)所得波数模式下的光谱数据的拉曼光谱背底荧光。
进一步的，所述辨识信息包括：特征峰所在光谱的区间范围、峰值强度和面积。
进一步的，所述检测软件客户端设置一查询接口，检测软件客户端模块根据登录用户的权限和查询请求向云端进行查询，并返回对应的查询信息。
进一步的，所述检测软件客户端包括：光谱处理模块，配置管理模块，加密模块，物质类别管理模块，用户管理模块，报表管理模块，谱图操作模块，谱图显示模块，SOP帮助模块，检测结果显示模块。
进一步的，对检测出的物质，采用固定层次进行分类，即通过文件夹名称加同名配置文件的方式来组织类别结构进行分类；或者采用自由层次进行分类，即通过数据库按检材、按物质来自由组合类别结构进行分类；或者根据用户购买的检测项目来对检测进行分类，即按照不同用户购买的检测项目来组织检测类别结构进行分类。
进一步的，对检测出的物质，采用层次化方式进行显示：通过大的圆角图形图标加上底部的描述字来显示一级目录，通过不同颜色的背景结合背景图片上的汉字来显示二级目录，通过将类别项的图标灰色处理或将描述灰色处理来区分用户是否购买了该检测项目。
进一步的，按照标准作业程序SOP制备出所述待检物质样品。
一种面向多行业检测的激光拉曼光谱智能化辨识***，其特征在于包括激光拉曼光谱仪模块，行业检测软件客户端，云端；其中，
所述激光拉曼光谱仪模块，用于在客户端控制下，对置于激光拉曼光谱仪的检测池中的待检物质样品进行光谱数据的采集，并将其发送至行业检测软件客户端；
所述行业检测软件客户端，用于对收到的光谱数据进行检测识别，并将检测结果保存到云端；或者将该光谱数据发送到云端进行检测识别；
所述云端，用于对光谱数据进行检测识别、存储和检测结果管理服务，以及对客户端软件进行和用户权限管理服务、软件模块升级服务和检测类别更新服务；
其中，所述行业检测软件客户端或云端设有一行业物质的拉曼光谱数据库，每一物质设有一辨识方法；进行检测识别时，首先对该光谱数据进行拉曼特征峰提取；如果从该光谱数据选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征峰辨识方法的物质，将其辨识信息与所选取出的拉曼 特征峰的阈值信息进行对比，如果满足条件则检测为存在此物质；如果从该光谱数据未选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征辨识峰方法的物质，利用小波分析方法对该光谱数据处理并提取特征峰，如果与该物质的特征峰匹配，则检测为存在此物质；对于设置的辨识方法为模式识别中有监督学习方法的物质，根据每一物质已标注样本数据利用有监督学习分类器对该光谱数据进行分类，检测是否存在对应的物质；对于设置的辨识方法为模式识别中无监督学习方法的物质，计算每一物质的样本数据的微分值作为该物质的特征向量，计算该光谱数据的微分值作为特征向量，然后计算两特征向量的相似度，如果大于设定阈值，则检测为存在对应的物质。
进一步的，所述检测软件客户端包括一光谱数据预处理模块，用于对采集的光谱数据进行处理：对采集的光谱数据进行微分，确定光谱数据中的热像素点位置，如果存在热像素则采用临近点均值方法对热像素点进行均值补偿；对于出现连续多个热像素点，在先从左向右判定一次热像素值的大小，然后做均值计算后，再从右向左判定一次热像素值的大小，然后做均值计算，得到热像素移除后的光谱数据；对热像素移除后的光谱数据进行Boxcar等滤波器进行滤波平滑处理；采用三次均匀有理B样条曲线对滤波平滑后的光谱数据进行建模，得到建模后的像素模式下光谱数据；选取若干标准物质，并对每一标准物质建立一拟合方程，通过拟合方程将像素模式下的光谱数据转换为波数下的光谱数据；采用极值算法找到波数下的光谱数据的光谱基点位置，然后将所有基点做成基线，以基线对应的光谱强度为参考“0”值，对所得光谱数据的背底荧光进行移除。
进一步的，所述检测软件客户端包括：客户端监测模块，客户端浏览模块，光谱处理模块，配置管理模块，加密模块，物质类别管理模块，用户管理模块，报表管理模块，谱图操作模块，谱图显示模块，SOP帮助模块，检测结果显示模块；所述检测软件客户端设置一查询接口，检测软件客户端模块根据登录用户的权限和查询请求向云端进行查询，并返回对应的查询检测结果。
进一步的，所述辨识信息包括：特征峰所在光谱的区间范围、峰值强度和面积。
本***组成包括前处理模块、激光拉曼光谱仪模块、客户端模块和服务端(云端)模块；其中，服务端(云端)模块与客户端模块通过网络连接，激光拉曼光谱仪模块与客户端模块通过数据线有线或无线网络连接，对于前处理模块处理后的样品，可放置于激光拉曼光谱仪的检测池中，客户端模块通过激光拉曼光谱采集控制模块控制激光拉曼光谱仪模块获取待测物质原始光谱，并经激光拉曼光谱分析模块及激光拉曼光谱智能化辨识模块定性或定量检出待测物质。
本***的客户端检测模块的检测流程为：样品前处理→上机检测→自动辨识(辨识有两种方式：分别为客户端辨识和云端辨识，云端辨识结束后将结果返回给检测客户端)→生成并打印检测报表；其中，客户端和云端辨识模块中都采用了大量的智能算法可根据光谱数据对样品进行智能化自动辨识，其中主要包括两大类：光谱预处理算法和光谱辨识算法。前者包括，诸如：滤波器、导数微分、多项式拟合和小波分 析等方法。后者包括：特征峰辨识和模式识别方法，其中模式识别方法又包括有监督的分类器方法和无监督的聚类方法。本***客户端浏览模块的主要流程为：登陆***→权限审核→提交数据查看请求→生成并打印查看检测报表。
与现有技术相比，本发明的积极效果为：
本发明拉曼光谱的数据处理和结果分析过程都由计算机自动完成，判定结果在软件界面上直观显示。用户不需了解拉曼谱图的产生过程及分析过程等细节，只需按照检测流程向导的几个简单步骤就可以得到检测结果及报表。拉曼光谱自动辨识软件之所以能快速准确的获得辨识结果，一方面依赖于硬件采集模块及生物前处理手段的稳定可靠，还主要依赖于拉曼光谱自动辨识***检测流程及数据处理分析方法；为了方便用户使用及适应平板电脑的操作习惯，我们借助微软WPF技术实现了很多特定的软件界面技术，正是有了这些人性化的检测流程及智能的数据处理分析方法，拉曼光谱自动辨识软件给我们提供了比传统光谱分析软件更大的优势。
本***可以检测食品行业中的食品安全问题，包括：非食用化学物质、滥用食品添加剂、掺假伪劣食品、农药、兽药和激素残留等；本***可检测保健品行业中降糖、降压、安神、抗疲劳和减肥类产品中添加处方药成份等问题；本***可检测化妆品行业中美白、祛痘类、去屑、染发和抗衰老产品中添加处方药成份等问题；检测时间不超过20分钟，达到了快速检测的要求。在检测***的应用过程中，用户可以在不同终端设备(如：平板电脑、手机等)通过有线或无线网络连接方式，远程操控拉曼光谱仪模块，采集拉曼光谱数据，这些光谱数据经过***内置的智能化自动辨识模块处理后,可以在检测终端设备中显示检测结果。在结果判定时，该***的自动辨识软件采用了大量的智能算法，使得拉曼光谱的分析、处理和辨识过程均由计算机自动完成，判定结果在软件界面上直观显示，极大的提升了检测效率，提供了更好的用户体验。本***的服务端(云端)模块可通过数据服务接口实现软件升级、检测物质类别更新服务，检测物质类别可根据需求扩展，当云端增加新的检测物质类别时，用户可在检测客户端同步更新。由于本***应用于多行业，包括：食品、药品、保健品及医学领域的研发应用，这些行业的用户在具体应用研发中地理位置是分散的，各用户检测客户端数据库又是独立的，而检测应用每天又在不停的进行，日积月累,在用户的检测客户端会产生大量的光谱数据。当这些光谱数据通过网络从用户客户端数据库传递到云端数据库，这些海量数据就形成了对企业有价值的大数据。通过对云端大数据的数据挖掘和数据分析有助于我们更好的服务于不同行业的检测需求，给用户提供更有意义的增值服务。
