CN114858782A - Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discrimination model - Google Patents
Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discrimination model Download PDFInfo
- Publication number
- CN114858782A CN114858782A CN202210782533.4A CN202210782533A CN114858782A CN 114858782 A CN114858782 A CN 114858782A CN 202210782533 A CN202210782533 A CN 202210782533A CN 114858782 A CN114858782 A CN 114858782A
- Authority
- CN
- China
- Prior art keywords
- milk powder
- raman
- raman hyperspectral
- countermeasure
- discrimination
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention discloses a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model, which belongs to the technical field of dairy product safety detection and comprises the steps of collecting Raman hyperspectral data of a normal milk powder sample and generating a credible Raman hyperspectral database; defining a probability distribution discrimination boundary corresponding to the credible Raman hyperspectral data by using a deep countermeasure learning algorithm; and collecting Raman hyperspectral data of the milk powder sample to be detected, and judging whether the Raman hyperspectral data of the milk powder sample to be detected obey the probability distribution of the credible Raman hyperspectrum. According to the method, the Raman hyperspectral imaging technology and the anti-discriminant model algorithm are organically combined, any other doped sample except a normal sample can be accurately identified without specific dopant information, and the efficiency of non-directional screening of milk powder doping is remarkably improved.
Description
Technical Field
The invention relates to the technical field of dairy product safety detection, in particular to a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model.
Background
The dairy product industry represented by infant milk powder is one of the food industries with the highest growth speed in China, and is closely related to the daily life of people. Milk powder has become an important nutrient source for people, particularly non-breast-fed infants, and the quality safety of the milk powder directly influences the life health of related people. However, the milk powder market is faced with economic benefit driven adulteration and counterfeiting, and a large number of unknown adulterants are added to milk powder by illegal vendors to seek higher economic benefits. These adulteration behaviors seriously affect the public credibility of the Chinese dairy industry and seriously endanger the health and safety of consumers. Therefore, in order to deal with various adulteration risks in the milk powder products, the development of a novel and efficient milk powder adulteration detection technology is urgently needed.
At present, detection of illegal additives in milk powder mainly depends on a directional detection method in a laboratory, and the challenges mainly comprise two aspects, namely, application of the directional detection method in practice is limited by various illegal additives, and exhaustive detection is difficult to carry out; secondly, most of the directional detection methods need complicated sample pretreatment processes to enter a detection stage, so that the method is time-consuming, labor-consuming, high in cost and complex in operation, and the rapid detection of large-batch milk powder is difficult to realize. The above two problems cause that the directional detection method has considerable limitations in the aspect of milk powder quality safety screening, and is difficult to actively find problems in daily supervision and adapt to the current severe milk powder quality safety situation. In order to overcome the defects of the directional detection method and achieve the aim of early warning of the quality safety of the milk powder, the non-directional detection method is produced.
Different from the directional detection method, the non-directional detection method has a relatively wide detection range, and can realize screening and prediction of unknown dopants in a certain range. However, most of the existing non-directional detection related methods are based on a chromatography-mass spectrometry combined method, and still need to rely on a large amount of known compound spectrum databases to combine various molecular structures so as to deduce an unknown dopant structure. Although this considerably extends the coverage of unknown dopants, the deduced unknown dopant structure remains quite limited, making it difficult to cover the entire chemical structure of the potential unknown dopant species. Therefore, in principle, the existing non-directional detection technology needs to rely on the prior knowledge of the doped samples to establish a discriminant model, and the problem of limited sample learning exists, namely, samples of a training set are difficult to cover the diversity of the doped substances, so that the discriminant model is extremely easy to be over-fitted. The defect in principle causes that the current non-directional detection technology is difficult to generate a doping discriminant model with high precision and high confidence.
Therefore, how to provide an efficient and accurate milk powder doping detection method is a problem that needs to be solved urgently by the technical personnel in the field.
Disclosure of Invention
In order to overcome various defects of the current non-directional detection technology, the invention provides a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discrimination model, which realizes the non-directional screening of milk powder samples in an in-situ lossless state and has the obvious advantages of simplicity, high efficiency, high precision and the like.
