WO2019195971A1 - 光谱分析方法、装置、电子设备及计算机可读存储介质 - Google Patents

光谱分析方法、装置、电子设备及计算机可读存储介质 Download PDF

Info

Publication number
WO2019195971A1
WO2019195971A1 PCT/CN2018/082239 CN2018082239W WO2019195971A1 WO 2019195971 A1 WO2019195971 A1 WO 2019195971A1 CN 2018082239 W CN2018082239 W CN 2018082239W WO 2019195971 A1 WO2019195971 A1 WO 2019195971A1
Authority
WO
WIPO (PCT)
Prior art keywords
spectral
component
spectral analysis
proportion
measured object
Prior art date
Application number
PCT/CN2018/082239
Other languages
English (en)
French (fr)
Inventor
牟涛涛
骆磊
Original Assignee
深圳达闼科技控股有限公司
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 深圳达闼科技控股有限公司 filed Critical 深圳达闼科技控股有限公司
Priority to CN201880001146.4A priority Critical patent/CN108780037A/zh
Priority to PCT/CN2018/082239 priority patent/WO2019195971A1/zh
Publication of WO2019195971A1 publication Critical patent/WO2019195971A1/zh

Links

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
    • 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
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28—Investigating the spectrum
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28—Investigating the spectrum
    • G01J3/44—Raman spectrometry; Scattering spectrometry ; Fluorescence spectrometry
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2201/00—Features of devices classified in G01N21/00
    • G01N2201/12—Circuits of general importance; Signal processing
    • G01N2201/129—Using chemometrical methods
    • G01N2201/1293—Using chemometrical methods resolving multicomponent spectra

