[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN107067017B - Burn depth prediction system based on near infrared spectrum of CAGA and SVM - Google Patents

Burn depth prediction system based on near infrared spectrum of CAGA and SVM Download PDF

Info

Publication number
CN107067017B
CN107067017B CN201611072092.XA CN201611072092A CN107067017B CN 107067017 B CN107067017 B CN 107067017B CN 201611072092 A CN201611072092 A CN 201611072092A CN 107067017 B CN107067017 B CN 107067017B
Authority
CN
China
Prior art keywords
population
near infrared
burn
infrared spectrum
optimal
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.)
Active
Application number
CN201611072092.XA
Other languages
Chinese (zh)
Other versions
CN107067017A (en
Inventor
王品
吴烨
李勇明
尹美芳
吴军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Depp Medical Polytron Technologies Inc
Original Assignee
Individual
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 Individual filed Critical Individual
Priority to CN201611072092.XA priority Critical patent/CN107067017B/en
Publication of CN107067017A publication Critical patent/CN107067017A/en
Application granted granted Critical
Publication of CN107067017B publication Critical patent/CN107067017B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/359—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Pathology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Immunology (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention provides a burn depth prediction system based on near infrared spectrum of CAGA and SVM: collecting near infrared spectrum signals in a sample to be predicted, and performing pretreatment and feature extraction; the prediction device predicts the burn depth of a sample to be predicted by using a CAGA-SVM integrated detection model; the model builder collects burn training samples and adopts a support vector machine to build a regression model of near infrared spectrum signals and burn depths; and performing global optimization on kernel function parameters and feature subsets of the regression model by adopting a chain type intelligent agent genetic algorithm to obtain a CAGA-SVM integrated detection model. The system constructs a burn depth inversion integration model, accurately establishes a complex relation between near infrared spectrum signals and skin burn depth, and realizes accurate quantitative detection of the skin burn depth. The experimental result shows that the algorithm of the invention has high accuracy, good stability and strong generalization capability, and can accurately predict the skin burn depth.