In order to achieve the purpose, the invention adopts the following technical scheme:
a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model comprises the following steps:
s1, collecting Raman hyperspectral data of a normal milk powder sample, and generating a credible Raman hyperspectral database;
s2, determining a probability distribution discrimination boundary corresponding to the credible Raman hyperspectral data by using a depth counterattack learning algorithm;
s3, collecting Raman hyperspectral data of the milk powder sample to be detected, and judging whether the Raman hyperspectral data of the milk powder sample to be detected obey probability distribution of credible Raman hyperspectrum.
Preferably, the raman hyperspectral data collected in steps S1 and S3 includes raman spectrum information data and spectral feature spatial distribution information.
Preferably, step S2 specifically includes:
based on a deep countermeasure learning algorithm, a Raman hyperspectral countermeasure discrimination model is constructed, and a probability distribution boundary corresponding to credible Raman hyperspectral data is determined by the Raman hyperspectral countermeasure discrimination model.
Preferably, the establishing of the raman hyperspectral countermeasure discrimination model and the determining of the probability distribution boundary corresponding to the credible raman hyperspectral data by using the raman hyperspectral countermeasure discrimination model specifically comprise the following steps:
s21, creating a discriminator D by using a convolutional neural network, randomly initializing network parameters of the discriminator D, and constructing a sparse initial discrimination boundary;
s22, creating a generator G by using a convolutional neural network, constructing a random initial mapping function, randomly sampling from standard normal distribution to obtain a latent feature z, and mapping the feature z into artificial Raman hyperspectral data G (z) by using the initial mapping function;
and S23, iteratively updating the discriminator D and the generator G to enable the initial discrimination boundary to continuously converge inwards and generate a probability distribution discrimination boundary which compactly surrounds the credible Raman hyperspectral data.
Preferably, the convolutional neural network for creating the discriminator D includes four convolutional layers and two fully-connected layers, where the four convolutional layers are used for latent feature z extraction, and the two fully-connected layers are used for integrating and compressing latent feature z, so as to finally obtain the artificial raman hyperspectral data probability distribution boundary.
Preferably, the convolutional neural network of the creation generator G includes four convolutional layers, the first three convolutional layers are used for latent feature z extraction, and the last convolutional layer is used for mapping the latent feature z to generate artificial raman hyperspectral data G (z).
Preferably, step S23 specifically includes:
s231, optimizing cost function by utilizing generator
s232, optimizing cost function by using discriminator
and S233, iterating the optimization generator G and the discriminator D, wherein the optimization direction is a gradient descending direction until the initial discrimination boundary of the discriminator D converges.
Preferably, the judging that the initial discrimination boundary of the discriminator D converges specifically includes:
respectively collecting Raman hyperspectral data of 20 parts of normal milk powder samples and 100 parts of doped milk powder samples as test sets;
marking the Raman hyperspectral data of the normal milk powder samples in the test set as negative samples, and marking the Raman hyperspectral data of the doped milk powder samples as positive samples;
respectively and sequentially inputting the negative samples and the positive samples in the test set into a countermeasure discrimination model for iterative training, and obtaining the times of false negative FN, true negative TN, false positive FP and true positive TP output results of the countermeasure discrimination model;
calculating three evaluation index values of a false negative rate FNR, a false positive rate FPR and an overall accuracy rate AR according to the times of false negative FN, true negative TN, false positive FP and true positive TP, and evaluating the convergence effect of the confrontation discrimination model by using the specific formula:
and when the obtained FNR and FPR values are less than 2 percent, the obtained AR value is greater than 95 percent, and the results of three evaluation index values in the ten times of iterative training of the test set are kept stable, judging that the initial judgment boundary of the discriminator D reaches convergence.
According to the technical scheme, compared with the prior art, the invention discloses a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model, and the method has the following beneficial effects:
(1) according to the method, the Raman hyperspectral imaging technology is adopted to represent the key component information of the milk powder, and the rapid nondestructive testing of the milk powder sample can be realized in situ without complex sample preparation.