Definitions

  • the present application relates to the field of spectral measurement technologies, and in particular, to a spectral analysis method, apparatus, electronic device, and computer readable storage medium.
  • Spectral analysis refers to a method of identifying a substance based on its spectrum and determining its chemical composition and relative content to obtain the molecular structure of the substance.
  • the inventors have found that at least the following problems exist in the prior art: the existing spectral analysis method can only perform single substance spectral recognition or mixture analysis on the measured object, and when analyzing the measured object composed of a plurality of substances, It is impossible to determine the specific substances contained in it and the proportion of each substance.
  • the present application provides a spectral analysis method, apparatus, electronic device, and computer readable storage medium to solve the above technical problems.
  • An embodiment of the present application provides a spectral analysis method, including: acquiring a spectrum of a measured object; performing spectral analysis on a spectral image of the measured object according to a previously obtained spectral analysis model, and determining the measured The composition of the object and the proportion of each component; wherein the spectral analysis model is obtained by training a large amount of spectral data in the spectral sample data set; the spectral analysis model is used to define a mapping relationship between the spectral image and the composition and the proportion of each component.
  • An embodiment of the present application provides a spectral analysis configuration including: an acquisition module and a determination module; an acquisition module for acquiring a spectral image of the measured object; and a determination module for performing spectral analysis according to a pre-acquisition
  • the model performs spectral analysis on the spectrum of the measured object to determine the composition of the measured object and the proportion of each component; wherein the spectral analysis model is obtained by training a large amount of spectral data in the spectral sample data set; the spectral analysis model is used for Define the mapping relationship between the spectrum and the composition and the proportion of each component.
  • An embodiment of the present application provides an electronic device including at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being At least one processor is executed to enable at least one processor to perform the spectral analysis methods involved in any of the method embodiments of the present application.
  • One embodiment of the present application provides a computer readable storage medium storing computer instructions for causing a computer to perform a spectral analysis method as referred to in any of the method embodiments of the present application.
  • the embodiment of the present application obtains a spectral analysis model for defining a mapping relationship between a spectral image and a component and a proportion of each component by training a large amount of spectral data in the spectral sample data set, and needs to be analyzed and measured.
  • the spectroscopic analysis model is used to perform spectral analysis on the spectrogram of the object to be measured, so that the ratio of various components and components contained in the object to be tested can be quickly analyzed.
  • the spectral analysis model is obtained by training a large amount of spectral data, and adding a predetermined offset to the training data during the training process, the influence of the spectral shift on the analysis result can be effectively reduced, and the result is accurate.
  • Sexuality in addition to sharing the database between different resolutions and spectral ranges to achieve sharing between databases.
  • FIG. 1 is a flow chart of a method of spectral analysis in a first embodiment of the present application
  • FIG. 2 is a schematic diagram of training of a spectral analysis model in the first embodiment of the present application
  • FIG. 3 is a flow chart of a method of spectral analysis in a second embodiment of the present application.
  • FIG. 4 is a flow chart of a spectral analysis method in a third embodiment of the present application.
  • Figure 5 is a block diagram showing a spectrum analyzing apparatus in a fourth embodiment of the present application.
  • FIG. 6 is a block schematic diagram of an electronic device in a fifth embodiment of the present application.
  • the first embodiment of the present application relates to a spectral analysis method, and the specific flow is shown in FIG.
  • step 101 a spectrogram of the measured object is acquired.
  • step 102 spectral analysis of the spectrum of the measured object is performed according to the spectral analysis model.
  • step 103 the composition of the analyte and the proportion of each component are determined.
  • the method for determining the composition of the measured object and the proportion of each component in the embodiment may be: inputting the spectral data in the spectral image of the measured object into a spectral analysis model, and then obtaining the output of the spectral analysis model.
  • the result of the spectral analysis is obtained by extracting the components contained in the spectral analysis results and the proportion of each component, and determining the ratio of the extracted components and the components to the components of the analyte and the proportion of each component.
  • the spectral analysis model is obtained by training a large amount of spectral data in the spectral sample data set.
  • the spectral analysis model obtained in the training in this embodiment is mainly used to define a mapping relationship between the spectral image and the composition and the proportion of each component, that is, after inputting the spectral data in the spectral image to be analyzed into the spectral analysis model, the spectrum
  • the analysis model outputs the components corresponding to the spectrum and the proportion of each component by analysis.
  • the spectral data included in the spectral sample data set may specifically be spectral data of a single pure substance, or spectral data of a single mixture, or both.
  • the spectral analysis model is only to determine the composition of the pure substance for analysis, then only the spectral data set of the pure component of the known component needs to be constructed, and then the spectral data in the spectral data set of the pure object is trained based on the deep learning algorithm to obtain A spectral analysis model for analyzing the composition of pure matter.
  • the spectral analysis model is only for the purpose of analyzing and determining the composition of the mixture and the proportion of each component, it is only necessary to construct a spectral data set of a mixture of known components and components, and then focus the spectral data of the mixture based on the depth learning algorithm.
  • the spectral data is trained to obtain a spectral analysis model.
  • the spectral analysis model can analyze and determine the composition of the pure substance, and analyze and determine the composition of the mixture and the proportion of each component, it is necessary to separately train the spectral data of the above two data sets to obtain a purity that can be analyzed and determined.
  • the composition of the material can be analyzed to determine the composition of the mixture and the spectral analysis model of the proportion of each component.
  • training the spectral data of the spectral data set of the pure substance to obtain a spectral analysis model for analyzing the components of the pure substance may specifically include the following steps :
  • the spectral data of the pure substance of the known component (which may be the spectral data of the pure component of the known component stored in the historical database), and the random noise data and/or the horizontal ordinate offset data are added to the spectral data of the pure object.
  • a spectral data set of a pure substance of a known composition can be constructed (can be rationally added by a person skilled in the art according to the actual situation).
  • the spectral data in the spectral data set of the pure object is repeatedly trained until the spectral analysis model is obtained.
  • the spectral data of the spectral data set of the mixture is trained based on a deep learning algorithm, and specifically includes the following steps:
  • spectral data of known different substances are randomly added to the spectral data sets of the pure components of the known components to be constructed in different proportions, and a spectral data set of a mixture of known components and proportions of the components is constructed.
  • the spectral data in the spectral data set of the mixture is iteratively trained until a spectral analysis model is obtained.
  • the above-mentioned spectral data in the spectral data set is repeatedly trained until a spectral analysis model is obtained, specifically, by repeated training on a large amount of spectral data, such as for pure substance A (component B), when passing As long as you input the spectrum of pure A, add different noise and offset, or even different spectral resolution and range, you can output B, you can think that the training is completed.
  • a spectral analysis model such as for pure substance A (component B)
  • the above operation of acquiring the spectral analysis model is performed before performing spectral analysis on the spectrum of the measured object to determine the composition of the measured object and the proportion of each component.
  • the spectrum analysis model obtained by the training may be directly saved on the spectrum analysis device, may be saved in the remote server, or both may be saved, and the synchronization is periodically updated according to a certain synchronization mechanism.
  • the specific implementation manner is known to those skilled in the art. It can be set as needed, and there is no limit here.
  • the training of the spectral analysis model shown in FIG. 2 is that the spectral data stored in the database ranges from 200 cm -1 to 3000 cm -1 , the spectral data is 1400 points, and each dot is separated by 2 cm -1 .
  • the spectral data of 10,000 kinds of pure materials is trained.
  • L1 to L9 in FIG. 2 are an input layer (ie, an input layer), a first fully connected layer (ie, a Dense Layer), a first batch layer (ie, a Batch Normalization Laye), and a second fully connected layer ( Dense Layer), Dropout Layer, and second batch layer (Batch) Normalization Laye), the third fully connected layer (Dense Layer), the filter layer (that is, the Threshold Filter layer), and the output layer.
  • the input 1400 ⁇ 1 tensor is transmitted to L2 after passing through L1.
  • the input 1400 ⁇ 1 tensor is compiled based on the ReLU (Rectified Linear Unit) activation function to determine whether it is necessary to add a bias. Set the vector.
  • ReLU Rectified Linear Unit
  • activation is an activation function in the above formula
  • kernel is the weight matrix of this layer
  • bias is the offset vector
  • result of output is output.
  • the output is defined as a tensor shape of 3200 ⁇ 1, and the first step is extended.
  • Other parameters are similar to those in L2 and will not be described here.
  • L5 is mainly used to randomly disconnect the input neurons according to the rate (ie, the output tensor part position is set to 0) to prevent over-fitting, and set the proportional parameter setting of this layer here. Is 0.5.
  • the output is defined as a tensor shape of 10000 ⁇ 1, and the second step is extended, and the data is activated by the Sigmoid activation function.
  • Other parameters are similar to those in L2, and are not described here.
  • the data is input to L8, and the components of the input sample are judged according to the output tensor value, and the output of each component larger than the threshold is represented by "1", and the others are represented by "0", and the above The structure is transmitted to L9 and output by L9.
  • the initial sample data that needs to be trained can be obtained, and then the sample data is trained based on the deep learning algorithm, as follows:
  • the Raman spectrum of the substance is composed of a plurality of Gaussian peaks (or Lorentz peaks) of different peak heights, and the peak width depends on the resolving power of the Raman spectrometer, and the peak position depends on the chemical bonds contained in the substance molecules.
  • the Raman spectrum of different substances mainly reflects the difference of peak position and peak height ratio.
  • the Raman spectrum of the mixture can be regarded as a linear stack of Raman spectra of the pure matter of the mixture.
  • the pure Raman spectroscopy data of the above 10,000 kinds of materials are divided into four groups of A, B, C and D (in practical applications, the data set can be reasonably divided according to the amount of spectral data. Training, no restrictions here), and generate the data sets participating in the training according to the following combination:
  • Training data of two substances (15000 kinds): A set and B set, one-to-one correspondence, using random concentration mixing to obtain 2500 kinds of mixture; A set and C set, one-to-one correspondence, using random concentration mixing, get 2500 kinds mixture; A set and D set, one-to-one correspondence, using random concentration mixing, to obtain 2500 kinds of mixture; B set and C set, one-to-one correspondence, using random concentration mixing to obtain 2500 kinds of mixture; B set and D set, one-to-one correspondence, Using a random concentration mixture, 2500 kinds of mixtures were obtained; C sets were combined with D sets, one-to-one, and mixed at random concentrations to obtain 2500 kinds of mixtures.
  • Training data for three substances (a total of 10,000): A+B+C, one-to-one correspondence, using random concentration mixing, to obtain 2500 kinds of mixture; B+C+D, one-to-one correspondence, using random concentration mixing to obtain 2500 kinds of mixture; A+B+D, one-to-one correspondence, Mixing at random concentrations gave 2500 mixtures; A+C+D, one-to-one correspondence, mixed at random concentrations to give 2500 mixtures.