本发明检测灵敏度高、时间短、成本低、无需前处理或仅需简单的前处理、设备体积小、重量轻、便于携带，因此可作为现场快速检测的有效手段，应用于多种行业的现场快速检测，满足食品药监等部门日常监管以及快速检测方面的需求。
附图说明
图1为本发明***总体框架示意图；
图2为本发明方法总体流程图；
图3为本发明***结构示意图；
图4为本发明检测客户端模块(行业检测软件用户端)软件逻辑流程图；
图5为本发明检测客户端模块(行业检测软件用户端)核心数据处理流程图；
图6为本发明检测客户端***架构图；
图7为本发明服务端(云端)***架构图；
图8为本发明浏览客户端(监管平台用户端)***架构图
图9为某牌减肥胶囊***检测结果；
图10为某牌减肥胶囊酚酞检测结果；
图11为某牌螺旋藻减肥胶囊***检测结果；
图12为某牌螺旋藻减肥胶囊酚酞检测结果；
图13为某牌佳丽胶囊***检测结果；
图14为某牌佳丽胶囊酚酞检测结果；
图15为某牌参苓胶囊西地那非检测结果；
图16为某牌参苓胶囊西地那非检测结果；
图17为某牌三七黄芪胶囊盐酸二甲双胍检测结果；
图18为某牌三七黄芪胶囊马来酸罗格列酮检测结果；
图19为某牌降糖类保健品盐酸二甲双胍检测结果；
图20为某牌降糖类保健品马来酸罗格列酮检测结果；
图21为某牌糖立宁片盐酸苯乙双胍检测结果。
具体实施方式
下面结合附图对本发明进行进一步详细描述。
如图1所示，本发明的激光拉曼光谱自动辨识***由四大模块组成：1)前处理模块，2)激光拉曼光谱仪模块，3)客户端模块(客户端检测模块、客户端服务模块)，4)服务端(云端)模块，其中：
1)前处理模块：按照SOP(Standard Operation Procedure,标准作业程序)制备出待检物质样品，为检测***最终得到准确可靠的结果提供了必要的保证。
2)激光拉曼光谱仪模块：对于经前处理模块处理后的待检物质样品，可置于激光拉曼光谱仪的检测池中，通过与客户端拉曼光谱采集控制模块的协同工作，将采集的光谱数据发送至行业检测软件客户端模块。
3)客户端模块：客户端模块包括客户端监测模块和客户端浏览模块。客户端检测模块由激光拉曼光谱采集控制模块、激光拉曼光谱分析模块、激光拉曼光谱智能化辨识模块等核心数据处理模块组成。客户端软件架构具有良好的扩展性，可满足多个行业的物质检测需求，不同行业的物质检测类别库相互独立，方便检测物质类别扩充并可实现在线更新，检测结果伴有直观的红绿指示灯及醒目的文字信息自动显示，操作步骤简便，用户界面人性友好。客户端浏览模块主要功能如下：各类用户根据权限对全部或部分监控检测信息进行浏览和查询服务，例如：统计报表相关的月报表、年报表等；安全预警相关的合格率统计、风险预警信息发布等；信息服务相关的安全动态、政策法规、质量标准等。
4)服务端(云端)模块：具有良好的软件模块扩展性，基于大数据服务，目前提供软件模块升级、检测类别更新服务、用户权限管理服务、检测结果存储、云端结果辨识服务和云端数据存储服务。
如图2所示，本***客户端检测模块的检测流程为：样品前处理→上机检测→自动辨识(辨识有两种方式：分别为客户端辨识和云端辨识，其中云端辨识结束后将结果返回给检测客户端→生成并打印检测报表；如用户采用客户端辨识，在检测结束后用户可以选择把检测结果上传到服务端(云端)的数据库中进行保存；本***客户端浏览模块的主要流程为：登陆***→权限审核→提交数据查看请求→生成并打印查看检测报表。
如图3所示，本***前处理模块中包括纳米增强模块和SOP操作管理模块。激光拉曼光谱仪模块包括激光器、光谱仪和光谱仪连接模块；客户端检测模块包括激光拉曼光谱采集控制模块、激光拉曼光谱分析模块、激光拉曼光谱辨识模块、***配置模块、物质类别管理模块、定量管理模块、光谱数据分类数据库和加密模块等。在激光拉曼光谱辨识模块中采用了大量的智能算法根据光谱数据对样品进行自动辨识，其中主要包括两大类算法：光谱预处理算法和光谱辨识算法。前者包括，诸如：滤波器、导数微分、多项式拟合和小波分析等方法。后者包括：特征峰辨识和模式识别方法，其中模式识别方法又包括有监督的分类器方法和无监督的聚类方法。客户端浏览模块包括数据请求模块、数据显示模块和报表打印模块。服务端(云端)模块包括：检测类别配置管理模块及相应数据库，用于对物质类别、检材类别进行管理；用户管理模块及相应数据库，用于对用户权限、用户组别进行管理；检测结果管理模块及数据库，用于对检测结果记录和管理；云端辨识算法模块主要包括光谱辨识算法和机器学习算法，由于客户端训练样本有限以及客户端处理器运算能力对算法支持有限，可能对辨识算法结果有一定的影响，因此用户在光谱数据辨识过程中可以选择云端数据库进行辨识，同时可以通过机器学习的方法把新辨识的结果(物质)加入到云端数据库中优化辨识算法模型。
如图4所示为本发明检测客户端处理流程图，用户登录客户端后可以选择待检测样品的行业类别和物质检测类别、选择对待检样品是进行定量分析还是定性分析、检测客户端根据用户的选择对***进行配置，完成对样品的检测并输出检测结果。其中检测客户端包括：
1)拉曼光谱的数据采集、数据处理和结果辨识算法集。
●拉曼光谱的数据采集手段：对于公司不同型号的拉曼光谱仪，我们在客户端软件模块中，对通讯连接函数库进行了统一的接口封装，使得我们能够按照统一的数据格式采集光谱数据。即通过封装的函数，传递积分时间、扫描次数和仪器型号参数从而获取光谱的像素位置信息和光谱强度信息。
●拉曼光谱数据的优化处理手段：底层获取的数据往往包含噪声、荧光干扰信号，为了减少噪声及荧光干扰信号的影响，我们采用了如下的技术手段来对拉曼光谱数据进行分析和处理：
■数据平滑算法。采用多点连续平滑。(参考文献：Abraham.Savitzky,M.J.E.Golay.Smoothing and Differentiation ofData by Simplified Least Squares Procedures.[J]Anal.Chem.,1964,36(8),pp 1627–1639)
■数据拟合算法。采用高次曲线拟合。(参考文献：Meier,R.J.Vib.Spectrosc.2005,39,266–269)；
■基线处理算法。采用高阶(含一阶)微分方法、极大极小数值法，这种算法在一定程度上有助于消除背底荧光对有效信号的影响。(参考文献：Andrzej Kwiatkowski1，Marcin Gnyba1，Janusz Smulko1，
Figure PCTCN2015077755-appb-000006
Wierzba1.Alogrithms ofChemicals ofDetection Using Raman Spectra MetrologyAnd Measurement Systems.[J]MetrologyAnd Measurement Systems.Vol XVII(2010),No 4,Pages 549–560.)；
●拉曼光谱特征峰辨识算法手段:
单(多)峰辨识阈值算法。通过设定的辨识信息(如特征峰所在光谱的区间范围、峰值强度和面积)从而确定是否含有待查的拉曼特征峰。对于辨识峰阈值判定失效的情况，采用用小波分析等分峰技术手段，能有效解决拉曼特征峰重叠的问题。
●拉曼光谱模式识别的方式：
■有监督模型分类器的方法：对于无显著特征峰同时有一定标注样本的问题，我们采用有监督模型的分类器来这类问题。
■无监督模型聚类的方法：对于无显著特征峰又没有标注样本的问题，我们采用层次聚类、非层次聚类的方式来进行聚类。
2)拉曼光谱数据库和机器学习方法。
●建立拉曼光谱数据库:建立的过程如下,首先通过拉曼光谱仪测出标准物质的拉曼图谱(和上文拉曼光谱数据同),然后用软件通过执行数据库操作程序,将谱图和相关信息存入数据库。其中相关信息详见下文提到的拉曼光谱属性集。