(2) The method organically combines the confrontation discrimination model and the Raman hyperspectral imaging technology, simultaneously obtains the spectrum and the spatial information of the milk powder sample, and carries out deep fusion on the related information so as to obtain the hyperspectral topological distribution of the normal milk powder sample to the maximum extent, and the method has excellent detection sensitivity and spatial resolution. The spectral information refers to Raman spectral information of a single-point sample in a micro area by a Raman hyperspectral imaging technology, and accurately reflects material component information of the micro area; the spatial information refers in particular to spatial distribution information obtained in the whole area by a Raman hyperspectral imaging technology, and accurately reflects the spatial distribution information of the milk powder sample. Therefore, the Raman hyperspectral imaging technology has the capacity of both micro-area analysis and integral analysis, and the non-directional screening efficiency of milk powder doping is greatly improved.
(3) According to the method, the credible distribution of a normal milk powder sample is fitted by adopting antagonistic learning, then an artificial Raman hyperspectral sample is generated in an antagonistic training mode in the credible distribution boundary, and the discrimination boundary is continuously updated in an iterative manner to distinguish the artificial sample from the normal sample, so that the credible distribution boundary continuously shrinks inwards to form a compact boundary until the compact boundary converges. Therefore, the method skillfully avoids the dilemma that doping detection needs to exhaust doping materials and the mode thereof to cover unknown dopants in an unsupervised countertraining mode, overcomes the problem of limited data in the traditional analysis method, greatly expands the coverage range of non-directional screening, and theoretically has the screening capability of all dopants.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic flow diagram of a method provided by the present invention;
FIG. 2 is a schematic diagram of a network structure of a countermeasure decision model according to the present invention;
FIG. 3 is a schematic diagram illustrating the principle of formation of a discrimination boundary of the countermeasure discrimination model credible Raman hyperspectral data provided by the invention;
FIG. 4 is a schematic diagram illustrating the effect of the patterned doped milk powder positioning distribution region provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in FIGS. 1-3, the embodiment of the invention discloses a milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model, which comprises the following steps:
s1, collecting Raman hyperspectral data of a normal milk powder sample, and generating a credible Raman hyperspectral database;
s2, determining a probability distribution discrimination boundary corresponding to the credible Raman hyperspectral data by using a depth counterattack learning algorithm;
step S2 mainly includes the steps of constructing a Raman hyperspectral countermeasure discrimination model based on a depth countermeasure learning algorithm, and determining a probability distribution boundary corresponding to credible Raman hyperspectral data by using the Raman hyperspectral countermeasure discrimination model. The method specifically comprises the following steps:
s21, creating a discriminator D by using a convolutional neural network, randomly initializing network parameters of the discriminator D, and constructing a sparse initial discrimination boundary;
s22, creating a generator G by using a convolutional neural network, constructing a random initial mapping function, randomly sampling from standard normal distribution to obtain a latent feature z, and mapping the feature z into artificial Raman hyperspectral data G (z) by using the initial mapping function;
s23, iteratively updating the discriminator D and the generator G to enable the initial discrimination boundary to continuously converge inwards and generate a probability distribution discrimination boundary which compactly surrounds the credible Raman hyperspectral data;
step S23 specifically includes:
s231, optimizing a cost function by using a generator
s232, optimizing cost function by using discriminator
and S233, iterating the optimization generator G and the discriminator D, wherein the optimization direction is a gradient descending direction until the initial discrimination boundary of the discriminator D converges.
S3, collecting Raman hyperspectral data of the milk powder sample to be detected, and judging whether the Raman hyperspectral data of the milk powder sample to be detected obey probability distribution of credible Raman hyperspectrum.
The raman hyperspectral data collected in steps S1 and S3 includes raman spectrum information data and spectral feature spatial distribution information. The Raman spectrum information refers to Raman spectrum information of a single-point sample in a micro area by a Raman hyperspectral imaging technology, and accurately reflects material component information of the micro area; the spectral characteristic spatial distribution information refers in particular to spatial distribution information obtained in the whole area by a Raman hyperspectral imaging technology, and accurately reflects the spatial distribution information of the milk powder sample.