  • Training data of four substances (a total of 10,000 kinds): A+B+C+D, one-to-one correspondence, using random concentration mixing to obtain 2500 kinds of mixtures; from all the materials, 4 kinds of substances were randomly selected and mixed at random concentration. This was carried out 7500 times (in practical applications, the mixing can be set here as needed by a person skilled in the art, and is not limited here), and 7,500 kinds of mixtures are obtained.
  • the generated 45000 sets of data are divided into a batch (sample size used in one iteration) per 1000 pieces of data as input data, and then a certain number of trainings are performed based on the deep learning algorithm until loss (training loss value) and The val_loss (average training loss value) is stable and reduced to below 10 -6 , at which point the spectral analysis model is considered available.
  • the spectral analysis method obtaineds a spectral analysis model for defining a mapping relationship between a spectral image and a component and a proportion of each component by training a large amount of spectral data in the spectral sample data set.
  • the spectral analysis model is used to perform spectral analysis on the spectrum of the measured object, thereby being able to quickly analyze the proportion of various components and components contained in the measured object. .
  • the spectral analysis model is obtained by training a large amount of spectral data, and adding a predetermined offset to the training data during the training process, the influence of the spectral shift on the analysis result can be effectively reduced, and the result is accurate.
  • Sexuality in addition to sharing the database between different resolutions and spectral ranges to achieve sharing between databases.
  • a second embodiment of the present application relates to a method of spectral analysis. This embodiment is further improved on the basis of the first embodiment.
  • the specific improvement is: when determining the composition of the measured object and the proportion of each component, by determining whether the spectral analysis result carries the label with successful analysis. And according to the judgment result, the corresponding operation, the specific process is shown as in Fig. 3.
  • the steps 301 to 306 are included, wherein the steps 301, 302, and 306 are substantially the same as the steps 101 to 103 in the first embodiment, and are not described herein again.
  • the steps 301, 302, and 306 are substantially the same as the steps 101 to 103 in the first embodiment, and are not described herein again.
  • the touch spectrum analysis method provided in the first embodiment, and details are not described herein again.
  • step 303 it is determined whether the spectrum analysis result carries a label with a successful analysis.
  • step 306 if the spectrum analysis result carries the label with successful analysis, the process proceeds to step 306; otherwise, the process proceeds to step 304.
  • the label with the analysis success is carried, and when the step 306 is performed to determine the composition of the measured object and the proportion of each component, specifically, the component and each component extracted from the spectral analysis result are obtained.
  • the proportion of the measured object is determined as the composition of the measured object and the proportion of each component.
  • the label used to indicate whether the spectral analysis operation is successful may be an attribute set separately, such as carrying “success” or “failure” in the output spectral analysis result after the analysis, or The content of "YES” or "NO” to determine if the spectral analysis operation was successful.
  • the field of the success or failure of the identification may not be separately set, and the output of the material that successfully identifies is directly set to “1”, and the output of the unidentified material is “0”, so that the spectrum of the measured object is based on the spectral analysis model.
  • the spectral analysis results of the output are all "1"
  • it is determined that the spectral analysis operation is successful otherwise it fails.
  • step 304 a new spectral analysis model is acquired.
  • the acquired new spectral analysis model described in this embodiment is based on a deep learning algorithm, and spectral data of pure components of other known components and components taken from the network, and/or The spectral data of the known component and the mixture of the components are trained and obtained.
  • step 305 spectral analysis of the spectrogram of the measured object is performed according to the new spectral analysis model.
  • the spectral analysis method provided in this embodiment is based on deep learning when the composition of the measured object and the proportion of each component cannot be determined by using the existing spectral analysis model.
  • the algorithm is trained by the spectroscopic analysis device or the remote server by spectral data of a large number of known components in the network and the purity of each component, and/or spectral data of a mixture of known components and components.
  • a new spectral analysis model is obtained to analyze the spectrum of the measured object, further enhancing the user experience.
  • the spectral analysis operation can ensure the accuracy and timeliness of the analysis results as much as possible, and can greatly enhance the user experience.
  • the spectral data in the network can be trained based on the deep learning algorithm to obtain a new spectral analysis model, and the acquired spectral data is obtained based on the deep learning algorithm. Before training, you need to make sure you have access to the network.
  • the existing chemometric method is used for analysis when the spectral analysis results carry a label with successful analysis.
  • a third embodiment of the present application relates to a method of spectral analysis. This embodiment is further improved on the basis of the first or second embodiment, and the specific improvement is: after determining the composition of the measured object and the proportion of each component, the components of the measured object and the components are preserved. The ratio and the environmental information of the measured object.
  • the following description will be specifically made on the improvement made in the first embodiment, and the specific process is as shown in FIG. 4.
  • the steps 401 to 404 are included, wherein the steps 401 to 403 are substantially the same as the steps 101 to 103 in the first embodiment, and details are not described herein.
  • the steps 401 to 403 are substantially the same as the steps 101 to 103 in the first embodiment, and details are not described herein.
  • the technical details that are not described in detail in this embodiment refer to the touch spectrum analysis method provided in the first embodiment, and details are not described herein again.
  • step 404 the components of the test object, the proportion of each component, and the environmental information of the object to be tested are stored.
  • the spectral analysis method determines the composition of the measured object, the proportion of each component, and the ratio of the components of the measured object after determining the composition of the measured object and the proportion of each component.
  • the environmental information of the measured object obtained by the spectral analysis device is saved (for example, stored in a memory inside the spectral analysis device or uploaded to a cloud server), thereby facilitating the tester to complete the spectral analysis of the measured object. After that, it is possible to perform subsequent operations based on the analyzed data, such as updating data in the spectral sample data set.
  • the environmental information of the measured object in the embodiment may specifically include GPS (Global Positioning System) information of the position of the measured object, and image information of the surrounding environment is other Relevant information to facilitate management or management of the spectrum analysis equipment and operations.
  • GPS Global Positioning System
  • obtaining and preserving the environmental information of the object to be tested also facilitates screening for dangerous events around the object to be tested, and then promptly notifies the operator to respond to avoid injury to the operator and equipment.
  • the result may be displayed on the display interface of the spectrum analysis device or transmitted to other devices for subsequent processing.
  • a fourth embodiment of the present application relates to a spectral analysis apparatus, the specific structure of which is shown in FIG.
  • the spectrum analysis device mainly includes an acquisition module 501 and a determination module 502.
  • the obtaining module 501 is configured to acquire a spectrum of the measured object.
  • the determining module 502 is configured to perform spectral analysis on the spectrum of the measured object according to the spectral analysis model obtained in advance, and determine the composition of the measured object and the proportion of each component.
  • the spectral analysis model is obtained by training a large amount of spectral data in the spectral sample data set.
  • the spectral analysis model obtained by the training in this embodiment is mainly used to define a mapping relationship between the spectrum and the components and the proportion of each component.
  • the spectral analysis apparatus obtains a spectral analysis model for defining a mapping relationship between a spectral image and a component and a proportion of each component by training a large amount of spectral data in the spectral sample data set.
  • the spectral analysis model is used to perform spectral analysis on the spectrum of the measured object, thereby being able to quickly analyze the proportion of various components and components contained in the measured object.
  • the spectral analysis model is obtained by training a large amount of spectral data, the influence of spectral shift on the analysis result can be effectively reduced, and the accuracy of the result is ensured.
  • a fifth embodiment of the present application relates to an electronic device, and the specific structure is as shown in FIG. 6.
  • the electronic device in this embodiment may be a spectrum analyzer, such as a portable spectrum analyzer similar to a notebook computer, a small handheld spectrum analyzer, or a smart terminal device equipped with a spectrum analysis application, such as a mobile phone. Tablet PCs, etc., are not mentioned here, and are not specifically limited.
  • the electronic device may specifically include one or more processors 601 and a memory 602 and a communication component 603 communicatively coupled to the at least one processor 601.
  • processors 601 are exemplified in FIG.
  • each functional module in the spectrum analysis device involved in the above embodiment is deployed on the processor 601, and the processor 601, the memory 602, and the communication component 603 can be connected to each other through a bus or other manner, FIG. Take the bus connection as an example.
  • the memory 602 as a computer readable storage medium, can be used to store software programs, computer executable programs, and modules, such as program instructions/modules corresponding to the spectral analysis methods involved in any of the method embodiments of the present application.
  • the processor 601 performs the spectral analysis method involved in any of the method embodiments of the present application by running software programs, instructions, and modules stored in the memory 602 and controlling the communication component 603 to receive and/or transmit data.
  • the memory 602 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function; and the storage data area may establish an analysis result database for saving The composition of the measured object, the proportion of each component, and the environmental information of the measured object.
  • the memory 602 may include a high speed random access memory, and may also include a read/write memory (Random Access Memory, RAM) or the like.
  • RAM Random Access Memory
  • memory 602 can optionally include memory remotely located relative to processor 601 that can be connected to the terminal device over a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • At least one processor 601 may be stored in the memory 602, and the instructions are executed by the at least one processor 601, so that the at least one processor 601 can perform the spectral analysis method according to any method embodiment of the present application.
  • the respective operations in the spectral analysis method are completed.
  • the electronic device in this embodiment may also be a cloud intelligent electronic device, that is, an electronic device for performing processing operations.
  • the device's processor is in the cloud.
  • the cloud intelligent electronic device makes the intelligent computing capability of the cloud a convenient service, which greatly reduces the research and development costs and operating costs of the intelligent electronic device, and utilizes the powerful computing power of the cloud to facilitate more convenient training.
  • a spectral analysis model is obtained.
  • a sixth embodiment of the present application is directed to a computer readable storage medium, which is a computer readable storage medium having stored therein computer instructions that enable a computer to perform any of the present application
  • the method of spectral analysis involved in the method examples is directed to a computer readable storage medium, which is a computer readable storage medium having stored therein computer instructions that enable a computer to perform any of the present application The method of spectral analysis involved in the method examples.
  • a program instructing related hardware may be completed by a program instructing related hardware, and the program is stored in a storage medium, and includes a plurality of instructions for making a device (which may be a single chip microcomputer). , a chip, etc. or a processor performs all or part of the steps of the methods described in various embodiments of the present application.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .

Landscapes

  • Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)
  • Spectrometry And Color Measurement (AREA)

Abstract

本申请涉及光谱测量技术领域，公开了一种光谱分析方法、装置、电子设备及计算机可读存储介质。本申请中，光谱分析方法包括：获取被测物的光谱图；根据预先获得的光谱分析模型，对被测物的光谱图进行光谱分析，确定被测物的成分及各成分的占比；其中，光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得；光谱分析模型用于定义光谱图与成分及各成分占比的映射关系。该光谱分析方法，能够分析出被测物中包含的多种成分及各成分的占比，并且有效降低了光谱偏移对分析结果的影响。

Description

光谱分析方法、装置、电子设备及计算机可读存储介质 技术领域
本申请涉及光谱测量技术领域，特别涉及一种光谱分析方法、装置、电子设备及计算机可读存储介质。
 背景技术
光谱分析，是指根据物质的光谱来鉴别物质及确定它的化学组成和相对含量，从而得到物质的分子结构的方法。
当前的光谱分析都是基于查找、寻峰和化学计量等方式，对被测物进行单一物质光谱识别或混合物分析。具体的说，其分析流程如下：
先通过光谱寻峰，找出待分析光谱中所含有的主要峰值（能够体现被测物特征的峰值）位置，然后通过查表的方式，找出可能含有的物质，并根据不合理的峰值位置，剔除其中不可能含有的物质，最后对剩余的物质进行相关系数计算，选取相关系数较大的物质，并将选取出的相关系数较大的物质确定为识别出的物质。
 技术问题
但是，发明人发现现有技术中至少存在如下问题：现有的光谱分析方法只能对被测物进行单一物质光谱识别或混合物分析，在对由多种物质组成的被测物进行分析时，无法确定其包含的具体物质及各物质的占比。
另外，现有的光谱分析方法，其分析结果受光谱偏移影响较大，因此对光谱分析设备的要求较高，并且为了降低光谱偏移对分析结果的影响，需要定期确认光谱分析设备的定标系数，以保证分析输出的数据能够标准化。
而且由于现有的光谱分析设备的分辨率、光谱范围等参数的不同，也会使得不同光谱分析设备之间不能共用同一个数据库。
另外，随着光谱数据库中光谱数据的增加，分析过程中的耗费时间也会线性增加，无疑会增大分析过程的计算量，从而严重影响分析速度。
 技术解决方案
本申请提供了一种光谱分析方法、装置、电子设备及计算机可读存储介质，以解决上述技术问题。
本申请的一个实施例提供了一种光谱分析方法，该光谱分析方法包括：获取被测物的光谱图；根据预先获得的光谱分析模型，对被测物的光谱图进行光谱分析，确定被测物的成分及各成分的占比；其中，光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得；光谱分析模型用于定义光谱图与成分及各成分占比的映射关系。
本申请的一个实施例提供了一种光谱分析配置，该光谱分析配置包括：获取模块和确定模块；获取模块，用于获取被测物的光谱图；确定模块，用于根据预先获得的光谱分析模型，对被测物的光谱图进行光谱分析，确定被测物的成分及各成分的占比；其中，光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得；光谱分析模型用于定义光谱图与成分及各成分占比的映射关系。
本申请的一个实施例提供了一种电子设备，该电子设备包括至少一个处理器；以及与至少一个处理器通信连接的存储器；其中，存储器存储有可被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行本申请任意方法实施例中涉及的光谱分析方法。
本申请的一个实施例提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机指令，计算机指令用于使计算机执行本申请任意方法实施例中涉及的光谱分析方法。
 有益效果
本申请实施例相对于现有技术而言，通过对光谱样本数据集中的大量光谱数据进行训练获得用于定义光谱图与成分及各成分占比的映射关系的光谱分析模型，在需要分析被测物的成分及成分占比时，利用该光谱分析模型对被测物的光谱图进行光谱分析，从而能够快速分析出被测物中包含的多种成分及各成分的占比。另外，由于光谱分析模型是对大量光谱数据进行训练获得的，并且在训练的过程中向训练数据添加了预定的偏移，因而可以有效降低光谱偏移对分析结果的影响，保证了结果的准确性，另外通过训练不同分辨率和光谱范围的数据库实现了数据库之间的共用。
 附图说明
一个或多个实施例通过与之对应的附图中的图片进行示例性说明，这些示例性说明并不构成对实施例的限定，附图中具有相同参考数字标号的元件表示为类似的元件，除非有特别申明，附图中的图不构成比例限制。
图1是本申请第一实施例中光谱分析方法的流程图；
图2是本申请第一实施例中光谱分析模型的训练的示意图；
图3是本申请第二实施例中光谱分析方法的流程图；
图4是本申请第三实施例中光谱分析方法的流程图；
图5是本申请第四实施例中光谱分析装置的方框示意图；
图6是本申请第五实施例中电子设备的方框示意图。
 本发明的实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请部分实施例进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请的第一实施例涉及一种光谱分析方法，具体流程如图1所示。
在步骤101中，获取被测物的光谱图。
在步骤102中，根据光谱分析模型，对被测物的光谱图进行光谱分析。
在步骤103中，确定被测物的成分及各成分的占比。
具体的说，在本实施例中确定被测物的成分及各成分的占比的方式具体可以是：将被测物的光谱图中的光谱数据输入光谱分析模型，然后获取光谱分析模型输出的光谱分析结果，并从光谱分析结果中提取包含的成分及各成分的占比，将提取到的成分及各成分的占比，确定为被测物的成分及各成分的占比。
需要说明的是，在本实施例中，光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得。
另外，本实施例中训练获得的光谱分析模型主要是用于定义光谱图与成分及各成分占比的映射关系，即在将需要分析的光谱图中的光谱数据输入至光谱分析模型后，光谱分析模型通过分析会输出与该光谱图对应的成分及各成分的占比。
另外，在本实施例中，光谱样本数据集中包括的光谱数据具体可以是单独的纯净物的光谱数据，或者单独混合物的光谱数据，或者两则均包括。
为了便于理解，以下对光谱分析模型的获取进行具体说明：
若光谱分析模型仅仅是为分析确定纯净物的成分，则仅需构建已知成分的纯净物的光谱数据集，然后基于深度学习算法，对纯净物的光谱数据集中的光谱数据进行训练，获得用于分析纯净物的成分的光谱分析模型。
若光谱分析模型仅仅是为了分析确定混合物的成分及各成分的占比，则仅需构建已知成分及各成分占比的混合物的光谱数据集，然后基于深度学习算法，对混合物的光谱数据集中的光谱数据进行训练，获得光谱分析模型。
若光谱分析模型既可以分析确定纯净物的成分，又可以分析确定混合物的成分及各成分的占比，则需要分别针对上述两种数据集中的光谱数据进行训练，从而得到一个既可以分析确定纯净物成分，又可以分析确定混合物成分及各成分占比的光谱分析模型。
关于构建已知成分的纯净物的光谱数据集，基于深度学习算法，对纯净物的光谱数据集中的光谱数据进行训练，获得用于分析纯净物的成分的光谱分析模型的操作具体可以包括以下步骤：
首先，获取已知成分的纯净物的光谱数据（可以是历史数据库中存储的已知成分的纯净物的光谱数据），向纯净物的光谱数据添加随机噪声数据和/或横纵坐标偏移数据（可以由本领域的技术人员，根据实际情况合理添加），构建已知成分的纯净物的光谱数据集。
然后，基于深度学习算法，对纯净物的光谱数据集中的光谱数据进行反复训练，直到得到光谱分析模型。
关于构建已知成分及各成分占比的混合物的光谱数据集，基于深度学习算法，对混合物的光谱数据集中的光谱数据进行训练，具体可以包括以下步骤：
首先，按照不同比例，向构建的已知成分的纯净物的光谱数据集中随机添加已知的不同物质的光谱数据，构建已知成分及各成分占比的混合物的光谱数据集。
然后，基于深度学习算法，对混合物的光谱数据集中的光谱数据进行反复训练，直到得到光谱分析模型。
需要说明的是，以上仅为举例说明，并不对本申请的技术方案和要保护的范围构成限定，本领域的技术人员可以根据需要进行设置，此处不做限制。