●拉曼光谱增加属性集：为了解决拉曼光谱在数据库中存储查询方便的问题，可以通过创建拉曼光谱属性集的方式来帮助对光谱数据的分析和理解。未知光谱可以通过机器学习的方法来处理：当进行新的物质检测时，如果新物质的光谱数据不能在我们已经建立的光谱数据中查找到，那可直接将该物质拉曼光谱添加到数据库中，作为下一次判定的依据。
如图5所示，为本发明检测客户端对光谱数据进行分析处理的流程图。首先通过激光拉曼光谱采集控制模块对激光拉曼光谱仪设置采集参数，然后在激光拉曼光谱采集控制模块中调用封装的函数来读取光谱的像素和强度数据，从而得到原始光谱，经激光拉曼光谱分析模块对原始光谱进行优化处理，再通过激光拉曼光谱辨识模块对前述步骤分析处理，经过这样处理后的光谱数据可以选择进行客户端辨识或者云端辨识。如果选择客户端辨识，可采用模式识别方法或特征峰辨识方法进行处理，最终可以得到检测结果。
1 激光拉曼光谱采集控制模块及激光拉曼光谱分析模块：
我们应用自主研发的RamTracer-200系列激光拉曼光谱仪，通过在激光拉曼光谱采集控制模块中设置激光功率、平滑系数、扫描次数等参数,获取像素模式下光谱数据，再通过X轴校正转化获取波数模式下光谱数据，从而得到波数模式下光谱数据。
拉曼光谱中原始像素模式下光谱数据通常不可避免的会含有热像素干扰，这种现象为公知现象，热像素产生的原因主要是由于激光拉曼光谱仪中CCD感光元器件含有坏点及死点。
1.1)热像素移除：
从拉曼光谱采集控制模块中输出的数据为像素模式下的原始光谱数据，这份像素模式下的原始光谱数据将作为其后续拉曼光谱分析模块的输入数据，通过拉曼光谱分析模块的数据处理方法将对原始光谱数据进一步加工，从而得到我们研究分析需要的理想光谱数据。拉曼光谱的热像素表现为在光谱数据中突然出现一个或者两个与相邻拉曼强度绝对值相差极大的点，对于这样的点我们采用微分的手段，找到其位置，采用临近点均值方法，对热像素点进行均值补偿。对连续多个热像素我们增加了双向判断机制，即首先从左向右(从拉曼光谱数据的起始位置到拉曼光谱数据的结束位置,下文同)判定一次热像素值的大小，然后做均值计算后，再从右向左做一次同样的判定和计算，这样通过两次判定能避免热像素遗漏。对于处理过后的数据我们称为经过热像素移除后像素模式下的光谱数据。
1.2)数据滤波平滑算法：
经过步骤1.1处理后的像素模式下光谱数据，不可避免的还是会存在一些噪声，为了更好的满足我们的分析需求既保护特征峰强度又同时能较为有效的去除噪声，我们采用了多点连续平滑方法，其实质是窗口移动多项式最小二乘拟合，首先计算出拉曼光谱的信噪比，然后根据拉曼光谱的信噪比自动调整窗口大小，如果信噪比大窗口小，反之亦然，通过这样的处理，能够对数据进行更好的平滑。对于处理后的数据，我们称之为经过数字滤波平滑后像素模式下的光谱数据。
1.3)数据拟合算法:
经过步骤1.2处理后的像素模式下光谱数据，已经具备了较高的分析价值，但由于离散的点列几何属性不多，为了便于进一步分析拉曼光谱数据的特点，我们选择了三次均匀有理B样条曲线对上述数据进行拟合。以下是三次均匀有理B样条的方程(下文中Pi为控制点)：
Figure PCTCN2015077755-appb-000007
其中t∈[0，1]
我们的创新点在于，采用的控制点并不是原始的光谱数据，而是经过平滑处理后的光谱数据，然后进行这样的拟合目的是为了后续更方便的求一阶二阶微分。对于经过这种数据拟合处理后的数据我们称为经过建模后的像素模式下的光谱数据。
1.4)X轴校正:
在以上1.1-1.3多个步骤中的数据处理对象为像素模式下的数据拟合。在拉曼光谱分析研究中，通常以波数下的拉曼光谱数据作为沟通及交流标准，可以针对特定标准物质(比如：乙腈、甲苯、苯甲腈等)建立拟合方程，将上一步所得像素模式下的光谱数据转化为波数下的光谱数据，即通过拟合方程将像素坐标下光谱强度数据转化为波数坐标下光谱强度数据。我们的创新点在于，对于标准物质像素模式下光谱数据的拉曼峰选取，拿乙腈为例，我们选取的信噪比大于3最能表征分子基团的特征峰，根据公知，n个点最高可以拟合出n-1阶多项式，我们选择了公知的三次多项式的表达方式来进行拟合。对于处理后的数据我们称为波数下的光谱数据。
1.5)基线处理算法:
对于采用1.4步骤处理后的数据，还可能存在荧光干扰有效信号等问题，我们采用极值算法来找到参考光谱的基点位置，然后以这些基点做基线。设定基线对应的光谱强度为参考“0”值，将这些相邻“0”值作为参考点，顺次两两相连得到基线,然后用原始拉曼光谱的强度值和基线上相应位置的强度值做减法，从而实现移除拉曼光谱背底荧光的功能。我们的创新点在于，对于极小值基点增加了阈值筛查功能，只有在满足阈值条件下的极小值点才是真正基点。经过这个步骤处理后的拉曼光谱数据可以做为后续拉曼光谱辨识模块的输入数据。
2.激光拉曼光谱辨识模块：
尽管拉曼光谱比较复杂，但拉曼光谱仍然是一种富含信息的“分子指纹图谱”，经过光谱分析模块处理后的光谱数据对于分析待检物质信息带了极大的便利。主流的光谱辨识手段是特征峰辨识方法和模式识别方法。前者需要已知物质的特征峰，结合寻峰算法来判定是否存在特定物质成分，我们通过多年的研究与探索，结合不同行业的大量实践，积累了大量的拉曼光谱行业应用经验数据，建立了一套行业物质的拉 曼光谱数据库，该数据库包括含相应种物质的辨识方法、辨识峰(例如：图9、图11、图13中***的辨识峰是818cm-1和1086cm-1(辨识峰位范围一般为所标示的辨识峰位±3cm-1)，图10、图12、图14中酚酞的辨识峰是822cm-1、1012cm-1和1150cm-1，图15、图16中西地那非的辨识峰是624cm-1、810cm-1、1232cm-1和1574cm-1，图17、图19中盐酸二甲双胍的辨识峰是718cm-1和1440cm-1，图18、图20中马来酸罗格列酮的辨识峰是616cm-1、734cm-1、1176cm-1、1250cm-1和1322cm-1，图21中盐酸苯乙双胍的辨识峰是986cm-1和1192cm-1)、区间范围、阈值强度和面积。通过这个数据库可以有效解决物质辨识认定问题。我们的辨识峰和公知的拉曼特征峰的区别在于，我们只选择增强效果显著的拉曼特征峰，其中增强效果显著是指，增强模式下该峰强度大于非增强模式下该峰强度2倍及以上。后者经常用在解决无显著特征峰的拉曼光谱数据辨识上。
2.1)特征峰辨识方法(如果处理结果有显著特征峰)；
2.1.1)单(多)峰辨识算法:
将原始光谱经过预处理(热像素移除、滤波平滑、数据拟合、X轴校正、基线处理)后，根据数据库中设定的辨识信息(特征峰所在光谱的区间范围、峰值强度和面积)来进行逐一比对，确定是否含有待查的拉曼特征峰。
2.1.2)非显著特征峰的辨识算法(如果未找到显著辨识峰，同时数据库中该物质辨识方法设定为特征辨识峰法):
在做针对微恒量物质增强实验的时候，尽管我们SOP来指导每次的实验，但是由于操作人员的经验差异，还是可能出现：2.1.1判定失败的情况,于是我们引入了小波分析方法,该方法的实质是把原始信号不同频率段的信息抽取出来，并将其显示在时间轴上，这样既可反映信号的时域特征也可反映信号的频域特征。在拉曼光谱辨识处理中引入小波分析的算法，能较好的解决拉曼光谱中弱信号特征峰的分离。
2.2)模式识别方法
2.2.1)有监督学习分类器算法:
对于无显著特征峰且有一定标注样本集的数据，(其中：标注样本是指在实验数据中已经分类的阳性样本数据和阴性样本数据)我们采用有监督学习分类器来对未知光谱数据进行分类。分类器设计的要点在于选择分类模型、建立相似度评价指标和选取特征范围及特征向量。其中分类器包括K近邻、感知器、朴素贝叶斯、支持向量机等分类模型，相似度评价指标包括余弦相似度、欧式距离、马氏距离等指标，在特征向量的选取方面，针对不同行业应用类别，自动选取特征范围，对光谱分析处理后的数据采用微分值作为特征向量值。
2.2.2)无监督学习聚类算法:
对于无显著特征峰又没有标注样本的数据，我们采用无监督聚类算法来对未知光谱数据进行聚类。聚类方法包括层次聚类，k均值聚类(非层次聚类)等方法，这些算法的关键在于：选取相似度评价指标和选取特征范围和特征向量，采用微分值为每个样本建立特征向量。根据样本的特征向量计算样本之间的相似度，可将相似度大的样本族聚合到一个类别中，其中相似度评价指标包括余弦相似度、欧式距离、马氏距离等指标。通过这些技术手段，我们可将同一样本中含有相近的信息的物质归类。
如图6所示，为本发明检测客户端***架构图；包括光谱处理模块，配置管理模块，加密模块，物质类别管理模块，用户管理模块，报表管理模块，谱图操作模块，谱图显示模块，SOP帮助模块，检测结果显示模块等。上述模块协同工作，实现了对待检物质的自动化检测流程，其中涉及到软件UI风格特点体现在如下几个方面：
1)检测物质的分类方法和层次化的显示方式。
●检测物质的分类方法：
-固定层次检测类别实现方式：通过文件夹名称加同名配置文件的方式来组织类别结构。
■自由层次检测类别实现方式：通过数据库按检材(待测物质所在的载体)、按物质来自由组合类别结构。
■分用户管理检测类别实现方式：按照不同用户购买的检测项目来组织检测类别结构。
●检测物质的层次化显示方式上：
■通过大的圆角图形图标加上底部的描述字来显示一级目录。
■通过不同颜色的背景结合背景图片上的汉字来显示二级目录。
■通过将类别项的图标灰色处理或将描述灰色处理来区分用户是否购买了该检测项目。
■通过界面飞入(匀速和非匀速)、淡入淡出等效果来实现检测项目的更迭。
2)检测物质的扩充、发布和更新方式。
●软件的基础框架有良好的扩展性：通过服务端(云端)软件建立检测物质类别、编辑检测参数和用户购买信息，客户端软件通过网络同步服务端(云端)软件后来实现客户端软件配置文件和数据库的自动更新。
●检测物质类别采用在线更新的方式：当发布新检测类别时候，只需要统一在服务端(云端)软件更新特定物质类别的检测模块即可,用户通过登录客户端软件的授权账号，同服务器上数据在线比对，了解自己的购买信息和厂家更新的检测物质类别，可对新增物质执行购买等操作。
3)为了适应平板电脑的操作习惯，采用新颖的人性化软件设计风格。
●一键式的检测操作方式：校正、检测、查看操作帮助等都是一键式操作。
●物质检测结显示方式：通过红绿黄灯、红绿黄字、不同的声音来描述结果。
●物质检测进度表达方式：动态图片、进度条、进度条上方的描述字、进度条下方的状态提示字，共同构成了进度提示及表达方式。
●隐藏弹出菜单组的表达方式：为了不占用检测界面布局中更多的空间，菜单组可隐藏可弹出。
●软件界面的整体布局方式：软件界面采用外边框加中间圆角矩形框的方式来表现整体的设计风格。
●自定义软件盘的操作方式：为了便于在平板电脑上输入操作，增加了自定义软键盘的操作功能。
4)客户端软件可按需求增添软件工具模块：
●定量曲线分析工具模块：可通过采集不同浓度的样本进行定量实验，同时该工具软件可以制作检测物质的定量曲线。
●谱图管理工具模块：对当前采集的谱图和历史谱图都可查看和浏览，并可执行相关的谱图操作。
●报表管理工具模块：编辑、修改和预览检测报表，并可将检测报表保存为PDF格式文件。
如图7所示，为本发明服务端(云端)***架构图。包含云端数据库(含大数据)、物质类别管理模块、用户管理模块、检测结果模块、云端辨识模块和云端数据存储模块。一方面，厂家可通过服务端(云端)软件模块对云端数据库(含大数据)进行数据分析和处理，进而得到有价值的分类数据信息。厂家还可以通过服务端(云端)软件模块配置、编辑、发布新的检测类别。另一方面，用户可根据需求通过检测客户端软件获取厂家云端用户所需要的分类数据，用户还可根据需求来选择是否购买新的检测物质类别。
如图8所示，为本发明用户浏览端的***架构图。包括质量管理模块(重要商品、不合格商品、质量检测)、各类统计图表、各类台账打印、文件法规界面、基础信息管理、报表管理模块和用户管理模块等。
实施例一：
采购了5种市售声称具有减肥功能的保健品，对其进行检测，检测项目为常见非法添加化学成分***、酚酞，判断其是否含有非法添加化学成分，其中3种样品均检出***及酚酞，结果如表4所示：
表4 减肥类保健品检测结果
Figure PCTCN2015077755-appb-000008
具体情况如下：
1.1 某牌减肥胶囊
标示名称 某牌减肥胶囊
剂型 胶囊剂
检测项目 ***，酚酞
***拉曼方法检测结果 检出，含量>0.1％；仪器检出限：0.5×10-3mg/g；检测结果如图9所示
酚酞的拉曼方法检测结果 检出，含量>0.1％；仪器检出限：10×10-3mg/g；检测结果如图10所示
备注 采用HPLC进行验证，结果一致。含***0.23％，酚酞7.67％
1.2 某牌螺旋藻减肥胶囊
标示名称 某牌螺旋藻减肥胶囊
剂型 胶囊剂
检测项目 ***，酚酞
***拉曼方法检测结果 检出，含量>0.1％；仪器检出限：0.5×10-3mg/g，检测结果如图11所示
酚酞的拉曼方法检测结果 检出，含量>0.1％；仪器检出限：10×10-3mg/g；检测结果如图12所示
1.3 某牌佳丽胶囊(绿瘦)
标示名称 某牌佳丽胶囊
剂型 胶囊剂
检测项目 ***，酚酞
***拉曼方法检测结果 检出，含量>0.1％；仪器检出限：0.5×10-3mg/g；检测结果如图13所示
酚酞的拉曼方法检测结果 检出，含量>0.1％；仪器检出限：10×10-3mg/g；检测结果如图14所示
备注 采用HPLC进行验证，结果一致。含***0.24％，酚酞12.37％
实施例二：抗疲劳类保健品
采购了3种市售声称具有抗疲劳、增强免疫力功能的保健品，对其进行检测，检测项目为常见非法添加化学成分西地那非，判断其是否含有非法添加化学成分，其中2种样品检出西地那非，结果如表6：
表6 抗疲劳类保健品检测结果
Figure PCTCN2015077755-appb-000009
具体情况如下：
2.1 某牌参苓胶囊
标示名称 某牌参苓胶囊
剂型 胶囊剂
检测项目 西地那非
西地那非拉曼方法检测结果 检出，含量>0.1％；仪器检出限：0.01×10-3mg/g；检测结果如图15所示
备注 采用HPLC进行验证，结果一致。含西地那非7.61％
2.2 某牌参苓胶囊
标示名称 某牌参苓胶囊
剂型 胶囊剂
检测项目 西地那非
西地那非拉曼方法检测结果 检出，含量>0.1％；仪器检出限：0.01×10-3mg/g；检测结果如图16所示
实施例三、降糖类保健品
采购了6种市售声称具有降血糖功能的保健品，对其进行检测，检测项目为常见非法添加化学成分盐酸二甲双胍、盐酸苯乙双胍、马来酸罗格列酮、盐酸吡格列酮，判断其是否含有非法添加化学成分，其中3种样品检出含非法添加化学成分，结果如表7：
表7 降糖类保健品的拉曼方法检测结果
Figure PCTCN2015077755-appb-000010
具体情况如下：
3.1 某牌三七黄芪胶囊
Figure PCTCN2015077755-appb-000011
3.2 某降糖类保健品
Figure PCTCN2015077755-appb-000012
3.3 某牌糖立宁片
标示名称 某牌糖立宁片
剂型 胶囊剂
检测项目 盐酸二甲双胍，盐酸苯乙双胍，马来酸罗格列酮，盐酸吡格列酮
盐酸二甲双胍检测结果 阴性，未检出；仪器检出限：0.1×10-3mg/g
盐酸苯乙双胍检测结果 检出，含量>0.1％；仪器检出限：0.1×10-3mg/g；检测结果如图21所示
马来酸罗格列酮检测结果 阴性，未检出；仪器检出限：0.1×10-3mg/g
盐酸吡格列酮检测结果 阴性，未检出；仪器检出限：0.5×10-3mg/g
格列吡检测结果 阴性，未检出；仪器检出限：1×10-3mg/g
格列齐特检测结果 阴性，未检出；仪器检出限：0.5×10-3mg/g
瑞格列奈检测结果 阴性，未检出；仪器检出限：1×10-3mg/g

Claims (12)

  1. 一种面向多行业检测的激光拉曼光谱智能化辨识方法，其步骤为：
    1)将待检物质样品置于激光拉曼光谱仪的检测池中进行光谱数据采集，然后将采集的光谱数据发送至行业检测软件客户端；
    2)选择客户端辨识或云端辨识；若选择客户端辨识，检测软件客户端对该光谱数据进行检测识别，在客户端保存结果，同时将检测结果传送到云端保存；若选择云端辨识，检测软件客户端将该光谱数据发送到云端进行检测识别并保存检测结果；其中，对该光谱数据进行检测识别的方法为：
    21)建立一行业物质的拉曼光谱数据库，其中每一物质设有一辨识方法；
    22)对该光谱数据进行拉曼特征峰提取；如果从该光谱数据选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征峰辨识方法的物质，将其辨识信息与所选取出的拉曼特征峰的阈值信息进行对比，如果存在符合条件的拉曼特征峰，则检测为存在此物质；如果从该光谱数据未选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征辨识峰方法的物质，利用小波分析方法对该光谱数据处理并提取特征峰，如果与该物质的特征峰匹配，则检测为存在此物质；
    23)对于设置的辨识方法为模式识别中有监督学习方法的物质，根据每一物质已标注样本数据利用有监督学习分类器对该光谱数据进行分类，检测是否存在对应的物质；
    24)对于设置的辨识方法为模式识别中无监督学习方法的物质，计算每一物质的样本数据的微分值作为该物质的特征向量，计算该光谱数据的微分值作为特征向量，然后计算两特征向量的相似度，如果大于设定阈值，则检测为存在对应的物质。
  2. 如权利要求1所述的方法，其特征在于该光谱数据进行检测识别之前，对该光谱数据进行预处理，其方法为：
    1)对采集的光谱数据进行微分，确定光谱数据中的热像素点位置，如果光谱数据中存在热像素则采用临近点均值方法对热像素点进行均值补偿；对于光谱数据中出现连续多个热像素点，首先对光谱数据从左向右判定一次热像素值的大小，然后做均值计算，再对光谱数据从右向左判定一次热像素值的大小，然后做均值计算，得到热像素移除后的光谱数据；
    2)对热像素移除后的光谱数据采用Boxcar滤波器进行滤波平滑处理；
    3)采用三次均匀有理B样条曲线对滤波平滑后的光谱数据进行建模，得到建模后的像素模式下的光谱数据；
    4)选取若干标准物质，并对每一标准物质建立一拟合方程，通过拟合方程将像素模式下的光谱数据转换为波数模式下的光谱数据；
    5)采用极值算法找到波数模式下的光谱数据的光谱基点位置，然后将所有基点做成基线，以基线对 应的光谱强度为参考“0”值，移除步骤4)所得波数模式下的光谱数据的拉曼光谱背底荧光。
  3. 如权利要求1或2所述的方法，其特征在于所述辨识信息包括：特征峰所在光谱的区间范围、峰值强度和面积。
  4. 如权利要求1或2所述的方法，其特征在于所述检测软件客户端设置一查询接口，检测软件客户端模块根据登录用户的权限和查询请求向云端进行查询，并返回对应的查询信息。
  5. 如权利要求1或2所述的方法，其特征在于所述检测软件客户端包括：光谱处理模块，配置管理模块，加密模块，物质类别管理模块，用户管理模块，报表管理模块，谱图操作模块，谱图显示模块，SOP帮助模块，检测结果显示模块。
  6. 如权利要求1所述的方法，其特征在于对检测出的物质，采用固定层次进行分类，即通过文件夹名称加同名配置文件的方式来组织类别结构进行分类；或者采用自由层次进行分类，即通过数据库按检材、按物质来自由组合类别结构进行分类；或者根据用户购买的检测项目来对检测进行分类，即按照不同用户购买的检测项目来组织检测类别结构进行分类。
  7. 如权利要求1所述的方法，其特征在于对检测出的物质，采用层次化方式进行显示：通过大的圆角图形图标加上底部的描述字来显示一级目录，通过不同颜色的背景结合背景图片上的汉字来显示二级目录，通过将类别项的图标灰色处理或将描述灰色处理来区分用户是否购买了该检测项目。
  8. 如权利要求1或2所述的方法，其特征在于按照标准作业程序SOP制备出所述待检物质样品。
  9. 一种面向多行业检测的激光拉曼光谱智能化辨识***，其特征在于包括激光拉曼光谱仪模块，行业检测软件客户端，云端；其中，
    所述激光拉曼光谱仪模块，用于在客户端控制下，对置于激光拉曼光谱仪的检测池中的待检物质样品进行光谱数据的采集，并将其发送至行业检测软件客户端；
    所述行业检测软件客户端，用于对收到的光谱数据进行检测识别，并将检测结果保存到云端；或者将该光谱数据发送到云端进行检测识别；
    所述云端，用于对光谱数据进行检测识别、存储和检测结果管理服务，以及对客户端软件进行用户权限管理服务、软件模块升级服务和检测类别更新服务；
    其中，所述行业检测软件客户端或云端设有一行业物质的拉曼光谱数据库，每一物质设有一辨识方法；进行检测识别时，首先对该光谱数据进行拉曼特征峰提取；如果从该光谱数据选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征峰辨识方法的物质，将其辨识信息与所选取出的拉曼特征峰的阈值信息进行对比，如果满足条件则检测为存在此物质；如果从该光谱数据未选取出增强效果显著的拉曼特征峰，对于设置的辨识方法为特征辨识峰方法的物质，利用小波分析方法对该光谱数据处理并提取特征峰，如果与该物质的特征峰匹配，则检测为存在此物质；对于设置的辨识方法为模 式识别中有监督学习方法的物质，根据每一物质已标注样本数据利用有监督学习分类器对该光谱数据进行分类，检测是否存在对应的物质；对于设置的辨识方法为模式识别中无监督学习方法的物质，计算每一物质的样本数据的微分值作为该物质的特征向量，计算该光谱数据的微分值作为特征向量，然后计算两特征向量的相似度，如果大于设定阈值，则检测为存在对应的物质。
  10. 如权利要求9所述的***，其特征在于所述检测软件客户端包括一光谱数据预处理模块，用于对采集的光谱数据进行处理：对采集的光谱数据进行微分，确定光谱数据中的热像素点位置，如果存在热像素则采用临近点均值方法对热像素点进行均值补偿；对于出现连续多个热像素点，在先从左向右判定一次热像素值的大小，然后做均值计算后，再从右向左判定一次热像素值的大小，然后做均值计算，得到热像素移除后的光谱数据；对热像素移除后的光谱数据进行Boxcar滤波器进行滤波平滑处理；采用三次均匀有理B样条曲线对滤波平滑后的光谱数据进行建模，得到建模后的像素模式下光谱数据；选取若干标准物质，并对每一标准物质建立一拟合方程，通过拟合方程将像素模式下的光谱数据转换为波数下的光谱数据；采用极值算法找到波数下的光谱数据的光谱基点位置，然后将所有基点做成基线，以基线对应的光谱强度为参考“0”值，对所得光谱数据的背底荧光进行移除。
  11. 如权利要求9所述的***，其特征在于所述检测软件客户端包括：客户端监测模块，客户端浏览模块，光谱处理模块，配置管理模块，加密模块，物质类别管理模块，用户管理模块，报表管理模块，谱图操作模块，谱图显示模块，SOP帮助模块，检测结果显示模块；所述检测软件客户端设置一查询接口，检测软件客户端模块根据登录用户的权限和查询请求向云端进行查询，并返回对应的查询检测结果。
  12. 如权利要求9或10或11所述的***，其特征在于所述辨识信息包括：特征峰所在光谱的区间范围、峰值强度和面积。
PCT/CN2015/077755 2013-05-31 2015-04-29 面向多行业检测的激光拉曼光谱智能化辨识方法及*** WO2015165394A1 (zh)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
CN201310214324 2013-05-31
CN201410181459.6 2014-04-30
CN201410181459.6A CN104215623B (zh) 2013-05-31 2014-04-30 面向多行业检测的激光拉曼光谱智能化辨识方法及***