In one embodiment, the step S223 of determining that the initial discrimination boundary of the discriminator D reaches convergence specifically includes:
respectively collecting Raman hyperspectral data of 20 parts of normal milk powder samples and 100 parts of doped milk powder samples as test sets;
marking the Raman hyperspectral data of the normal milk powder samples in the test set as negative samples, and marking the Raman hyperspectral data of the doped milk powder samples as positive samples;
respectively and sequentially inputting the negative samples and the positive samples in the test set into a countermeasure discrimination model for iterative training, and obtaining the times of false negative FN, true negative TN, false positive FP and true positive TP output results of the countermeasure discrimination model;
calculating three evaluation index values of a false negative rate FNR, a false positive rate FPR and an overall accuracy rate AR according to the times of false negative FN, true negative TN, false positive FP and true positive TP, and evaluating the convergence effect of the confrontation discrimination model by using the specific formula:
and when the obtained FNR and FPR values are less than 2 percent, the obtained AR value is greater than 95 percent, and the results of three evaluation index values in the ten times of iterative training of the test set are kept stable, judging that the initial judgment boundary of the discriminator D reaches convergence.
The false negative in the embodiment of the invention refers to that the true result of the sample is the adulterated sample, but the adulterated sample is negative after being judged, namely the actual adulterated sample is judged as a normal sample, which is also called as the omission detection; true negatives refer to: the true result of the sample is a normal sample which is negative after being judged, namely the normal sample is finally determined to be negative, and the negative judgment is correct. Similarly, false positive means that the real result of the sample is a normal sample, but is judged to be a doped sample; and the true positive indicates that the true result of the sample is the doped sample, and the judgment result is the doped sample.
The method skillfully avoids the problem of doped milk powder labeling, converts the problem of milk powder doping into unsupervised calculation of depth confidence boundary probability, obtains the tight support discrimination boundary distribution of normal sample Raman hyperspectral distribution in the mode of 'artificial Raman spectrum data generation' and 'discrimination boundary update' alternate iterative update, avoids the dilemma that unknown doped milk powder can be covered by exhausting doped substances and doping modes in mathematical principles, greatly improves the non-directional detection efficiency and accuracy of doped milk powder, and completely meets the requirement of daily milk powder doping screening with detection precision.
In order to further explain the detection steps and the detection effect of the invention, the following further explains two embodiments of single-point sampling doped milk powder non-directional doping detection only containing Raman spectrum information data and area sampling non-uniform doped milk powder dopant positioning detection simultaneously containing Raman spectrum information data and spectral characteristic spatial distribution information.
Example a: non-directional doping detection of doped milk powder based on single sampling point
The method comprises the following steps:
(1) trusted sample set preparation
Selecting a plurality of mainstream brand milk powders on the market verified by an authoritative laboratory as a credible sample set, and selecting 120 samples of different brands and different batches from the mainstream brand milk powders, wherein the weight of each sample is 20 g.
(2) Doped sample preparation
In order to verify the reliability of the confrontation discrimination model, the milk powder sample is artificially doped. According to the embodiment of the invention, a plurality of reported common illegal additives including melamine, cyanuric acid, starch, whey powder, maltodextrin and the like are selected and uniformly added into a credible milk powder sample so as to effectively simulate possible milk powder doping behaviors in reality. The doping mode mainly comprises a single-additive doping sample and a multi-additive doping sample, wherein the concentration of the additive in the doping sample is 0.5% -1.5%, and all the doping samples are effectively mixed through a turbine stirring device. In the embodiment of the invention, 150 single-additive-doped samples and 200 multi-additive-doped samples are prepared.
(3) Data packet
And randomly selecting 70 normal milk powder samples as a training set, 20 normal milk powder samples and 100 doped samples as a test set, and taking 30 normal samples and 250 doped samples as a verification set in all the other samples.
(4) Data acquisition
The Raman hyperspectral collection of all samples is completed by adopting a point scanning Raman hyperspectral imaging system, wherein the adopted Raman spectrometer has the excitation wavelength of 785nm, the power of 100mW, the integration time of 200ms, the moving step length of the adopted two-dimensional imaging displacement platform signal is 1mm, the sampling area is a square area of 50mm multiplied by 50mm, and 2500 sample points are scanned in each sampling.