另外，上述所说的对光谱数据集中的光谱数据进行反复训练，直到得到光谱分析模型，具体是指，通过对大量光谱数据的反复训练，如对于纯净物A（成分为B），当通过多次训练只要输入纯净物A的光谱图加不同噪声和偏移，甚至不同光谱分辨率和范围，就可以输出B时，即可以认为训练完成。
另外，上述获取光谱分析模型的操作是在对被测物的光谱图进行光谱分析，确定被测物的成分及各成分的占比之前完成的。并且，训练获得的光谱分析模型可以直接保存在光谱分析设备上，也可以保存在远端服务器，或者二者均保存，根据某种同步机制定时更新同步，具体的实现方式，本领域的技术人员可以根据需要合理设置，此处不做限制。
为了对本申请的技术方案中涉及的关于光谱分析模型的训练有一个更加清楚、完整的理解，以下拉曼光谱分析为例进行具体说明，具体的训练示意图如图2所示。
需要说明的是，图2所示的光谱分析模型的训练是以数据库中存储的光谱数据的范围为200cm -1~3000cm -1，光谱数据为1400个点，每个点间隔2cm -1，库中有一万种纯净物的光谱数据为例，进行的训练。
具体的说，图2中的L1至L9分别为输入层（即Input Layer）、第一全连接层（即Dense Layer）、第一批量化层（即Batch Normalization Laye）、第二全连接层（即Dense Layer）、下降层（即Dropout Layer）、第二批量化层（即Batch Normalization Laye）、第三全连接层（即Dense Layer）、过滤层（即Threshold Filter层）、输出层。
在进行训练之前，需要将原有的1400个点（即1400*1的张量）延拓为10000个点（即10000*1的张量），即构建训练的光谱样本数据集。
具体的，先定义输入为1400×1的张量输入至L1。其中，Batch Size（批尺寸）在对输入的数据进行处理后获得。
输入的1400×1的张量经过L1后，传送至L2，在L2中，基于ReLU（修正线性单元，Rectified linear unit）激活函数对输入的1400×1的张量进行编译，确定是否需要加入偏置向量。
具体的公式如下：
output = activation（dot（input， kernel）+ bias），
需要说明的是，上述公式中activation是激活函数，kernel是本层的权值矩阵，bias为偏置向量，output输出的结果。
另外，需要说明的是ReLU激活函数的原理是将负数输入置0，正数保留，默认不加入偏置向量。关于该公式的使用，本领域的技术人员可以根据掌握的技术手段进行使用，此处不再赘述。
在L4中，定义输出为3200×1的张量形状，完成第一步延拓。其他的参数与L2中的类似，此处不再赘述。
L5主要用于在训练过程中每次更新参数时，按照比例（rate）随机断开输入神经元（即将输出张量部分位置置0），防止过度拟合，此处设置本层的比例参数设置为0.5。
在L7中，定义输出为10000×1的张量形状，完成第二步延拓，并利用Sigmoid激活函数对数据进行激活操作，其他的参数与L2中的类似，此处不再赘述。
在完成上述操作后，数据输入至L8，根据输出的张量值判断输入样本的组成成分，将每一个大于阈值的成分的输出用“1”表示，其他的用“0”表示，并将上述结构传输至L9，由L9输出。
另外，需要说明的是，图2中的L3和L6的主要作用是为了在每个批次的训练上，将L2和L4的输出的激活值规范化，使其输出数据均值接近0，标准差接近1。
在完成上述操作后，即可得到需要进行训练的初始样本数据，接着基于深度学习算法，对上述样本数据进行训练，具体如下：
需要说明的是，物质的拉曼光谱是由多个不同峰高的高斯峰（或洛伦兹峰）构成，峰宽取决于拉曼光谱仪的分辨能力，峰位取决于物质分子包含的化学键。不同物质的拉曼光谱主要体现峰位、峰高比的不同，混合物的拉曼光谱可以认为是组成混合物的纯净物的拉曼光谱的线性叠。
根据上述拉曼光谱的特点，将上述获得的10000种物质的纯净拉曼光谱数据均分成A、B、C和D四组（实际应用中，可以根据光谱数据的量，合理划分数据集，进行训练，此处不做限制），并且按照以下所属组合方式分别生成参与训练的数据集：
一种物质的训练数据（共10000种）： A集合：第1至第2500号物质； B集合：第2501至第5000号物质； C集合：第5001至第7500号物质； D集合：第7501至第10000号物质。
两种物质的训练数据（共15000种）： A集合与B集合，一一对应，采用随机浓度混合，得到2500种混合物；A集合与C集合，一一对应，采用随机浓度混合，得到2500种混合物；  A集合与D集合，一一对应，采用随机浓度混合，得到2500种混合物；B集合与C集合，一一对应，采用随机浓度混合，得到2500种混合物；B集合与D集合，一一对应，采用随机浓度混合，得到2500种混合物；C集合与D集合，一一对应，采用随机浓度混合，得到2500种混合物。
三种物质的训练数据（共10000种）：  A+B+C，一一对应，采用随机浓度混合，得到2500种混合物；B+C+D，一一对应，采用随机浓度混合，得到2500种混合物；A+B+D，一一对应，采用随机浓度混合，得到2500种混合物；A+C+D，一一对应，采用随机浓度混合，得到2500种混合物。
四种物质的训练数据（共10000种）：A+B+C+D，一一对应，采用随机浓度混合，得到2500种混合物；从全部物质中，随机抽取4种物质，采用随机浓度混合，进行7500次（在实际应用中，混合的此处可以由本领域的技术人员根据需要设置，此处不做限制），得到7500种混合物。
最后，将生成的45000组数据按照每1000条数据划分为一个Batch（1次迭代所使用的样本量）作为输入数据，然后基于深度学习算法进行一定次数的训练，直到loss（训练损失值）和val_loss（平均训练损失值）稳定且降低至10 -6以下，此时便可认为光谱分析模型可用。
需要说明的是，以上仅为举例说明，并不对本申请的技术方案和要保护的范围构成限定，本领域的技术人员可以根据需要进行设置，此处不做限制。
通过上述描述不难发现，本实施例中提供的光谱分析方法，通过对光谱样本数据集中的大量光谱数据进行训练获得用于定义光谱图与成分及各成分占比的映射关系的光谱分析模型，在需要分析被测物的成分及成分占比时，利用该光谱分析模型对被测物的光谱图进行光谱分析，从而能够快速分析出被测物中包含的多种成分及各成分的占比。另外，由于光谱分析模型是对大量光谱数据进行训练获得的，并且在训练的过程中向训练数据添加了预定的偏移，因而可以有效降低光谱偏移对分析结果的影响，保证了结果的准确性，另外通过训练不同分辨率和光谱范围的数据库实现了数据库之间的共用。
本申请的第二实施例涉及一种光谱分析方法。本实施例在第一实施例的基础上做了进一步改进，具体改进之处为：在确定被测物的成分及各成分的占比时，通过判断光谱分析结果中是否携带有分析成功的标签，并根据判断结果进行相应操作，具体流程如图3所示。
具体的说，在本实施例中，包含步骤301至步骤306，其中，步骤301、步骤302、步骤306分别与第一实施例中的步骤101至步骤103大致相同，此处不再赘述，下面主要介绍不同之处，未在本实施例中详尽描述的技术细节，可参见第一实施例所提供的碰光谱分析方法，此处不再赘述。
在步骤303中，判断光谱分析结果中是否携带有分析成功的标签。
具体的说，若光谱分析结果中携带有分析成功的标签，进入步骤306，否则，进入步骤304。
需要说明的是，在确定光谱分析结果中携带有分析成功的标签，进入步骤306执行确定被测物的成分及各成分的占比时，具体为将光谱分析结果中提取到的成分及各成分的占比，确定为被测物的成分及各成分的占比。
另外，在本实施例中，用于表示光谱分析操作是否成功的标签可以是单独设置的一个属性，比如在分析完后，在输出的光谱分析结果中携带“成功”或“失败”，或者是“YES”或“NO”的内容，从而来确定光谱分析操作是否成功。
另外，也可以不单独设置标识成功或失败的字段，直接设置识别成功的物质输出为“1”，未识别成功的物质输出为“0”，这样在根据光谱分析模型，对被测物的光谱图分析完成后，如果输出的光谱分析结果全部为“1”，则确定光谱分析操作成功，否则失败。并且，基于这种方式，还可以确定被测物是纯净物还是混合物（在确定光谱分析操作成功的前提下，如果“1”的个数为多个，则是混合物；反之则为纯净物）。
需要说明的是，以上仅为举例说明，并不对本申请的技术方案和要保护的范围构成限定，本领域的技术人员可以根据需要进行设置，此处不做限制。
在步骤304中，获取新的光谱分析模型。
具体的说，本实施例中所说的获取到的新的光谱分析模型是基于深度学习算法，对从网络中获取的其他已知成分及各成分占比的纯净物的光谱数据，和/或已知成分及各成分占比的混合物的光谱数据，进行训练得到的。
关于如何基于深度学习算法，对光谱数据进行训练得到光谱分析模式的流程，已经在本申请的第一实施例中详细记载，此处不再赘述。