Publications (1)

Publication Number Publication Date
WO2015165394A1 true WO2015165394A1 (zh) 2015-11-05

Family

ID=52097341

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/077755 WO2015165394A1 (zh) 2013-05-31 2015-04-29 面向多行业检测的激光拉曼光谱智能化辨识方法及***

Country Status (2)

Country Link
CN (1) CN104215623B (zh)
WO (1) WO2015165394A1 (zh)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107238595A (zh) * 2017-05-05 2017-10-10 浙江大学 封闭容器的酒精浓度测量装置及测量方法
CN107340547A (zh) * 2017-07-24 2017-11-10 山东省职业卫生与职业病防治研究院 一种无人机载光谱侦检***及其用于危险物侦检作业的控制方法
CN110927142A (zh) * 2019-12-12 2020-03-27 华侨大学 一种基于表面增强拉曼技术的便携式地沟油快速检测仪
CN111208112A (zh) * 2020-01-14 2020-05-29 山西省食品药品检验所(山西省药品包装材料监测中心) 一种食品和药品中吡格列酮和罗格列酮的定性检测方法
CN111239054A (zh) * 2018-11-28 2020-06-05 中移物联网有限公司 一种光谱分析模型应用方法和装置
CN111458309A (zh) * 2020-05-28 2020-07-28 上海海关动植物与食品检验检疫技术中心 一种基于近红外-拉曼联用的植物油定性方法
CN112763477A (zh) * 2020-12-30 2021-05-07 山东省食品药品检验研究院 一种基于拉曼光谱的仿制药质量快速评价***
CN112834481A (zh) * 2020-12-31 2021-05-25 宁波海关技术中心 一种拉曼光谱增强测量***及测量方法
CN112986210A (zh) * 2021-02-10 2021-06-18 四川大学 一种尺度自适应的微生物拉曼光谱检测方法及***
CN113466206A (zh) * 2021-06-23 2021-10-01 上海仪电（集团）有限公司中央研究院 一种基于大数据的拉曼光谱分析***
CN113655050A (zh) * 2021-08-17 2021-11-16 南京富岛信息工程有限公司 一种改进轻质油中痕量原油拉曼光谱检测限的方法
CN114088646A (zh) * 2021-11-17 2022-02-25 汕头海关技术中心 一种化妆品非法添加物快速鉴别方法
CN114184599A (zh) * 2021-12-30 2022-03-15 广州医科大学 单细胞拉曼光谱采集数目估计方法、数据处理方法及装置
CN114486774A (zh) * 2021-12-31 2022-05-13 中科谱光(郑州)应用科学技术研究院有限公司 一种基于高光谱大数据的人工智能算法匹配方法
CN114674352A (zh) * 2022-03-31 2022-06-28 天津大学 基于瑞利散射光谱非相似性的分布式扰动传感和解调方法
CN115931828A (zh) * 2023-02-17 2023-04-07 华谱智能科技(天津)有限公司 适应复杂土壤基体的成分分析与预测方法、单元及***
CN117405650A (zh) * 2023-12-14 2024-01-16 奥谱天成（厦门）光电有限公司 不可降解物质的检测方法、介质
CN117907309A (zh) * 2024-03-19 2024-04-19 夏芮智能科技有限公司 一种基于拉曼光谱的食品和药品安全检测***