(5) Model construction and training
And marking the normal sample as 1 and the abnormal sample as 0, performing repeated iterative training on the training set in an antagonistic learning mode described by the method, and performing model performance evaluation on the test set after each training to finally complete the compact distribution boundary probability fitting of the normal milk powder.
(6) Model validation
Inputting the verification set data into the discrimination model obtained in the fifth step, and according to the probability output by the discrimination model, if the probability is greater than 0.5, the milk powder is considered to be normal milk powder, if the probability is less than 0.5, the milk powder is considered to be doped milk powder, the sensitivity of the model obtained by final test is 98.3%, the specificity is 99.2%, and the overall accuracy is 99.1%.
Example b: non-uniform doped milk powder dopant positioning detection based on region sampling
(1) Preparation of test specimens
In this example, the test sample is a doped milk powder sample having a certain doped region, and is prepared by placing the uniformly doped milk powder sample prepared in example a into a funnel and coating the uniformly doped milk powder on the detection surface of the normal milk powder according to a certain shape. Different doping region shapes are used for different additives, including the letters "a" through "H", where the letter "G" is not used because it is more difficult to draw.
(2) Data acquisition
The same method as that used in example a was used to collect raman hyperspectral imaging data from the doped milk powder obtained in step (1), to obtain three-dimensional raman imaging data containing spectral feature spatial distribution information, and simultaneously, raman hyperspectral imaging data from a pure milk powder sample was used as a control sample.
The collection of spectral characteristic spatial distribution information can realize point-by-point or line-by-line scanning by carrying a milk powder sample through a two-dimensional imaging displacement platform to obtain three-dimensional Raman imaging data (comprising three dimensions of x, y and z, wherein the dimension of x and y corresponds to a spatial dimension, and the dimension of z corresponds to a spectral dimension).
(3) Dopant localization
Inputting the Raman hyperspectral imaging data containing spectral feature spatial distribution information obtained in the step (2) into a discrimination model to discriminate the authenticity of each sampling point in the doped milk powder sample, (in the embodiment, the training and verification process of the countermeasure discrimination model is not repeated), if the output value of the countermeasure discrimination model is greater than 0.5, the sample point is considered to have no doping, if the output value of the countermeasure discrimination model is less than 0.5, the sample point is considered to have doping, the normal sample point is marked to be black, the abnormal sample point is marked to be white, and as shown in fig. 4, the set non-uniform doping area is clearly reproduced by the obtained result.
In the discrimination process, the data integration of the spectral feature spatial distribution information is mainly embodied in a discriminator D, firstly, spatial information on x-y spatial dimension is extracted by adopting two-dimensional convolution, then, the spatial information is compressed to a one-dimensional vector, and the data feature is continuously extracted by adopting one-dimensional convolution operation to carry out final discrimination.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A milk powder doping non-directional detection method based on a Raman hyperspectral countermeasure discriminant model is characterized by comprising the following steps:
s1, collecting Raman hyperspectral data of a normal milk powder sample, and generating a credible Raman hyperspectral database;
s2, determining a probability distribution discrimination boundary corresponding to the credible Raman hyperspectral data by using a depth counterattack learning algorithm;
s3, collecting Raman hyperspectral data of the milk powder sample to be detected, and judging whether the Raman hyperspectral data of the milk powder sample to be detected obey probability distribution of credible Raman hyperspectrum.
2. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discriminant model as claimed in claim 1, wherein the Raman hyperspectral data collected in steps S1 and S3 comprises Raman spectrum information data and spectral feature spatial distribution information.
3. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discriminant model as claimed in claim 1, wherein step S2 specifically comprises:
based on a deep countermeasure learning algorithm, a Raman hyperspectral countermeasure discrimination model is constructed, and a probability distribution boundary corresponding to credible Raman hyperspectral data is determined by the Raman hyperspectral countermeasure discrimination model.
4. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discrimination model according to claim 3 is characterized in that the Raman hyperspectral countermeasure discrimination model is constructed, and the step of determining the probability distribution boundary corresponding to the credible Raman hyperspectral data by using the Raman hyperspectral countermeasure discrimination model specifically comprises the following steps:
s21, creating a discriminator D by using a convolutional neural network, randomly initializing network parameters of the discriminator D, and constructing a sparse initial discrimination boundary;
s22, creating a generator G by using a convolutional neural network, constructing a random initial mapping function, randomly sampling from standard normal distribution to obtain latent features z, and mapping the latent features z into artificial Raman hyperspectral data G (z) by using the initial mapping function;
and S23, iteratively updating the discriminator D and the generator G to enable the initial discrimination boundary to continuously converge inwards and generate a probability distribution discrimination boundary which compactly surrounds the credible Raman hyperspectral data.
5. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discriminant model is characterized in that the convolutional neural network for creating the discriminant D comprises four convolutional layers and two fully-connected layers, wherein the four convolutional layers are used for latent feature z extraction, and the two fully-connected layers are used for integrating and compressing the latent feature z to finally obtain an artificial Raman hyperspectral data probability distribution boundary.
6. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discriminant model is characterized in that the convolutional neural network of the generator G comprises four convolutional layers, the first three convolutional layers are used for latent feature z extraction, and the last convolutional layer is used for mapping the latent feature z to generate artificial Raman hyperspectral data G (z).
7. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discriminant model as claimed in claim 4, wherein the step S23 specifically comprises:
s231, optimizing cost function by utilizing generator
s232, optimizing cost function by using discriminator
and S233, iterating the optimization generator G and the discriminator D, wherein the optimization direction is a gradient descending direction until the initial discrimination boundary of the discriminator D converges.
8. The milk powder doping non-directional detection method based on the Raman hyperspectral countermeasure discrimination model according to claim 7, wherein the step of judging whether the initial discrimination boundary of the discriminator D converges specifically comprises the following steps:
respectively collecting Raman hyperspectral data of 20 parts of normal milk powder samples and 100 parts of doped milk powder samples as test sets;
marking the Raman hyperspectral data of the normal milk powder samples in the test set as negative samples, and marking the Raman hyperspectral data of the doped milk powder samples as positive samples;
respectively and sequentially inputting the negative samples and the positive samples in the test set into a countermeasure discrimination model for iterative training, and obtaining the times of false negative FN, true negative TN, false positive FP and true positive TP output results of the countermeasure discrimination model;
calculating three evaluation index values of a false negative rate FNR, a false positive rate FPR and an overall accuracy rate AR according to the times of false negative FN, true negative TN, false positive FP and true positive TP, and evaluating the convergence effect of the confrontation discrimination model by using the specific formula:
and when the obtained FNR and FPR values are less than 2 percent, the obtained AR value is greater than 95 percent, and the results of three evaluation index values in the ten times of iterative training of the test set are kept stable, judging that the initial judgment boundary of the discriminator D reaches convergence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210782533.4A CN114858782B (en) | 2022-07-05 | 2022-07-05 | Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discriminant model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210782533.4A CN114858782B (en) | 2022-07-05 | 2022-07-05 | Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discriminant model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114858782A true CN114858782A (en) | 2022-08-05 |