在步骤305中，根据新的光谱分析模型，对被测物的光谱图进行光谱分析。
通过上述描述不难方向，与现有技术相比，本实施例中提供的光谱分析方法，在利用现有的光谱分析模型无法确定被测物的成分及各成分的占比时，基于深度学习算法，由光谱分析设备或远端服务器通过对网络中的大量已知成分及各成分占比的纯净物的光谱数据，和/或已知成分及各成分占比的混合物的光谱数据进行训练，获得新的光谱分析模型来对被测物的光谱图进行分析，进一步提升了用户体验。
另外，值得一提的是，在实际应用中，如果利用新获得的光谱分析模型也无法确定被测物的成分及各成分的占比，还可以采用现有的化学计量方式进行分析。
通过上述三种方式的配合，使得光谱分析操作能够尽可能的保证其分析结果的准确性、及时性，还可以大大提升用户的体验。
另外，为了保证设备或服务器能够从网络中获取光谱数据，从而能够基于深度学习算法，对网络中的光谱数据进行训练，得到新的光谱分析模型，在基于深度学习算法，对获取到的光谱数据进行训练之前，需要先确定能够访问网络。
若无法访问网络，则在光谱分析结果中位携带有分析成功的标签时，直接利用现有的化学计量法进行分析。
需要说明的是，以上仅为举例说明，并不对本申请的技术方案和要保护的范围构成限定，本领域的技术人员可以根据需要进行设置，此处不做限制。
本申请的第三实施例涉及一种光谱分析方法。本实施例在第一或第二实施例的基础上做了进一步改进，具体改进之处为：在确定被测物的成分及各成分的占比之后，保存被测物的成分、各成分的占比及被测物所处的环境信息。为了便于说明，以下针对在第一实施例上做的改进进行具体说明，具体流程如图4所示。
具体的说，在本实施例中，包含步骤401至步骤404，其中，步骤401至步骤403分别与第一实施例中的步骤101至步骤103大致相同，此处不再赘述，下面主要介绍不同之处，未在本实施例中详尽描述的技术细节，可参见第一实施例所提供的碰光谱分析方法，此处不再赘述。
在步骤404中，保存被测物的成分、各成分的占比及被测物所处的环境信息。
与现有技术相比，本实施例中提供的光谱分析方法，在确定被测物的成分及各成分的占比之后，通过将分析所得的被测物的成分、各成分的占比，以及通过光谱分析设备获取到的被测物所处的环境信息进行保存（如保存到光谱分析设备内部的存储器中，或者上传到云端服务器），从而可以方便测试人员在完成对被测物的光谱分析后，能够根据分析所得的数据进行后续的操作，如更新光谱样本数据集中的数据等。
需要说明的是，本实施例中所说的被测物所处的环境信息，具体可以包括被测物所处位置的GPS（全球定位系统，Global Positioning System）信息，周围环境的图像信息即其他相关信息，从而便于管理人员或管理中心对光谱分析设备和操作任意的管理。
另外，获取并保存被测物所处的环境信息，也便于筛查被测物周边是否存在危险事件，进而及时通知操作人员作出应对，避免操作人员和设备受到伤害。
另外，值得一提的是，在实际应用中，在确定被测物的成分及各成分的占比之后，还可以将该结果在光谱分析设备的显示界面显示，或者传送给其他设备进行后续处理。
需要说明的是，以上仅为举例说明，并不对本申请的技术方案和要保护的范围构成限定，本领域的技术人员可以根据需要进行设置，此处不做限制。
本申请的第四实施例涉及一种光谱分析装置，具体结构如图5所示。
如图5所示，光谱分析装置主要包括获取模块501和确定模块502。
其中，获取模块501，用于获取被测物的光谱图。确定模块502，用于根据预先获得的光谱分析模型，对被测物的光谱图进行光谱分析，确定被测物的成分及各成分的占比。
需要说明的是，在本实施例中，光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得。
另外，本实施例中训练获得的光谱分析模型主要是用于定义光谱图与成分及各成分占比的映射关系。
另外，需要说明的是，由于本实施例为与方法实施例对应的虚拟装置实施例，因而未在本实施例中详尽描述的技术细节，可参见本申请任一实施例所提供的光谱分析方法，此处不再赘述。
通过上述描述不难发现，本实施例中提供的光谱分析装置，通过对光谱样本数据集中的大量光谱数据进行训练获得用于定义光谱图与成分及各成分占比的映射关系的光谱分析模型，在需要分析被测物的成分及成分占比时，利用该光谱分析模型对被测物的光谱图进行光谱分析，从而能够快速分析出被测物中包含的多种成分及各成分的占比。另外，由于光谱分析模型是对大量光谱数据进行训练获得的，因而可以有效降低光谱偏移对分析结果的影响，保证了结果的准确性。
另外，需要说明的是，以上所描述的装置实施例仅仅是示意性的，并不对本申请的保护范围构成限定，在实际应用中，本领域的技术人员可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的，此处不做限制。
本申请的第五实施例涉及一种电子设备，具体结构如图6所示。
本实施例中所说的电子设备可以是光谱分析仪，如类似笔记本电脑的便携式光谱分析仪、小型的手持式光谱分析仪，还可以是安装有光谱分析应用程序的智能终端设备，如手机、平板电脑等，此处不再一一例举，也不做具体限制。
具体的说，该电子设备内部具体可以包括一个或多个处理器601以及与至少一个处理器601通信连接的存储器602和通信组件603，图6中以一个处理器601为例。
在本实施例中，上述实施例中涉及到的光谱分析装置中的各功能模块均部署在处理器601上，处理器601、存储器602和通信组件603可以通过总线或其他方式相互连接，图6中以通过总线连接为例。
存储器602作为一种计算机可读存储介质，可用于存储软件程序、计算机可执行程序以及模块，如本申请任意方法实施例中涉及的光谱分析方法对应的程序指令/模块。处理器601通过运行存储在存储器602中的软件程序、指令以及模块，并控制通信组件603接收和/或发送数据，完成本申请任意方法实施例中涉及的光谱分析方法。
另外，需要说明的是，存储器602可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可建立分析结果数据库，用于保存被测物的成分、各成分的占比及被测物所处的环境信息等。此外，存储器602可以包括高速随机存取存储器，还可以包括可读写存储器（RandomAccessMemory，RAM）等。在一些实施例中，存储器602可选包括相对于处理器601远程设置的存储器，这些远程存储器可以通过网络连接至终端设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
另外，在实际应用中，存储器602中可以存储至少一个处理器601执行的指令，指令被至少一个处理器601执行，以使至少一个处理器601能够执行本申请任意方法实施例涉及的光谱分析方法，控制光谱分析装置中的各个功能模块完成光谱分析方法中的各个操作，未在本实施例中详尽描述的技术细节，可参见本申请任一实施例所提供的光谱分析方法。
另外，值得一提的是，随着云计算技术的发展，为了进一步提升电子设备的处理能力，本实施例中所说的电子设备还可以是云端智能电子设备，即用于进行处理操作的电子设备的处理器是位于云端的。
具体的说，云端智能电子设备使得云端的智能计算能力成为一种便捷的服务，从而极大地降低了智能电子设备的研发成本与运营成本，并且利用云端的强大计算能力，可以更加方便快速的训练获得光谱分析模型。
需要说明的是，上述所说的两种类型的电子设备仅为本实施例中的具体举例说明，并不对本申请的技术方案和要保护的范围构成限定，在实际应用中，本领域的技术人员可以根据现有机器设备的发展情况，基于上述光谱分析方法的实现流程进行实现，此处不做限制。
本申请的第六实施例涉及一种计算机可读存储介质，该可读存储介质为计算机可读存储介质，该计算机可读存储介质中存储有计算机指令，该计算机指令使计算机能够执行本申请任意方法实施例中涉及的光谱分析方法。
本领域技术人员可以理解实现上述实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序存储在一个存储介质中，包括若干指令用以使得一个设备（可以是单片机，芯片等）或处理器（processor）执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器（ROM，Read-Only Memory）、随机存取存储器（RAM，Random Access Memory）、磁碟或者光盘等各种可以存储程序代码的介质。
本领域的普通技术人员可以理解，上述各实施例是实现本申请的具体实施例，而在实际应用中，可以在形式上和细节上对其作各种改变，而不偏离本申请的精神和范围。