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105987753A (zh) * 2015-02-11 2016-10-05 河北伊诺光学科技有限公司 一种基于云计算的光谱专家***及其使用方法
DE102015203537B3 (de) * 2015-02-27 2016-06-23 Celltool Gmbh Vorrichtung und Verfahren zur Überprüfung eines Materials für eine Transplantation
US9933375B2 (en) * 2015-09-25 2018-04-03 Olympus Scientific Solutions Americas, Inc. XRF/XRD system with dynamic management of multiple data processing units
CN106290941B (zh) * 2016-08-31 2018-02-13 合肥领谱科技有限公司 基于云计算与云存储的毒品及易制毒化学品检测管理***及方法
CN108007913B (zh) * 2016-10-27 2020-08-14 中国人民解放军第二军医大学 光谱处理装置、方法以及药品真伪判定***
CN107995237B (zh) * 2016-10-27 2021-12-10 上海迪亚凯特生物医药科技有限公司 光谱数据兼容方法及***
CN108240978B (zh) * 2016-12-26 2020-01-21 同方威视技术股份有限公司 基于拉曼光谱的自学习式定性分析方法
CN108241846B (zh) * 2016-12-26 2021-03-05 同方威视技术股份有限公司 用于识别拉曼谱图的方法
CN108254351B (zh) * 2016-12-29 2023-08-01 同方威视技术股份有限公司 用于物品查验的拉曼光谱检测方法
CN107037028A (zh) * 2017-03-10 2017-08-11 北京华泰诺安探测技术有限公司 一种云平台拉曼光谱识别方法及装置
CN107167463A (zh) * 2017-04-29 2017-09-15 合肥国轩高科动力能源有限公司 一种锂离子电池中涂胶隔膜材料的定性及均一性分析方法
WO2019127352A1 (zh) * 2017-12-29 2019-07-04 深圳达闼科技控股有限公司 基于拉曼光谱的物质识别方法及云端***
CN108051425B (zh) * 2018-01-10 2020-08-21 南京简智仪器设备有限公司 一种拉曼光谱信噪比评估方法
CN108388634A (zh) * 2018-02-24 2018-08-10 颜召臣 利用多维度数据分析古董年代的***及方法
WO2019183882A1 (zh) * 2018-03-29 2019-10-03 深圳达闼科技控股有限公司 物质检测方法、装置、电子设备及计算机可读存储介质
CN109073537A (zh) * 2018-07-16 2018-12-21 深圳达闼科技控股有限公司 一种物质检测的方法、装置、终端和可读存储介质
KR20210078531A (ko) * 2018-10-23 2021-06-28 암젠 인크 실시간 예측을 위한 라만 분광 모델의 자동 교정 및 자동 유지 관리
CN109827943A (zh) * 2019-02-27 2019-05-31 山东省食品药品检验研究院 一种佐匹克隆片的鉴别方法
CN109975211A (zh) * 2019-04-28 2019-07-05 重庆冠雁科技有限公司 基于物联网的拉曼光谱物质监测***及监测方法
CN112686768A (zh) * 2020-12-30 2021-04-20 山东省食品药品检验研究院 一种药品制剂企业物料快速识别***
CN113378680B (zh) * 2021-06-01 2022-06-28 厦门大学 一种拉曼光谱数据的智能建库方法
CN114858779B (zh) * 2022-05-30 2024-03-12 南通朗地罗拉安全设备有限公司 一种智能化气体检测方法及装置
CN116933056B (zh) * 2023-07-24 2024-06-21 哈尔滨工业大学 无扣除拉曼背景确定拉曼光谱特征峰峰面积的方法及***

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1992939A1 (en) * 2007-05-16 2008-11-19 National University of Ireland, Galway A kernel-based method and apparatus for classifying materials or chemicals and for quantifying the properties of materials or chemicals in mixtures using spectroscopic data.
CN101532954A (zh) * 2008-03-13 2009-09-16 天津天士力现代中药资源有限公司 一种用红外光谱结合聚类分析鉴定中药材的方法
CN102023137A (zh) * 2009-09-18 2011-04-20 贵州仁怀茅台镇金士酒业有限公司 一种白酒鉴别方法
CN102590211A (zh) * 2011-01-11 2012-07-18 郑州大学 利用光谱和图像特征进行烟叶分级的方法
WO2012120775A1 (ja) * 2011-03-04 2012-09-13 パナソニック株式会社 結晶性評価方法、結晶性評価装置、及びそのコンピュータソフト
CN102982403A (zh) * 2012-10-31 2013-03-20 北京农业智能装备技术研究中心 基于rfid的果品光谱测试信息管理***
CN103411974A (zh) * 2013-07-10 2013-11-27 杭州赤霄科技有限公司 基于云端大数据的平面材料检测远程***及检测方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2003233585A1 (en) * 2002-05-20 2003-12-12 Rosetta Inpharmatics Llc Computer systems and methods for subdividing a complex disease into component diseases
ATE502343T1 (de) * 2002-06-14 2011-04-15 Pfizer Metabolische phänotypisierung
US7321791B2 (en) * 2003-09-23 2008-01-22 Cambridge Research And Instrumentation, Inc. Spectral imaging of deep tissue

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1992939A1 (en) * 2007-05-16 2008-11-19 National University of Ireland, Galway A kernel-based method and apparatus for classifying materials or chemicals and for quantifying the properties of materials or chemicals in mixtures using spectroscopic data.
CN101532954A (zh) * 2008-03-13 2009-09-16 天津天士力现代中药资源有限公司 一种用红外光谱结合聚类分析鉴定中药材的方法
CN102023137A (zh) * 2009-09-18 2011-04-20 贵州仁怀茅台镇金士酒业有限公司 一种白酒鉴别方法
CN102590211A (zh) * 2011-01-11 2012-07-18 郑州大学 利用光谱和图像特征进行烟叶分级的方法
WO2012120775A1 (ja) * 2011-03-04 2012-09-13 パナソニック株式会社 結晶性評価方法、結晶性評価装置、及びそのコンピュータソフト
CN102982403A (zh) * 2012-10-31 2013-03-20 北京农业智能装备技术研究中心 基于rfid的果品光谱测试信息管理***
CN103411974A (zh) * 2013-07-10 2013-11-27 杭州赤霄科技有限公司 基于云端大数据的平面材料检测远程***及检测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIU, YAN ET AL.: "Progress of Raman Spectroscopy in Fast Detection of Counterfeit Drugs", JOURNAL OF PHARMACEUTICAL PRACTICE, vol. 30, no. 6, 25 November 2012 (2012-11-25), pages 401 - 404 and 446, XP055234007 *
YUAN, YUFENG ET AL.: "Application of Chemometrics Combined with Raman Spectroscopy to the Analysis of Biomaterials", CHINESE JOURNAL OF SPECTROSCOPY LABORATORY, vol. 27, no. 6, 30 November 2010 (2010-11-30), pages 2442 - 2448 *

Cited By (28)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107238595B (zh) * 2017-05-05 2023-08-04 浙江大学 封闭容器的酒精浓度测量装置及测量方法
CN107238595A (zh) * 2017-05-05 2017-10-10 浙江大学 封闭容器的酒精浓度测量装置及测量方法
CN107340547A (zh) * 2017-07-24 2017-11-10 山东省职业卫生与职业病防治研究院 一种无人机载光谱侦检***及其用于危险物侦检作业的控制方法
CN111239054A (zh) * 2018-11-28 2020-06-05 中移物联网有限公司 一种光谱分析模型应用方法和装置
CN110927142A (zh) * 2019-12-12 2020-03-27 华侨大学 一种基于表面增强拉曼技术的便携式地沟油快速检测仪
CN111208112A (zh) * 2020-01-14 2020-05-29 山西省食品药品检验所(山西省药品包装材料监测中心) 一种食品和药品中吡格列酮和罗格列酮的定性检测方法
CN111458309A (zh) * 2020-05-28 2020-07-28 上海海关动植物与食品检验检疫技术中心 一种基于近红外-拉曼联用的植物油定性方法
CN111458309B (zh) * 2020-05-28 2023-07-07 上海海关动植物与食品检验检疫技术中心 一种基于近红外-拉曼联用的植物油定性方法
CN112763477A (zh) * 2020-12-30 2021-05-07 山东省食品药品检验研究院 一种基于拉曼光谱的仿制药质量快速评价***
CN112834481B (zh) * 2020-12-31 2023-09-05 宁波海关技术中心 一种拉曼光谱增强测量***及测量方法
CN112834481A (zh) * 2020-12-31 2021-05-25 宁波海关技术中心 一种拉曼光谱增强测量***及测量方法
CN112986210A (zh) * 2021-02-10 2021-06-18 四川大学 一种尺度自适应的微生物拉曼光谱检测方法及***
CN113466206A (zh) * 2021-06-23 2021-10-01 上海仪电（集团）有限公司中央研究院 一种基于大数据的拉曼光谱分析***
CN113655050A (zh) * 2021-08-17 2021-11-16 南京富岛信息工程有限公司 一种改进轻质油中痕量原油拉曼光谱检测限的方法
CN113655050B (zh) * 2021-08-17 2024-04-26 南京富岛信息工程有限公司 一种改进轻质油中痕量原油拉曼光谱检测限的方法
CN114088646A (zh) * 2021-11-17 2022-02-25 汕头海关技术中心 一种化妆品非法添加物快速鉴别方法
CN114088646B (zh) * 2021-11-17 2024-04-12 汕头海关技术中心 一种化妆品非法添加物快速鉴别方法
CN114184599B (zh) * 2021-12-30 2024-04-26 广州医科大学 单细胞拉曼光谱采集数目估计方法、数据处理方法及装置
CN114184599A (zh) * 2021-12-30 2022-03-15 广州医科大学 单细胞拉曼光谱采集数目估计方法、数据处理方法及装置
CN114486774A (zh) * 2021-12-31 2022-05-13 中科谱光(郑州)应用科学技术研究院有限公司 一种基于高光谱大数据的人工智能算法匹配方法
CN114674352A (zh) * 2022-03-31 2022-06-28 天津大学 基于瑞利散射光谱非相似性的分布式扰动传感和解调方法
CN114674352B (zh) * 2022-03-31 2023-09-15 天津大学 基于瑞利散射光谱非相似性的分布式扰动传感和解调方法
CN115931828B (zh) * 2023-02-17 2023-06-16 华谱智能科技(天津)有限公司 适应复杂土壤基体的成分分析与预测方法、单元及***
CN115931828A (zh) * 2023-02-17 2023-04-07 华谱智能科技(天津)有限公司 适应复杂土壤基体的成分分析与预测方法、单元及***
CN117405650A (zh) * 2023-12-14 2024-01-16 奥谱天成（厦门）光电有限公司 不可降解物质的检测方法、介质
CN117405650B (zh) * 2023-12-14 2024-03-12 奥谱天成（厦门）光电有限公司 不可降解物质的检测方法、介质
CN117907309A (zh) * 2024-03-19 2024-04-19 夏芮智能科技有限公司 一种基于拉曼光谱的食品和药品安全检测***
CN117907309B (zh) * 2024-03-19 2024-06-04 夏芮智能科技有限公司 一种基于拉曼光谱的食品和药品安全检测***

Also Published As

Publication number Publication date
CN104215623A (zh) 2014-12-17
CN104215623B (zh) 2018-09-25

Similar Documents

Publication Publication Date Title
WO2015165394A1 (zh) 面向多行业检测的激光拉曼光谱智能化辨识方法及***
Kiani et al. Integration of computer vision and electronic nose as non-destructive systems for saffron adulteration detection
Ropodi et al. Multispectral imaging (MSI): A promising method for the detection of minced beef adulteration with horsemeat
Li et al. Potential of smartphone-coupled micro NIR spectroscopy for quality control of green tea
Clarke Extracting process-related information from pharmaceutical dosage forms using near infrared microscopy
Li et al. Evaluation of black tea by using smartphone imaging coupled with micro-near-infrared spectrometer
Pereira et al. Evaluation and identification of blood stains with handheld NIR spectrometer
Dong et al. Deep learning for geographical discrimination of Panax notoginseng with directly near-infrared spectra image
Sereshti et al. Quality control of saffron and evaluation of potential adulteration by means of thin layer chromatography-image analysis and chemometrics methods
Anzanello et al. A multivariate-based wavenumber selection method for classifying medicines into authentic or counterfeit classes
Deconinck et al. Chemometrics and chromatographic fingerprints to discriminate and classify counterfeit medicines containing PDE-5 inhibitors
CN103411906A (zh) 珍珠粉和贝壳粉的近红外光谱定性鉴别方法
Dong et al. 2DCOS combined with CNN and blockchain to trace the species of boletes
Zhang et al. Rapid identification of the storage age of dried tangerine peel using a hand-held near infrared spectrometer and machine learning
Trentanni Hansen et al. NIR-based Sudan I to IV and Para-Red food adulterants screening
Wang et al. Classification modeling method for hyperspectral stamp‐pad ink data based on one‐dimensional convolutional neural network
Yu et al. Prediction of enological parameters and discrimination of rice wine age using least-squares support vector machines and near infrared spectroscopy
Si et al. Nondestructive qualitative and quantitative analysis of Yaobitong capsule using near-infrared spectroscopy in tandem with chemometrics
Custers et al. Chemometrical analysis of fingerprints for the detection of counterfeit and falsified medicines
Teng et al. Predictive data clustering of laser-induced breakdown spectroscopy for brain tumor analysis
Li et al. Discrimination of raw and sulfur-fumigated ginseng based on Fourier transform infrared spectroscopy coupled with chemometrics
Kamarudin et al. Tongue's substance and coating recognition analysis using HSV color threshold in tongue diagnosis
Hu et al. Intelligent spectral algorithm for pigments visualization, classification and identification based on Raman spectra
Ren et al. Digital depiction of the quality of Dianhong black tea based on pocket-sized near infrared spectroscopy
Liu et al. Recent advancement of chemical imaging in pharmaceutical quality control: From final product testing to industrial utilization

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15786451

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 15786451

Country of ref document: EP

Kind code of ref document: A1