| CN114858782B CN114858782B (en) | 2022-09-27 |
Family
ID=82626623
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210782533.4A Active CN114858782B (en) | 2022-07-05 | 2022-07-05 | Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discriminant model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114858782B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116310566A (en) * | 2023-03-23 | 2023-06-23 | 华谱科仪(北京)科技有限公司 | Chromatographic data graph processing method, computer device and computer readable storage medium |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563355A (en) * | 2017-09-28 | 2018-01-09 | 哈尔滨工程大学 | Hyperspectral abnormity detection method based on generation confrontation network |
| CN109145992A (en) * | 2018-08-27 | 2019-01-04 | 西安电子科技大学 | Cooperation generates confrontation network and sky composes united hyperspectral image classification method |
| CN109948693A (en) * | 2019-03-18 | 2019-06-28 | 西安电子科技大学 | Expand and generate confrontation network hyperspectral image classification method based on super-pixel sample |
| CN110119780A (en) * | 2019-05-10 | 2019-08-13 | 西北工业大学 | Based on the hyperspectral image super-resolution reconstruction method for generating confrontation network |
| CN110320162A (en) * | 2019-05-20 | 2019-10-11 | 广东省智能制造研究所 | A kind of semi-supervised high-spectral data quantitative analysis method based on generation confrontation network |
| CN110610156A (en) * | 2019-09-12 | 2019-12-24 | 河海大学 | Hyperspectral image classification method based on generation countermeasure network |
| WO2020029356A1 (en) * | 2018-08-08 | 2020-02-13 | 杰创智能科技股份有限公司 | Method employing generative adversarial network for predicting face change |
| CN111413316A (en) * | 2020-04-09 | 2020-07-14 | 天津大学 | Milk powder authenticity screening method based on Raman spectrum |
| CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
| US20200380366A1 (en) * | 2018-06-12 | 2020-12-03 | Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences | Enhanced generative adversarial network and target sample recognition method |
| CN112699899A (en) * | 2020-12-31 | 2021-04-23 | 杭州电子科技大学 | Hyperspectral image feature extraction method based on generation countermeasure network |
| CN112712857A (en) * | 2020-12-08 | 2021-04-27 | 北京信息科技大学 | Method for generating biological Raman spectrum data based on WGAN (WGAN) antagonistic generation network |
| WO2021174935A1 (en) * | 2020-03-03 | 2021-09-10 | 平安科技（深圳）有限公司 | Generative adversarial neural network training method and system |
| CN113378680A (en) * | 2021-06-01 | 2021-09-10 | 厦门大学 | Intelligent database building method for Raman spectrum data |
| WO2021253632A1 (en) * | 2020-06-19 | 2021-12-23 | 深圳新视智科技术有限公司 | Cloth defect detection method based on adversarial neural network, and terminal and storage medium |
| CN113947712A (en) * | 2021-09-02 | 2022-01-18 | 西安电子科技大学 | Hyperspectral anomaly detection method and system based on capsule differential countermeasure network |
| CN114218985A (en) * | 2021-12-09 | 2022-03-22 | 上海理工大学 | Model system for classifying types of rejection resistance reactions in kidney transplantation and construction method thereof |
-
2022
- 2022-07-05 CN CN202210782533.4A patent/CN114858782B/en active Active
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563355A (en) * | 2017-09-28 | 2018-01-09 | 哈尔滨工程大学 | Hyperspectral abnormity detection method based on generation confrontation network |
| US20200380366A1 (en) * | 2018-06-12 | 2020-12-03 | Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences | Enhanced generative adversarial network and target sample recognition method |
| WO2020029356A1 (en) * | 2018-08-08 | 2020-02-13 | 杰创智能科技股份有限公司 | Method employing generative adversarial network for predicting face change |
| CN109145992A (en) * | 2018-08-27 | 2019-01-04 | 西安电子科技大学 | Cooperation generates confrontation network and sky composes united hyperspectral image classification method |
| CN109948693A (en) * | 2019-03-18 | 2019-06-28 | 西安电子科技大学 | Expand and generate confrontation network hyperspectral image classification method based on super-pixel sample |
| CN110119780A (en) * | 2019-05-10 | 2019-08-13 | 西北工业大学 | Based on the hyperspectral image super-resolution reconstruction method for generating confrontation network |
| CN110320162A (en) * | 2019-05-20 | 2019-10-11 | 广东省智能制造研究所 | A kind of semi-supervised high-spectral data quantitative analysis method based on generation confrontation network |
| CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
| CN110610156A (en) * | 2019-09-12 | 2019-12-24 | 河海大学 | Hyperspectral image classification method based on generation countermeasure network |