Claims (11)

  1. 一种光谱分析方法，包括：
    获取被测物的光谱图；
    根据预先获得的光谱分析模型，对所述被测物的光谱图进行光谱分析，确定所述被测物的成分及各成分的占比；
    其中，所述光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得；
    所述光谱分析模型用于定义光谱图与成分及各成分占比的映射关系。
  2. 如权利要求1所述的光谱分析方法，其中，所述根据预先获得的光谱分析模型，对所述被测物的光谱图进行光谱分析，确定所述被测物的成分及各成分的占比之前，所述光谱分析方法还包括：
    构建已知成分的纯净物的光谱数据集，基于深度学习算法，对所述纯净物的光谱数据集中的光谱数据进行训练，获得所述光谱分析模型；
    和/或，构建已知成分及各成分占比的混合物的光谱数据集，基于深度学习算法，对所述混合物的光谱数据集中的光谱数据进行训练，获得所述光谱分析模型。
  3. 如权利要求2所述的光谱分析方法，其中，所述构建已知成分的纯净物的光谱数据集，基于深度学习算法，对所述纯净物的光谱数据集中的光谱数据进行训练，具体包括：
    获取已知成分的所述纯净物的光谱数据；
    向所述纯净物的光谱数据添加随机噪声数据和/或横纵坐标偏移数据，构建已知成分的纯净物的光谱数据集；
    基于深度学习算法，对所述纯净物的光谱数据集中的光谱数据进行反复训练，直到得到所述光谱分析模型。
  4. 如权利要求3所述的光谱分析方法，其中，所述构建已知成分及各成分占比的混合物的光谱数据集，基于深度学习算法，对所述混合物的光谱数据集中的光谱数据进行训练，具体包括：
    按照不同比例，向构建的已知成分的所述纯净物的光谱数据集中随机添加已知的不同物质的光谱数据，构建已知成分及各成分占比的混合物的光谱数据集；
    基于深度学习算法，对所述混合物的光谱数据集中的光谱数据进行反复训练，直到得到所述光谱分析模型。
  5. 如权利要求1至4任意一项所述的光谱分析方法，其中，所述根据预先获得的光谱分析模型，对所述被测物的光谱图进行光谱分析，确定所述被测物的成分及各成分的占比，具体包括：
    将所述被测物的光谱图中的光谱数据输入所述光谱分析模型；
    获取所述光谱分析模型输出的光谱分析结果，并从所述光谱分析结果中提取包含的成分及各成分的占比；
    将提取到的所述成分及各成分的占比，确定为所述被测物的成分及各成分的占比。
  6. 如权利要求5所述的光谱分析方法，其中，所述将提取到的所述成分及各成分的占比，确定为所述被测物的成分及各成分的占比，具体包括：
    判断所述光谱分析结果中是否携带有分析成功的标签；
    若所述光谱分析结果中携带有所述分析成功的标签，将提取到的所述成分及各成分的占比，确定为所述被测物的成分及各成分的占比；
    若所述光谱分析结果未携带有所述分析成功的标签，获取网络中存在的已知成分及各成分占比的纯净物的光谱数据，和/或已知成分及各成分占比的混合物的光谱数据，基于深度学习算法，对获取到的所述光谱数据进行训练，获得新的光谱分析模型，根据新的所述光谱分析模型，对所述被测物的光谱图进行光谱分析，确定所述被测物的成分及各成分的占比。
  7. 如权利要求6所述的光谱分析方法，其中，在基于深度学习算法，对获取到的所述光谱数据进行训练之前，所述光谱分析方法还包括：
    确定能够访问网络。
  8. 如权利要求1至7任意一项所述的光谱分析方法，其中，在确定所述被测物的成分及各成分的占比之后，所述光谱分析方法还包括：
    保存所述被测物的成分、各成分的占比及所述被测物所处的环境信息。
  9. 一种光谱分析装置，包括：获取模块和确定模块；
    所述获取模块，用于获取被测物的光谱图；
    所述确定模块，用于根据预先获得的光谱分析模型，对所述被测物的光谱图进行光谱分析，确定所述被测物的成分及各成分的占比；
    其中，所述光谱分析模型为对光谱样本数据集中的大量光谱数据进行训练获得；
    所述光谱分析模型用于定义光谱图与成分及各成分占比的映射关系。
  10. 一种电子设备，包括：
    至少一个处理器；以及，
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行如权利要求1至8任意一项所述的光谱分析方法。
  11. 一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至8任意一项所述的光谱分析方法。
PCT/CN2018/082239 2018-04-09 2018-04-09 光谱分析方法、装置、电子设备及计算机可读存储介质 WO2019195971A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201880001146.4A CN108780037A (zh) 2018-04-09 2018-04-09 光谱分析方法、装置、电子设备及计算机可读存储介质
PCT/CN2018/082239 WO2019195971A1 (zh) 2018-04-09 2018-04-09 光谱分析方法、装置、电子设备及计算机可读存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/082239 WO2019195971A1 (zh) 2018-04-09 2018-04-09 光谱分析方法、装置、电子设备及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2019195971A1 true WO2019195971A1 (zh) 2019-10-17

Family

ID=64029135

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/082239 WO2019195971A1 (zh) 2018-04-09 2018-04-09 光谱分析方法、装置、电子设备及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN108780037A (zh)
WO (1) WO2019195971A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111272669A (zh) * 2020-01-23 2020-06-12 深圳市大拿科技有限公司 基于粪便信息检测的健康评估方法及相关设备
CN112287426A (zh) * 2020-09-21 2021-01-29 广东众图互联网工程设计有限公司 一种调整建筑结构的构件的方法、装置以及电子设备

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110057757B (zh) * 2018-01-18 2022-04-26 深圳市理邦精密仪器股份有限公司 血红蛋白及其衍生物的识别、识别网络构建方法及装置
CN110532901A (zh) * 2019-08-12 2019-12-03 北京邮电大学 基于多目标检测的深度学习智能光谱分析方法及系统
CN111289455A (zh) * 2020-03-25 2020-06-16 欧梯恩智能科技(苏州)有限公司 分布式异味评价方法、装置、终端及可读存储介质
CN113358594B (zh) * 2020-06-30 2023-07-28 北京领主科技有限公司 基于光谱检测的物质成分分析系统、方法、装置及介质
CN113281291A (zh) * 2021-05-14 2021-08-20 深圳市八六三新材料技术有限责任公司 一种香精的组分分析方法、装置及计算机可读存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1201521A (zh) * 1995-11-08 1998-12-09 株式会社京都第一科学 能谱测量中用于谱处理的方法和设备
CN101299022A (zh) * 2008-06-20 2008-11-05 河南中医学院 利用近红外光谱技术评价中药药材综合质量的方法
CN107561033A (zh) * 2017-09-21 2018-01-09 上海理工大学 基于太赫兹光谱的混合物中关键物质定性和定量测定方法
CN107796766A (zh) * 2017-10-18 2018-03-13 盐城工学院 一种臭脚盐产地鉴别方法、装置及计算机可读存储介质

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1185477C (zh) * 2002-11-22 2005-01-19 中国人民解放军第二军医大学 分析混合物的化学信息修饰法和按该法工作的光谱分析仪
CN100458411C (zh) * 2006-07-24 2009-02-04 中国林业科学研究院木材工业研究所 植物纤维材料中化学成分含量模型的建立和含量测定方法
CN101419159B (zh) * 2008-11-24 2011-09-28 杨季冬 谱峰完全重叠的双组分混合物同时测定的光谱分析方法
MX2013003038A (es) * 2010-09-17 2013-05-01 Abbvie Inc Espectroscopia raman para operaciones de bioprocesos.
CN105651705A (zh) * 2015-01-14 2016-06-08 青海春天药用资源科技利用有限公司 无损检测冬虫夏草粉/粉片中伪品含量的方法
CN104792752A (zh) * 2015-04-03 2015-07-22 江南大学 三维荧光光谱结合parafac算法测定混合色素溶液中色素含量的方法
US10186026B2 (en) * 2015-11-17 2019-01-22 Kla-Tencor Corp. Single image detection
CN105548079A (zh) * 2015-12-28 2016-05-04 浙江中烟工业有限责任公司 一种基于近红外光谱的烟丝组成测定方法
CN105758836B (zh) * 2016-02-18 2018-07-27 安徽芯核防务装备技术股份有限公司 一种基于面积法的拉曼光谱即时定量分析方法
CN106124449B (zh) * 2016-06-07 2019-03-05 中国科学院合肥物质科学研究院 一种基于深度学习技术的土壤近红外光谱分析预测方法
CN107316046B (zh) * 2017-03-09 2020-08-25 河北工业大学 一种基于增量补偿动态自适应增强的故障诊断方法
CN107145846B (zh) * 2017-04-26 2018-10-19 贵州电网有限责任公司输电运行检修分公司 一种基于深度学习的绝缘子识别方法
CN107491784A (zh) * 2017-08-09 2017-12-19 云南瑞升烟草技术（集团）有限公司 基于深度学习算法的烟叶近红外光谱定量建模方法及应用
CN107679577A (zh) * 2017-10-12 2018-02-09 理光图像技术（上海）有限公司 基于深度学习的图像检测方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1201521A (zh) * 1995-11-08 1998-12-09 株式会社京都第一科学 能谱测量中用于谱处理的方法和设备
CN101299022A (zh) * 2008-06-20 2008-11-05 河南中医学院 利用近红外光谱技术评价中药药材综合质量的方法
CN107561033A (zh) * 2017-09-21 2018-01-09 上海理工大学 基于太赫兹光谱的混合物中关键物质定性和定量测定方法
CN107796766A (zh) * 2017-10-18 2018-03-13 盐城工学院 一种臭脚盐产地鉴别方法、装置及计算机可读存储介质

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111272669A (zh) * 2020-01-23 2020-06-12 深圳市大拿科技有限公司 基于粪便信息检测的健康评估方法及相关设备
CN112287426A (zh) * 2020-09-21 2021-01-29 广东众图互联网工程设计有限公司 一种调整建筑结构的构件的方法、装置以及电子设备
CN112287426B (zh) * 2020-09-21 2024-05-17 广东众图科技有限公司 一种调整建筑结构的构件的方法、装置以及电子设备

Also Published As

Publication number Publication date
CN108780037A (zh) 2018-11-09

Similar Documents

Publication Publication Date Title
WO2019195971A1 (zh) 光谱分析方法、装置、电子设备及计算机可读存储介质
CN110110743B (zh) 一种七类质谱谱图自动识别系统与方法
US10671684B2 (en) Method and apparatus for identifying demand
Wang et al. Mechanisms underlying local functional and phylogenetic beta diversity in two temperate forests
CN111368024A (zh) 文本语义相似度的分析方法、装置及计算机设备
CN107545000A (zh) 基于知识图谱的信息推送方法及装置
CN110688454A (zh) 咨询对话处理的方法、装置、设备及存储介质
KR102002024B1 (ko) 객체 라벨링 처리 방법 및 객체 관리 서버
CN110457677B (zh) 实体关系识别方法及装置、存储介质、计算机设备
WO2021253510A1 (zh) 基于双向交互网络的行人搜索方法、系统、装置
Gao et al. A Novel Deep Convolutional Neural Network Based on ResNet‐18 and Transfer Learning for Detection of Wood Knot Defects
CN108780048A (zh) 一种确定检测设备的方法、检测装置及可读存储介质
EP3965050A1 (en) Systems and methods for deriving rating for properties
Luo et al. A new algorithm for bilinear spectral unmixing of hyperspectral images using particle swarm optimization
WO2019196107A1 (zh) 物质成分的检测方法及相关装置和计算机可读存储介质
WO2022227759A1 (zh) 图像类别的识别方法、装置和电子设备
CN111289459A (zh) 混合物质组分浓度检测方法、装置、设备及存储介质
CN112180056A (zh) 一种基于稀土元素检测的酱香型白酒溯源方法及系统
CN115905524B (zh) 融合句法和语义信息的情感分析方法、装置以及设备
CN109060675A (zh) 一种用于铁矿石中铁含量检测方法及装置
Li et al. Development of a single-leaf disease severity automatic grading system based on image processing
CN103761530B (zh) 一种基于相关向量机的高光谱图像解混方法
CN114401285B (zh) 一种电力车联网智能算法模型协同下发方法及系统
CN113269433B (zh) 税收风险预测方法、设备、介质及计算机程序产品
CN109253981A (zh) 基于红外光谱的定量分析模型建立方法及装置

Legal Events

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

Ref document number: 18914883

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 11/02/2021

122 Ep: pct application non-entry in european phase

Ref document number: 18914883

Country of ref document: EP

Kind code of ref document: A1