Description

Burn depth prediction system based on near infrared spectrum of CAGA and SVM
Technical Field
The invention particularly relates to a burn depth prediction system based on near infrared spectrum of CAGA and SVM.
Background
About 1100 million burn patients are worldwide every year, and millions of new burn patients are only generated in China every year. The burn wound depth can be judged and evaluated quickly and accurately in the early stage after injury, so that an auxiliary basis is provided for a doctor to select a treatment means as early as possible, the problems of excessive paralysis hyperplasia, dysfunction, treatment cost increase and the like are avoided to the greatest extent, and the method is a necessary premise for improving diagnosis and treatment quality. The accurate detection of the burn depth can provide an important reference basis for clinical treatment, but the burn depth diagnosis still depends on the judgment of the experience of a clinician so far, and is lack of objectivity, stability, quantification, simplicity and the like. The degree of burn is generally expressed by the common third degree quartile method, and the degree of burn is generally classified into first degree burn, superficial second degree burn, deep second degree burn and third degree burn. The degree and depth of the burn, the area of the burn and the nature of the burn. According to the dividing mode, the accuracy rate of judging the depth can only reach 65% -70% even by the most experienced clinician, and the accuracy rate can only reach about 40% when judging the shallow second-degree burn and the deep second-degree burn. Therefore, the research on a method capable of accurately detecting the burn depth has important practical significance.
The near infrared spectrum is used as a spectrum detection tool for noninvasive, non-contact and rapid detection. Because the absorption rate of organic molecules to near infrared light shows a certain rule on the spectrum at different wavelengths, the near infrared spectrum can be used for qualitative and quantitative analysis of the organic. Based on this, researchers have studied the burn skin state by using the near infrared spectrum, for example, Space Frequency Domain Imaging (SFDI) is proposed, and the burn state is evaluated, the burn depth is a parameter in the burn depth detection, when detecting the burn depth, factors such as burn position, burn temperature, burn depth, skin condition, age, past diseases and the like are often required to be considered, and because the relation between the near infrared spectrum signal and the burn depth is complex, and in addition, accurate prediction of the burn depth is difficult, the burn depth detection is not realized at present.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a burn depth prediction system based on near infrared spectrum of CAGA and SVM, which can effectively and accurately predict the burn depth.
The burn depth prediction system based on the near infrared spectrum of the CAGA and the SVM comprises a near infrared spectrum collector, a model builder and a prediction device; wherein,
the near infrared spectrum collector collects near infrared spectrum signals in a sample to be predicted, and the near infrared spectrum signals are preprocessed and subjected to characteristic extraction and transmitted to the prediction device;
the prediction device predicts the burn depth of a sample to be predicted by utilizing the established CAGA-SVM integrated detection model according to the data uploaded by the near infrared spectrum collector;
the model builder comprises an estimation device and an optimization device, wherein the estimation device collects a plurality of burn training samples and adopts a support vector machine to build a regression model of near infrared spectrum signals and burn depths; and the optimization device adopts a chain type intelligent agent genetic algorithm to carry out global optimization on the kernel function parameters and the feature subsets of the regression model to obtain the CAGA-SVM integrated detection model.
Preferably, the estimation device uses the scalding head of the temperature-control scalding instrument to scald different areas of the pigskin at different depths, wherein the scalded pigskin at each depth point on each area forms a scalding depth training sample; and collecting near infrared spectrum signals for each type of burn depth training sample by using a near infrared spectrometer to form the burn training sample.
Preferably, the estimating device adopts a support vector machine to establish a regression model of the near infrared spectrum signal and the burn depth as follows:
Figure GDA0002367901360000021
wherein, the burn training sample is { xl,yl1, 2. m, m is the sample volume, xlThe intensity value of the near infrared spectrum signal; y islThe depth of burn; k (x, x)l) Is a kernel function of the support vector machine; b is a constant; lagrange factor al，
Figure GDA0002367901360000022
The constraint of the maximization function is:
Figure GDA0002367901360000023
wherein C is a kernel function parameter.
Preferably, the kernel function of the support vector machine is a linear kernel, and the expression is:
K(x,xl)＝(x·xl) (3)。
preferably, the kernel function of the support vector machine is a radial basis kernel, and the expression is:
K(x,xl)＝exp(-γ||x-xl||2) (4)
where γ is a kernel function parameter.
Preferably, the optimization device, first, sets the C optimization selection interval as [1, 2000], and the γ optimization selection interval as [0.001, 2 ]; then, an initial population is set: each population individual is identified by one binary code, different binary digits in the binary code represent different characteristics of the population individual, when a certain characteristic in the population individual is selected, the binary digit corresponding to the characteristic is 1, otherwise, the binary digit is 0; defining a kernel function parameter C as a decimal number represented by 1 st-11 th binary digits of binary codes in the population individuals, and defining a kernel function parameter gamma as a decimal number represented by 12 nd-22 th binary digits of binary codes in the population individuals; secondly, screening the initial population of each iteration by adopting a chain type intelligent agent genetic algorithm: performing neighborhood competition selection, adaptive crossing and adaptive variation on the initial population to obtain a variation population, combining the initial population and the variation population, obtaining a new population by adopting a bubble sorting method, and taking the new population as the initial population of the next iteration; finally, judging whether the current iteration meets the evolution termination condition, and if not, screening the initial population of the next iteration; if so, obtaining an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual; and obtaining an optimized regression model, namely a CAGA-SVM integrated detection model, based on the optimal kernel function parameters C and gamma and the optimal feature subset.
Preferably, the neighborhood competition selection is performed sequentially from left to right on a cyclic chain, and the current agent is set as
Figure GDA0002367901360000031
Its neighborhood is Nbs1,j，
Figure GDA0002367901360000032
j1, 2, …, popsize, the number of individuals in a population, wherein
Figure GDA0002367901360000033
The update method of (1) is as follows:
Figure GDA0002367901360000041
wherein,
Figure GDA0002367901360000048
representing competing operations between agents;
the formula of the adaptive intersection is as follows:
Figure GDA0002367901360000042
in the formula, pcFor crossover probability, GH (j, j') is Lt 1,jAnd max1,jHaiming distance of, max1,jIs Nbs1,jThe individual with the maximum moderate fitness value, and f' is Lt 1,jAnd max1,jThe fitness value with the highest fitness value in the intermediate fitness values,
Figure GDA0002367901360000043
the average fitness value of the population individuals in the iteration is obtained;
the method for judging the self-adaptive variation comprises the following steps: when U (0,1) < pmJudging the population variation, wherein pmAs the mutation probability, pmLength is the number of genes of a single agent, i.e. the number of bits of the binary code in the initial population;
the method for obtaining the new population by adopting the bubbling sorting method comprises the following steps: defining the previous generation population as popt-1The current population is popt，popt-1Maximum fitness value of middle agent
Figure GDA0002367901360000044
poptMinimum fitness value of middle agent
Figure GDA0002367901360000045
Will be provided with
Figure GDA0002367901360000046
Corresponding agent replacement
Figure GDA0002367901360000047
The corresponding agent.
Preferably, the termination condition is K ═ K, where K denotes an algebraic number in which the maximum fitness value is continuously maintained; k represents a set value of an algebra in which the optimal fitness value is continuously maintained; and when K is equal to K, stopping iteration to obtain an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual.
According to the technical scheme, the burn depth inversion integration model is constructed based on the CAGA and the SVM, the complex relation between the near infrared spectrum signal and the skin burn depth is accurately established, and the accurate quantitative prediction of the skin burn depth is realized. The experimental result shows that the algorithm of the invention has high accuracy, good stability and strong generalization capability, and can accurately predict the skin burn depth.
Drawings
In order to more clearly illustrate the detailed description of the invention or the technical solutions in the prior art, the drawings that are needed in the detailed description of the invention or the prior art will be briefly described below. Throughout the drawings, like elements or portions are generally identified by like reference numerals. In the drawings, elements or portions are not necessarily drawn to scale.
Fig. 1 is a flowchart of the establishment of the CAGA-SVM integrated detection model in the present embodiment.
FIG. 2 is a chart of the near infrared spectra of different burn depths in this example.
Fig. 3 is a graph showing the relationship between burn depth and burn time in this example.
Detailed Description
Near infrared spectroscopy can effectively detect changes of skin burned tissues, but because the relationship between signals and depth is complex, a simple data analysis method is difficult to be effective. Based on this, machine learning is introduced into near infrared spectrum signal analysis of skin burns for establishing a complex relationship between near infrared spectrum signals and burn depth. Because the near infrared spectrum signal in the pigskin scald model is in direct proportion to the burn depth, the inversion aiming at the burn depth can be used for detecting the detection performance of the burn depth. Firstly, establishing a regression model of near infrared spectrum signals and burn depth by adopting a Support Vector Machine (SVM); and then, screening the SVM kernel function parameters and the optimal feature subset by adopting a chain intelligent agent genetic algorithm (CAGA), thereby forming a CAGA-SVM integrated detection model for predicting the burn depth of the sample to be predicted. The invention adopts an integrated inversion model combining the packaged CAGA and the SVM to realize the detection of the skin burn depth, and optimizes SVM kernel function parameters and feature subsets through the CAGA, thereby realizing low detection error and realizing the accurate detection of the burn depth.
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and therefore are only examples, and the protection scope of the present invention is not limited thereby. It is to be noted that, unless otherwise specified, technical or scientific terms used herein shall have the ordinary meaning as understood by those skilled in the art to which the invention pertains.
Example (b):
the burn depth prediction system based on the near infrared spectrum of the CAGA and the SVM comprises a near infrared spectrum collector, a model builder and a prediction device as shown in figure 1; wherein,
the near infrared spectrum collector collects near infrared spectrum signals in a sample to be predicted, and the near infrared spectrum signals are preprocessed and subjected to characteristic extraction and transmitted to the prediction device;
the prediction device predicts the burn depth of a sample to be predicted by utilizing the established CAGA-SVM integrated detection model according to the data uploaded by the near infrared spectrum collector;
the model builder comprises an estimation device and an optimization device, wherein the estimation device collects a plurality of burn training samples and adopts a support vector machine to build a regression model of near infrared spectrum signals and burn depths; and the optimization device adopts a chain type intelligent agent genetic algorithm to carry out global optimization on the kernel function parameters and the feature subsets of the regression model to obtain the CAGA-SVM integrated detection model.
The burn training sample selects pigskin as a sample, is taken from dead pure piglet abdominal skin, has similar photothermal characteristics and structural hierarchy to human skin, and is an ideal skin sample model. The estimation device utilizes a scalding head of a temperature control scalding instrument to flatly spread the pigskin on a test bed, different depths (the burning time is 5s, 10s, 15s, 20s, 30s, 40s, 60s, 80s, 100s and 120s, the relation between the burning depth and the burning time is shown in figure 3, the unit of the burning depth is micron) are respectively scalded on different areas of the pigskin, and the scalded pigskin at each depth point on each area forms a scalding depth training sample; a plurality of near infrared spectrum signals (more than 48 near infrared spectrum signals can be acquired for each type of burn deep training sample by using a near infrared spectrometer to form a plurality of burn training samples. Wherein the spectral measurement range is 900-2100 nm. FIG. 2 is a near infrared spectrogram of each burn depth after filtering, and the invention performs smooth filtering processing on the spectral data as sample data for subsequent experiments.
The SVM is a learning machine with strong nonlinear inversion capability and generalization capability, and has been successfully used in classification and regression in various application fields. The SVM can map features which are difficult to distinguish to a high-dimensional feature space, and establishes a nonlinear mapping relation between complex vectors through high-dimensional mapping and support vector optimization based on the principle of structural risk minimization so as to be beneficial to classification and prediction. In addition, the SVM algorithm is a convex quadratic optimization problem, and the found extreme value solution is guaranteed to be the global optimal solution.
The estimation device adopts a support vector machine to establish a regression model of near infrared spectrum signals and burn depth as follows:
Figure GDA0002367901360000071
wherein, the burn training sample is { xl,yl1, 2. m, m is the sample volume, xlThe intensity value of the near infrared spectrum signal; y islThe depth of burn; when the estimation device adopts the burn training sample to establish the regression model, y is usedlCarry in (f) (x). K (x, x)l) Is a kernel function of the support vector machine; b is a constant; a isl，
Figure GDA0002367901360000072
The sum constant b is obtained by solving the dual problem of minimizing the structural risk by adopting a Lagrange optimization method, and the Lagrange factor a is obtained in the solving processl，
Figure GDA0002367901360000073
The constraint of the maximization function is
Figure GDA0002367901360000074
Wherein C is a kernel function parameter.
The common kernel functions of the support vector machine include linear, polynomial, RBF and sigmoid, and the most common kernel functions of the linear and the radial are selected for comparison experiments.
The kernel function of the support vector machine is a linear kernel, and the expression is as follows:
K(x,xl)＝(x·xl) (3)。
the kernel function of the support vector machine is a radial basis kernel, and the expression is as follows:
K(x,xl)＝exp(-γ||x-xl||2) (4)
where γ is a kernel function parameter.
In the formula (2), C and γ in the formula (4) are parameters of the SVM kernel function. The function f in the formula (1) is completely represented by alAnd
Figure GDA0002367901360000075
determine that there are only a few a that are not 0lAnd
Figure GDA0002367901360000076
the corresponding samples are called support vectors. And the obtained regression expression (1) is the inverse model of the burn depth of the pigskin of the experiment. In order to improve the inversion accuracy, the kernel function parameters C, gamma and the characteristic subset of the chain type intelligent agent genetic algorithm are optimized.
Firstly, setting a C optimization selection interval as [1, 2000] and a gamma optimization selection interval as [0.001, 2 ]; then, an initial population is set: each population individual is identified by one binary code, different binary digits in the binary code represent different characteristics of the population individual, each characteristic in the population individual is represented by one binary, the total number of the characteristics is 256, when a certain characteristic in the population individual is selected, the binary digit corresponding to the characteristic is 1, otherwise, the binary digit is 0; each population individual corresponds to a fitness value. Defining a kernel function parameter C as a decimal number represented by 1 st-11 th binary digits of binary codes in the population individuals, and defining a kernel function parameter gamma as a decimal number represented by 12 nd-22 th binary digits of binary codes in the population individuals; bits 23 to 156 are 256 band syndrome bits.
Secondly, screening the initial population of each iteration by using a chain type intelligent agent genetic algorithm (the genetic algorithm is shown in the research on feature selection of the chain type intelligent agent genetic algorithm based on competition strategies published in the system simulation report): performing neighborhood competition selection, adaptive crossing and adaptive variation on the initial population to obtain a variation population, combining the initial population and the variation population, obtaining a new population by adopting a bubble sorting method, and taking the new population as the initial population of the next iteration; the CAGA is an improved intelligent agent genetic algorithm, has strong global optimization capability, high precision, good search precision and good stability, and can well optimize the parameters of a regression model, thereby obtaining high accuracy and low error. Assume that the current agent is Li,jThen its chain neighborhood is Neibari,j＝{Li,j1,Li,j2In which i represents on-chainThe agent's serial number, j, represents the jth agent on the ith chain. Each agent has certain energy, namely a fitness value, and the fitness is obtained through a fitness function. The arrangement is shown as the following formula:
Figure GDA0002367901360000081
l is the total number of agents on the ith chain, the neighborhood competition selection is carried out on the circular chain from left to right in sequence,
Figure GDA0002367901360000082
as the j agent in the t generation population, the neighborhood is Nbs1,j，
Figure GDA0002367901360000083
j1, 2, …, popsize, the number of individuals in a population, wherein
Figure GDA0002367901360000084
The update method of (1) is as follows:
Figure GDA0002367901360000085
wherein omicron represents competing operations among agents; if agent Lt 1,jAnd Lt 1,j1Competition, then there is Lt 1,j＝(ct j, 1ct j,2...ct j,k...ct j,length)，Lt 1,j1＝(ct j1,1ct j1,2...ct j1,k...ct j1,length)，ct j,kThe k-th gene of the agent whose position is (1, j), ct j1,kThe kth gene of the agent at position (1, j1) is shown, and length is the number of genes, i.e., the number of signatures, of a single agent. Competition operations
Figure GDA0002367901360000099
Can be expressed as follows:
Figure GDA0002367901360000091
u (0,1) represents a random number generated under a uniform distribution of the regions (0, 1).
The formula of the adaptive intersection is as follows:
Figure GDA0002367901360000092
in the formula, pcFor crossover probability, GH (j, j') is Lt 1,jAnd max1,jHaiming distance of, max1,jIs Nbs1,jThe individual with the maximum moderate fitness value f'jIs the current individual Lt 1,jThe value of the fitness value of (a) is,
Figure GDA0002367901360000093
is the maximum fitness value of the generation individual, f' is Lt 1,jAnd max1,jThe fitness value with the highest fitness value in the intermediate fitness values,
Figure GDA0002367901360000094
the average fitness value of the population individuals in the iteration is obtained;
the method for judging the self-adaptive variation comprises the following steps: when U (0,1) < pmJudging the population variation, wherein pmAs the mutation probability, pmLength is the number of genes of a single agent, i.e. the number of bits of the binary code in the initial population; p is a radical ofmRelated to chromosome length.
The method for obtaining the new population by adopting the bubbling sorting method comprises the following steps: defining the previous generation population as popt-1The current population is popt，popt-1Maximum fitness value of middle agent
Figure GDA0002367901360000095
poptMinimum fitness value of middle agent
Figure GDA0002367901360000096
Will be provided with
Figure GDA0002367901360000097
Corresponding agent replacement
Figure GDA0002367901360000098
And the corresponding agent ensures that the optimal individual fitness value of each generation is in an increasing trend.
Finally, judging whether the current iteration meets the evolution termination condition, if not, repeatedly screening the initial population of the next iteration; if so, obtaining an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual; setting k to represent an algebra in which the maximum fitness value is continuously maintained; k represents a set value of an algebra in which the optimal fitness value is continuously maintained; and when K is equal to K, stopping iteration to obtain an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual. The fitness value corresponding to the fitness function is an inversion average relative error value (ERR) obtained by the SVM aiming at the verification set under the selection of a certain C and gamma value and a feature subset. And obtaining an optimized regression model, namely a CAGA-SVM integrated detection model, based on the optimal kernel function parameters C and gamma and the optimal feature subset.
The method is based on the CAGA and the SVM, a burn depth inversion integration model is constructed, the complex relation between the near infrared spectrum signal and the skin burn depth is accurately established, and the accurate quantitative detection of the skin burn depth is realized, so that the burn depth is detected. The experimental result shows that the algorithm of the invention has high accuracy, good stability and strong generalization capability, can accurately predict the skin burn depth, and the error meets the clinical requirement.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the present invention, and they should be construed as being included in the following claims and description.

Claims (5)

1. The burn depth prediction system based on the near infrared spectrum of the CAGA and the SVM is characterized by comprising a near infrared spectrum collector, a model builder and a prediction device; wherein,
the near infrared spectrum collector collects near infrared spectrum signals in a sample to be predicted, and the near infrared spectrum signals are preprocessed and subjected to characteristic extraction and transmitted to the prediction device;
the prediction device predicts the burn depth of a sample to be predicted by utilizing the established CAGA-SVM integrated detection model according to the data uploaded by the near infrared spectrum collector;
the model builder comprises an estimation device and an optimization device, wherein the estimation device collects a plurality of burn training samples and adopts a support vector machine to build a regression model of near infrared spectrum signals and burn depths; the optimization device adopts a chain type intelligent agent genetic algorithm to carry out global optimization on kernel function parameters and feature subsets of a regression model to obtain the CAGA-SVM integrated detection model;
the estimation device adopts a support vector machine to establish a regression model of near infrared spectrum signals and burn depth as follows:
Figure FDA0002367901350000011
wherein, the burn training sample is { xl,yl1,2 … m, m being the sample volume, xlThe intensity value of the near infrared spectrum signal; y islThe depth of burn; k (x, x)l) Is a kernel function of the support vector machine; b is a constant; lagrange factor al，
Figure FDA0002367901350000012
The constraint of the maximization function is:
Figure FDA0002367901350000013
wherein C is a kernel function parameter;
the kernel function of the support vector machine is a radial basis kernel, and the expression is as follows:
K(x,xl)＝exp(-γ||x-xl||2) (4)
wherein γ is a kernel function parameter;
firstly, setting a C optimization selection interval as [1, 2000] and a gamma optimization selection interval as [0.001, 2 ]; then, an initial population is set: each population individual is identified by one binary code, different binary digits in the binary code represent different characteristics of the population individual, when a certain characteristic in the population individual is selected, the binary digit corresponding to the characteristic is 1, otherwise, the binary digit is 0; defining a kernel function parameter C as a decimal number represented by 1 st-11 th binary digits of binary codes in the population individuals, and defining a kernel function parameter gamma as a decimal number represented by 12 nd-22 th binary digits of binary codes in the population individuals; secondly, screening the initial population of each iteration by adopting a chain type intelligent agent genetic algorithm: performing neighborhood competition selection, adaptive crossing and adaptive variation on the initial population to obtain a variation population, combining the initial population and the variation population, obtaining a new population by adopting a bubble sorting method, and taking the new population as the initial population of the next iteration; finally, judging whether the current iteration meets the evolution termination condition, and if not, screening the initial population of the next iteration; if so, obtaining an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual; and obtaining an optimized regression model, namely a CAGA-SVM integrated detection model, based on the optimal kernel function parameters C and gamma and the optimal feature subset.
2. The system of claim 1, wherein the estimation device uses a scalding head of a temperature-controlled scalding instrument to scald different areas of the pigskin at different depths, wherein the scalded pigskin at each time point in each area forms a type of scalding depth training sample; and collecting near infrared spectrum signals for each type of burn depth training sample by using a near infrared spectrometer to form the burn training sample.
3. The system of claim 1, wherein the support vector machine kernel function is a linear kernel, and the expression is:
K(x,xl)＝(x·xl) (3)。
4. the system of claim 1, wherein the neighborhood competition selection is performed sequentially from left to right on a cyclic chain, and the current agent is set to
Figure FDA0002367901350000021
Its neighborhood is Nbs1,j，
Figure FDA0002367901350000022
popsize is the number of individuals in a population, wherein
Figure FDA0002367901350000023
The update method of (1) is as follows:
Figure FDA0002367901350000031
wherein,
Figure FDA0002367901350000032
representing competing operations between agents;
the formula of the adaptive intersection is as follows:
Figure FDA0002367901350000033
in the formula, pcFor the crossover probability, GH (j, j') is
Figure FDA0002367901350000034
And max1,jHaiming distance of, max1,jIs Nbs1,jThe individual with the largest fitness value fjIs a current individual
Figure FDA0002367901350000035
The value of the fitness value of (a) is,
Figure FDA0002367901350000036
is the maximum fitness value of the generation individual, f' is
Figure FDA0002367901350000037
And max1,jThe fitness value with the highest fitness value in the intermediate fitness values,
Figure FDA0002367901350000038
the average fitness value of the population individuals in the iteration is obtained;
the method for judging the self-adaptive variation comprises the following steps: when U (0,1) < pmJudging the population variation, wherein pmAs the mutation probability, pmLength is the number of genes of a single agent, i.e. the number of bits of the binary code in the initial population; u (0,1) represents a random number generated under a uniform distribution of the regions (0, 1);
the method for obtaining the new population by adopting the bubbling sorting method comprises the following steps: defining the previous generation population as popt-1The current population is popt，popt-1Maximum fitness value of middle agent
Figure FDA0002367901350000039
poptMinimum fitness value of middle agent
Figure FDA00023679013500000310
Will be provided with
Figure DEST_PATH_GDA0002367901360000046
Corresponding agent replacement
Figure FDA00023679013500000312
The corresponding agent.
5. The system of claim 4, wherein the termination condition is K ═ K, where K represents the number of generations where the maximum fitness value is continuously maintained; k represents a set value of an algebra in which the optimal fitness value is continuously maintained; and when K is equal to K, stopping iteration to obtain an optimal population individual, and obtaining optimal kernel function parameters C and gamma and an optimal characteristic subset by binary coding of the optimal population individual.
CN201611072092.XA 2016-11-29 2016-11-29 Burn depth prediction system based on near infrared spectrum of CAGA and SVM Active CN107067017B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611072092.XA CN107067017B (en) 2016-11-29 2016-11-29 Burn depth prediction system based on near infrared spectrum of CAGA and SVM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611072092.XA CN107067017B (en) 2016-11-29 2016-11-29 Burn depth prediction system based on near infrared spectrum of CAGA and SVM

Publications (2)

Publication Number Publication Date
CN107067017A CN107067017A (en) 2017-08-18
CN107067017B true CN107067017B (en) 2020-04-24

Family

ID=59618516

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611072092.XA Active CN107067017B (en) 2016-11-29 2016-11-29 Burn depth prediction system based on near infrared spectrum of CAGA and SVM

Country Status (1)

Country Link
CN (1) CN107067017B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990199B (en) * 2021-03-29 2024-04-26 杭州电子科技大学 Burn wound surface depth classification system based on support vector machine
CN118333161B (en) * 2024-06-12 2024-08-23 杭州电子科技大学 Optimization method based on multi-group competition multi-objective optimization confidence rule base

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5701902A (en) * 1994-09-14 1997-12-30 Cedars-Sinai Medical Center Spectroscopic burn injury evaluation apparatus and method
WO2016069788A1 (en) * 2014-10-29 2016-05-06 Spectral Md, Inc. Reflective mode multi-spectral time-resolved optical imaging methods and apparatuses for tissue classification

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5701902A (en) * 1994-09-14 1997-12-30 Cedars-Sinai Medical Center Spectroscopic burn injury evaluation apparatus and method
WO2016069788A1 (en) * 2014-10-29 2016-05-06 Spectral Md, Inc. Reflective mode multi-spectral time-resolved optical imaging methods and apparatuses for tissue classification

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Integrating support vector machine and genetic algorithm to implement dynamic wafer quality prediction system;Pao-HuaChou et al;《Expert Systems with Applications》;20100630;第37卷(第6期);4415-4416,4418-4419,图5 *
多类型磁共振影像特征融合算法用于阿尔兹海默病早期诊断;闫瑾;《中国优秀硕士学位论文全文数据库医药卫生科技辑》;20160615(第6期);全文 *

Also Published As

Publication number Publication date
CN107067017A (en) 2017-08-18

Similar Documents

Publication Publication Date Title
Singh et al. Feature selection of gene expression data for cancer classification: a review
US20050159896A1 (en) Apparatus and method for analyzing data
Mamani Machine Learning techniques and Polygenic Risk Score application to prediction genetic diseases
CN107067017B (en) Burn depth prediction system based on near infrared spectrum of CAGA and SVM
CN113380327B (en) Human biological age prediction and human aging degree assessment method
Zaman et al. Codon based back propagation neural network approach to classify hypertension gene sequences
Ruffieux et al. A global-local approach for detecting hotspots in multiple-response regression
CN110991494A (en) Method for constructing prediction model based on improved moth optimization algorithm
CN108427865B (en) Method for predicting correlation between LncRNA and environmental factors
CN111079074A (en) Method for constructing prediction model based on improved sine and cosine algorithm
CN116525108A (en) SNP data-based prediction method, device, equipment and storage medium
CN116172688A (en) Radiofrequency ablation damage depth prediction model and system based on support vector regression
Pyingkodi et al. Hybrid bee colony and weighted ranking firefly optimization for cancer detection from gene regulatory sequences
CN113159404A (en) Electric field seed treatment optimal parameter prediction method, system and terminal based on WOA-SVM
Khazen et al. Combinatorial expression rules of ion channel genes in juvenile rat (Rattus norvegicus) neocortical neurons
Kharis et al. Classification of lung cancer using support vector machine with feature selection based on artificial bee colony rate of change
CN112036432A (en) Spectral modeling sample set rapid partitioning method based on tabu optimization
Du et al. Sticky hidden Markov modeling of comparative genomic hybridization
CN111223522B (en) Method for identifying lncRNA based on fuzzy k-mer utilization rate
JP2013526274A5 (en)
Lahmer et al. DNA Microarray Analysis Using Machine Learning to Recognize Cell Cycle Regulated Genes
Sinha et al. A study of feature selection and extraction algorithms for cancer subtype prediction
Babatunde et al. A comparison of Boosting techniques for Classification of Microarray data
Muhammad et al. Gvdeepnet: Unsupervised deep learning techniques for effective genetic variant classification
Touchanti et al. A 2-stages feature selection framework for colon cancer classification using SVM

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
TR01 Transfer of patent right

Effective date of registration: 20221010

Address after: 313001 Building 1, No. 518, Zhongtian Road, Longxi Street, Huzhou Economic and Technological Development Zone, Huzhou City, Zhejiang Province

Patentee after: Zhejiang Depp medical Polytron Technologies Inc.

Address before: 400000 Communication Engineering College, Zone A, Chongqing University, Shapingba District, Chongqing

Patentee before: Wu Jun

TR01 Transfer of patent right