| WO2021174935A1 (en) * | 2020-03-03 | 2021-09-10 | 平安科技（深圳）有限公司 | Generative adversarial neural network training method and system |
| CN111413316A (en) * | 2020-04-09 | 2020-07-14 | 天津大学 | Milk powder authenticity screening method based on Raman spectrum |
| WO2021253632A1 (en) * | 2020-06-19 | 2021-12-23 | 深圳新视智科技术有限公司 | Cloth defect detection method based on adversarial neural network, and terminal and storage medium |
| CN112712857A (en) * | 2020-12-08 | 2021-04-27 | 北京信息科技大学 | Method for generating biological Raman spectrum data based on WGAN (WGAN) antagonistic generation network |
| CN112699899A (en) * | 2020-12-31 | 2021-04-23 | 杭州电子科技大学 | Hyperspectral image feature extraction method based on generation countermeasure network |
| CN113378680A (en) * | 2021-06-01 | 2021-09-10 | 厦门大学 | Intelligent database building method for Raman spectrum data |
| CN113947712A (en) * | 2021-09-02 | 2022-01-18 | 西安电子科技大学 | Hyperspectral anomaly detection method and system based on capsule differential countermeasure network |
| CN114218985A (en) * | 2021-12-09 | 2022-03-22 | 上海理工大学 | Model system for classifying types of rejection resistance reactions in kidney transplantation and construction method thereof |
Non-Patent Citations (3)
| Title |
|---|
| 刘恋秋: "基于深度卷积生成对抗网络的图像识别算法", 《液晶与显示》 * |
| 张鹏强等: "面向高光谱影像分类的生成式对抗网络", 《测绘通报》 * |
| 王铎等: "基于深度卷积条件生成对抗网络的虫害分类算法研究", 《中国农机化学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116310566A (en) * | 2023-03-23 | 2023-06-23 | 华谱科仪(北京)科技有限公司 | Chromatographic data graph processing method, computer device and computer readable storage medium |
| CN116310566B (en) * | 2023-03-23 | 2023-09-15 | 华谱科仪(北京)科技有限公司 | Chromatographic data graph processing method, computer device and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114858782B (en) | 2022-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110322423B (en) | Multi-modal image target detection method based on image fusion | |
| CN112213287B (en) | Coastal beach salinity inversion method based on remote sensing satellite image | |
| CN110174359B (en) | Aviation hyperspectral image soil heavy metal concentration assessment method based on Gaussian process regression | |
| CN105486655B (en) | The soil organism rapid detection method of model is intelligently identified based on infrared spectroscopy | |
| CN114858782B (en) | Milk powder doping non-directional detection method based on Raman hyperspectral countermeasure discriminant model | |
| CN112069955B (en) | Typhoon intensity remote sensing inversion method based on deep learning | |
| CN114460013B (en) | Coastal wetland vegetation overground biomass GAN model self-learning remote sensing inversion method | |
| CN117789038B (en) | Training method of data processing and recognition model based on machine learning | |
| CN114676779A (en) | Atmospheric quality monitoring data calibration method and system based on robust width network | |
| Tian et al. | Improving change detection in forest areas based on stereo panchromatic imagery using kernel MNF | |
| CN117541626B (en) | Remote sensing image change detection method based on deep learning | |
| CN113959974A (en) | Universal near-infrared spectrum model transfer method | |
| CN109670408A (en) | A kind of object-based remote sensing images Clean water withdraw method | |
| CN116842416B (en) | Calculation method suitable for coral reef sand foundation pipe pile side friction resistance | |
| TWI731689B (en) | Tomography method, system and apparatus based on time-domain spectroscopy | |
| CN111222576B (en) | High-resolution remote sensing image classification method | |
| CN116363536B (en) | Unmanned aerial vehicle inspection data-based power grid infrastructure equipment defect archiving method | |
| CN117011759A (en) | Method and system for analyzing multi-element geological information of surrounding rock of tunnel face by drilling and blasting method | |
| CN111666999A (en) | Remote sensing image classification method | |
| CN111199209A (en) | Bearing time-frequency spectrogram identification method based on IWO-KFCM algorithm | |
| CN114780917B (en) | Milk powder doping non-directional detection method based on flow discrimination model | |
| CN115078327A (en) | Rapid detection method for dangerous chemicals based on neural network architecture search | |
| CN112801028A (en) | Spectrum and space map hyperspectral image classification method based on inductive representation learning network | |
| Li et al. | Study on characteristic spectrum and multiple classifier fusion with different particle size in marine sediments | |
| CN112070761B (en) | Prawn freshness nondestructive testing method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |