KR20220098909A - Method and System for identifying multi-radioisotope using sparse representation of gamma spectrum - Google Patents

Abstract

The present invention goes one step further from the existing nuclide identification algorithm and converts a spectrum signal acquired from a radiation detector mounted on an unmanned device into a nuclide identification sparse signal that is easy to transmit and store. And to the system, there is an effect of facilitating the application of the nuclide identification algorithm in an environment using a micro controller unit (MCU), which has significantly poorer CPU and memory performance compared to a PC.

Description

Method and System for identifying multi-radioisotope using sparse representation of gamma spectrum

The present invention relates to a method and system for determining polynuclides through sparse expression of a gamma spectrum.

In the event of radioactive disasters such as the Chernobyl nuclear accident in 1986 and the Fukushima nuclear accident in 2011, it is necessary to quickly detect the characteristics of the leaked radioactive material in order to minimize environmental pollution and human damage.

In general, a high purity Ge detector (HPGE), which is used for accurate analysis of nuclides, involves processes such as sample collection and sample pretreatment prior to analysis, so it is difficult to quickly characterize it.

Recently, unmanned devices used in various fields, such as drones, are expected to be effective in rapidly detecting radioactive materials in areas that are difficult for humans to access at the site of a nuclear accident in which a high-radioactive material leaked.

Accordingly, the need for real-time on-site measurement technology development through the mounting of an unmanned device of the radiation detector is increasing.

The scintillation detector is widely used as a detector for on-site measurement due to its relatively low price, ease of manufacture, and high portability, so it is easy to mount unmanned devices. However, the low energy resolution characteristic of the scintillator detector makes it difficult to discriminate nuclides having similar gamma-ray energies. To overcome these physical properties, the scintillator detector must include a radioisotope identification algorithm.

In existing studies, the presence or absence of radioactive materials is determined using the measured counting number and library comparison, or special nuclear materials and naturally occurring radioactive materials are determined through a region of interest method. materials) are separated.

In addition, the nuclide identification algorithm through feature learning in the gamma spectrum based on a neural network overcomes the shortcomings that the aforementioned existing methods are limited to a single nuclide, so it is possible to distinguish and the spectrum measurement time must be long enough. It has shown the ability to classify polynuclides in However, these methods did not take into consideration that the CPU (central processing unit) and memory resources are limited to be applied to an unmanned device environment.

Accordingly, there is a need to construct an easier nuclide identification algorithm for efficient signal transmission and storage under limited conditions.

Patent Registration Publication No. 10-1918241 (Announcement Date: 2018. 11. 13) Patent Registration Publication No. 10-1680067 (Announcement Date: 2016. 11. 29)

An object of the embodiments of the present invention for improving the above-described problems is to go one step further from the existing nuclide identification algorithm and convert a spectrum signal obtained from a radiation detector mounted on an unmanned device into a gamma signal that is easy to transmit and store. It is to provide a method and system for determining polynuclides through sparse expression of a spectrum.

In order to achieve the above object, the method for determining multi-nuclide through sparse expression of a gamma spectrum according to an embodiment of the present invention is training data for specifying a label for each element included in a gamma spectrum obtained from a target. preparatory stage; A pre-learning step of learning a dictionary through singular vector deposition (SVD) for each labeled element; and a nuclide classification step of obtaining a sparse vector through a sparse representation of the learned dictionary, and discriminating the nuclide as a label corresponding to the term having the largest element based on the result of the dot product of the sparse vector with a classifier. characterized by including.

The nuclide classification step is characterized in that the sparse expression is used to extract distinguishable features from a gamma spectrum corresponding to each class among a plurality of classes.

The sparse representation is characterized in that the input signal is expressed as a linear combination of base elements constituting the dictionary.

The classifier is characterized in that the nuclide discrimination is performed on each component of the sparse vector of the input signal.

In addition, the multi-nuclide discrimination system through sparse expression of the gamma spectrum according to another embodiment of the present invention includes: a training data preparation unit for designating a label for each element included in a gamma spectrum obtained from a target by a radiation detector; a dictionary learning unit for learning a dictionary through singular vector deposition (SVD) for each labeled element; and obtaining a sparse vector through a sparse representation of the dictionary learned in the dictionary learning unit, and discriminating the nuclide as a label corresponding to the term having the largest element based on the result of the dot product of the sparse vector with a classifier. It is characterized in that it includes a nuclide classification unit.

The nuclide classifying unit includes: a sparse vector generator for obtaining a sparse vector through a sparse representation of the dictionary learned by the dictionary learning unit; and a nuclide discrimination unit for discriminating a nuclide as a label corresponding to a term having the largest element based on a result of dot product of the sparse vector with a classifier.

A method and system for determining multiple nuclide through sparse expression of a gamma spectrum according to an embodiment of the present invention go one step further from the existing nuclide identification algorithm and transmit and store a spectrum signal acquired from a radiation detector mounted on an unmanned device. Since it is converted into a rare signal for easy nuclide identification, it is easy to apply the nuclide identification algorithm in an environment using a micro controller unit (MCU), which has significantly poorer CPU and memory performance compared to a PC.

)에서 보이는 각 항의 계수(α, β, γ) 그리고 희소 벡터의 유효 원소 개수(T)의 최적화 과정을 통해 핵종 판별의 정확도를 향상시킬 수 있다.In addition, an embodiment of the present invention can be used not only for organic scintillation detectors but also for energy spectrum-based radiation detectors such as inorganic scintillators and semiconductor detectors, and in the pre-learning step, the number of elements in the dictionary ( K ), Equation 6 (

), it is possible to improve the accuracy of nuclide discrimination through the optimization process of the coefficients (α, β, γ) of each term and the number of effective elements (T) of the sparse vector.

1 is a block diagram schematically illustrating a system for determining polynuclides through sparse representation of a gamma spectrum according to an embodiment of the present invention.
2 is a flowchart schematically illustrating a discrimination method using a multinuclide discrimination system through sparse representation of a gamma spectrum according to an embodiment of the present invention.
3 is a graph showing measured and simulated spectra for each nuclide in Table 1 using a discrimination method using a multi-nuclide discrimination system through sparse expression of a gamma spectrum according to an embodiment of the present invention.
4 is a graph showing the sparse representation of a measured spectrum according to Table 2 using a discrimination method using a multinuclide discrimination system through sparse representation of a gamma spectrum according to an embodiment of the present invention.
5 is a graph showing the accuracy of nuclide classification in advance according to the measurement time and source location according to Table 2 using the discrimination method using the multi-nuclide discrimination system through sparse expression of the gamma spectrum according to an embodiment of the present invention. .

The present invention as described above will be described in detail with reference to the accompanying drawings and embodiments.

It should be noted that the technical terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. In addition, the technical terms used in the present invention should be interpreted as meanings generally understood by those of ordinary skill in the art to which the present invention belongs, unless otherwise defined in the present invention, and excessively comprehensive It should not be construed as a human meaning or in an excessively reduced meaning. In addition, when the technical term used in the present invention is an incorrect technical term that does not accurately express the spirit of the present invention, it should be understood by being replaced with a technical term that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted as defined in advance or according to the context before and after, and should not be interpreted in an excessively reduced meaning.

Also, the singular expression used in the present invention includes the plural expression unless the context clearly dictates otherwise. In the present invention, terms such as "consisting of" or "comprising" should not be construed as necessarily including all of the various elements or several steps described in the invention, and some of the elements or some steps are included. It should be construed that it may not, or may further include additional components or steps.

In addition, although the terms used in the present invention may be used to describe the components, the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of reference numerals, and redundant description thereof will be omitted.

In addition, in the description of the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention, and should not be construed as limiting the spirit of the present invention by the accompanying drawings.

1 is a block diagram schematically showing a multinuclide discrimination system through sparse representation of a gamma spectrum according to an embodiment of the present invention, and FIG. 2 is a sparse representation of a gamma spectrum according to an embodiment of the present invention. It is a flowchart schematically showing a discrimination method using a multinuclide discrimination system, and FIG. 3 is each nuclide in Table 1 using a discrimination method using a multinuclide discrimination system through sparse expression of a gamma spectrum according to an embodiment of the present invention. is a graph showing the measured and simulated spectrum for , and FIG. 4 is a sparse expression of the measured spectrum according to Table 2 using the discrimination method using the multinuclide discrimination system through sparse expression of the gamma spectrum according to an embodiment of the present invention is a graph showing, and FIG. 5 is a prior classification of nuclides according to measurement time and source location according to Table 2 using a discrimination method using a multi-nuclide discrimination system through sparse expression of a gamma spectrum according to an embodiment of the present invention. This graph shows the accuracy.

As shown in FIG. 1 , the multi-nuclide identification system through sparse expression of the gamma spectrum according to an embodiment of the present invention is obtained from the radiation detector 10 mounted on the unmanned device, going one step further from the existing nuclide identification algorithm. It is a system that converts a spectral signal into a nuclide discrimination sparse signal that is easy to transmit and store.

To this end, the multi-nuclide discrimination system through sparse representation of the gamma spectrum according to an embodiment of the present invention includes a learning data preparation unit 100 , a pre-learning unit 200 , and a nuclide classification unit 300 .

The training data preparation unit 100 specifies a label for each element included in the gamma spectrum obtained from the target by the radiation detector 10 .

On the other hand, a gamma-ray spectrum can be obtained by the radiation detector 10 (eg, a gamma-ray detector). This may be, for example, a thallium-doped sodium iodide (NaI(Tl))-based gamma-ray detector. The gamma ray detector may optionally be based on other materials such as high purity germanium (HPGe), cadmium telluride (CdTe), cadmium zinc telluride (CZT) and lanthanum bromide (LaBr). NaI(Tl)-based detectors can be used in thallium-doped sodium iodide-based spectral radioactivity detectors (RPMs) in boundary observation applications. The raw obtained gamma ray signals (either the target or reference samples) may be passed to a signal amplifier to amplify the signals. The (amplified) gamma-ray signals (one of the target or reference samples) can be passed to a multi-channel analyzer, which divides the signals into many bins (or energy ranges). The bin values are collectively referred to as a spectrum. The bins represent the smallest increase in the energy interval of the gamma-ray spectrum to which the coefficients contribute. Typically, the multi-channel analyzer will generate values within about 1,024 data bins, but depending on the analyzer this number may be more or less, e.g., 128 bins to 16,384 bins, 128 bins to 512 bins, 512 bins to 512 bins. There may be 2,048 bins, 2,048 bins to 8,192 bins, 8,192 bins to 16,384 bins, 512 bins to 4,096 bins, or 256 bins to 8,192 bins. The number of bins may advantageously be equal to an integrated power of two. Typically, the bins cover energy values in the range of 40 keV to 3,000 keV, but these endpoints can be large or small depending on the analyzer, for example 30 keV and 2,700 keV, respectively. The range is 30 keV to 2,700 keV, 35 keV to 2,700 keV, 40 keV to 2,700 keV, 30 keV to 3,000 keV, 35 keV to 3,000 keV, 40 keV to 3,000 keV, 30 keV to 4,000 keV, 35 keV to 4,000 keV and 40 keV to 4,000 keV can be

In addition to the gamma ray detector, the radiation detector 10 may be a spectroscopic portal detector arranged to acquire a gamma spectrum from a vehicle passing through a detection area.

The dictionary learning unit 200 learns a dictionary through singular vector deposition (SVD) for each labeled element.

The nuclide classifying unit 300 obtains a sparse vector through a sparse representation of the dictionary learned in the dictionary learning unit 200, and the term having the largest element based on the result of the dot product of the sparse vector with the classifier. A nuclide is identified as a label corresponding to

To this end, the nuclide classifying unit 300 includes a sparse vector generator (not shown) that obtains a sparse vector through a sparse representation of the dictionary learned in the dictionary learning unit 200, and divides the sparse vector into a classifier and a dot product. A nuclide discrimination unit for discriminating a nuclide as a label corresponding to the term having the largest element based on one result may be included.

To describe a method for discriminating polynuclides using a system for discriminating polynuclides through sparse expression of a gamma spectrum according to an embodiment of the present invention configured as described above, as shown in FIG. 2 , gamma obtained from a target A learning data preparation step (S10) of specifying a label for each element included in the spectrum, and a dictionary learning a dictionary through singular vector deposition (SVD) for each labeled element In the learning step (S20), obtaining a sparse vector through a sparse representation of the learned dictionary, and discriminating the nuclide as a label corresponding to the term with the largest element based on the result of the dot product of the sparse vector with the classifier and a nuclide classification step (S30).

The nuclide classification step (S30) uses a sparse expression to extract distinguishable features from a gamma spectrum corresponding to each class among a plurality of classes. means to express

Meanwhile, the classifier 20 performs nuclide discrimination for each component of the sparse vector of the input signal.

Hereinafter, a method for discriminating polynuclides through sparse expression of a gamma spectrum according to an embodiment of the present invention driven in the above order will be described in more detail.

First, the present invention relates to a method capable of spectral compression and nuclide classification through a sparse representation of a gamma spectrum based on dictionary learning.

In general, since a signal is composed of high-dimensional overlapping data, it is difficult to extract valid information from it. For this reason, effective information is extracted by reducing the dimension of a multi-dimensional signal through dimension reduction and removing unnecessary noise and redundant signals.

The sparse representation used in the present invention aims to express an input signal as a linear combination of base elements (atoms) constituting a dictionary as a method of dimensionality reduction. This can be expressed by Equation 1 below.

[Equation 1]

은 p 차수의 norm을 의미한다. Norm은 벡터의 길이 혹은 크기를 측정하는 함수로 norm이 측정한 벡터의 크기는 원점에서 벡터 좌표까지의 거리이다. 행렬 S에 대한 p 차수 norm은 다음 수학식 2와 같이 표현된다.here

is the norm of the p order. Norm is a function that measures the length or magnitude of a vector. The magnitude of the vector measured by norm is the distance from the origin to the vector coordinates. The p-order norm for the matrix S is expressed by Equation 2 below.

[Equation 2]

Y is an input signal composed of N spectra composed of M windows and is described by Equation 3 below.

[Equation 3]

D is a dictionary matrix composed of K base elements, and is expressed by Equation (4) below.

[Equation 4]

X is a sparse vector for the input signal Y , and all elements except for T non-zero sparse coefficients are 0, which is expressed by Equation 5 below.

[Equation 5]

It can be seen from Equation 1 that how to construct the dictionary D is very important to intuitively obtain the vector X obtained by sparsely expressed the input signal Y.

In the present invention, a prior learning technique is used to obtain a discriminative dictionary for efficient nuclide classification as well as sparse representation of the input spectrum. In order for the sparse vector of the input spectrum to be able to discriminate the nuclide, a label is assigned to each element constituting the learning sample in the pre-learning step.

Pre-learning is expressed as in Equation 6 below.

[Equation 6]

Each of the three terms means reconstruction error, discriminative sparse vector error, and classification error, and α, β, and γ are coefficients that adjust the contribution of each term. The linear transformation matrix A of the second term serves to suppress input signals having different labels to have distinct sparse vectors and is expressed as Equation 7 below.

[Equation 7]

The label matrix H is composed of m types of labels, meaning labels for each of the N input spectra, and is expressed by Equation (8).

[Equation 8]

The classifier W serves as a nuclide discrimination for each component of the sparse vector X of the input signal Y , and is expressed by Equation (9).

[Equation 9]

Equation 6 is expressed as the following Equation 10 through the K-SVD method.

[Equation 10]

=

=

및

은

,

을 의미한다.

은 각각

에서 0을 갖는 원소를 제거한 행렬을 의미한다.

은 각각

,

을 의미한다.

은 다음 수학식 11에서 확인할 수 있다.

and

silver

,

means

is each

It means a matrix from which elements with 0 are removed.

is each

,

means

can be found in Equation 11 below.

[Equation 11]

은 사전 행렬 D의 K번째 열 및 그에 상응하는 희소 벡터 X의 K번째 열,

을 제외한 오차 행렬이다.

은

크기의 행렬로, (

) 번째 항을 제외한 모든 항이 0을 갖는다.

은 수학식 12로 표현된다.in other words,

is the Kth column of the dictionary matrix D and the corresponding Kth column of the sparse vector X ,

It is an error matrix excluding .

silver

as a matrix of size, (

), all terms except the 1st term have 0.

is expressed by Equation (12).

[Equation 12]

은 특이값 분해(singular vector decomposition, SVD)를 통해 다음 수학식 13)과 같은 결과로 학습된다.In Equation 10 described through the above process, each element in the dictionary,

is learned through singular vector decomposition (SVD) as a result of Equation 13) below.

[Equation 13]

,

,

Based on the learned dictionary, a test sample obtains a sparse vector through Equation 1, and as a result of dot product with the classifier W , the nuclide is classified as a label corresponding to the term having the largest element.

As shown in FIG. 2 , the above process may be largely composed of three steps: a training data preparation step ( S10 ), a pre-learning step ( S20 ), and a nuclide classification step ( S30 ).

The above-described methods may be implemented in one or more software applications executable in a computer system. In particular, the steps of the methods are affected by instructions of software being executed within a computer system. The software instructions may each be formed into one or more code modules for performing one or more specific tasks. The software may also be divided into two separate parts, wherein a first part and corresponding code modules perform the methods, and a second part and corresponding code modules are a user between the first part and the user. Manage interfaces.

The software is usually loaded into a computer system from a computer readable medium, then typically stored in a hard disk drive (HDD) or memory, after which the software can be executed by the computer system. In some examples, the application programs may be provided to the user encoded on one or more CD-ROMs and read through a corresponding drive prior to being stored in memory. Optionally, the software may be read by the computer system from networks, or it may be loaded into the computer system from another computer readable medium. Additionally or alternatively, data, eg, reference spectra used to prepare the dictionary data, may be stored in memory, from a CD or other computer readable medium, or on the Internet or by some other means. can be mounted. Computer-readable storage medium refers to any storage medium that participates in providing instructions and/or data to a computer system for execution and/or processing. Examples of such storage media are floppy disks, magnetic tape, CD-ROM, hard disk drive, ROM or integrated circuits, USB memory, magneto-optics, whether or not such devices are internal or external to a computer module. disks, or computer readable cards such as PCMCIA or the like. Further, examples of computer-readable transmission media that may participate in the provision of software, applications, instructions and/or data to the computer module include wireless or infrared transmission channels as well as networks for other computer or networked devices. connections, and the Internet or intranets containing e-mail communications and information recorded on web sites and the like.

The second portion of the application programs described above and the corresponding code module may be executed to implement one or more graphical user interfaces (GUIs) presented or otherwise presented on a display. Through normal manipulation of a keyboard and mouse, a user of a computer system and application interfaces in a functionally applicable manner to provide control commands and/or input for applications related to the graphical user interface (GUI)(s). can be manipulated. Other types of functionally applicable user interfaces may also be implemented, such as an acoustic interface utilizing voice prompt output via loudspeakers and user voice command input via a microphone.

Experimental example

A 2-inch organic scintillator (EJ-200) detector was used to verify the sparse expression-based gamma nuclide identification method presented in the present invention.

Since the organic scintillator is composed of a low atomic number material, a photoelectric peak is less generated, and thus it is the most difficult material to distinguish nuclides.

In other words, validating this method using an organic scintillation detector having the worst energy resolution among energy spectrum-based radiation detectors means that validation of other detectors is also valid.

[Table 1] shows 16 classes consisting of four radionuclides Ba-133, Na-22, Cs-137, Co-60 and natural radiation (background, BG) for verification of the proposed method.

Class #(Label #) isotopes Remarks Background ¹³³ Ba ²² Na ¹³⁷ Cs ⁶⁰ Co One O - - - - No RI 2 O O - - - RI (#1) 3 O - O - - 4 O - - O - 5 O - - - O 6 O O O - - RI (#2) 7 O O - O - 8 O O - - O 9 O - O O - 10 O - O - O 11 O - - O O 12 O O O O - RI (#3) 13 O O O - O 14 O O - O O 15 O - O O O 16 O O O O O RI (#4)

Here, Class #1 is the case where no artificial radioactive material is present and only natural radiation contributes to the spectrum. Also, class #2-5 is the case where a single nuclide is present. Classes other than those correspond to various combinations of each nuclide.

A pre-learning step was performed for sparse representation, which means extracting distinguishable features from the spectrum corresponding to each class.

In order to obtain various learning samples for each class, the spectrum was simulated through MCNP6 (Monte Carlo N-Particle) simulation based on Monte Carlo theory as well as experimental measurements.

As shown in FIG. 3 , it can be confirmed that the spectrum obtained by measuring the gamma rays emitted from the used nuclide and the simulated spectrum are similar. Based on this, 100 spectra for each class were simulated.

[Table 2] shows the configuration of the two dictionaries trained data and test data. The difference between Type 1 and Type 2 is the presence or absence of a measurement spectrum in the pre-training data. This difference can confirm whether the sparse vector of the input spectrum shows high nuclide classification performance even with a dictionary learned only from the simulated spectrum, and at the same time, the influence of the prior learning of the measured data can be seen.

Type 1 Type 2 learning data testing data learning data testing data MCNP o - MCNP o - Experiment o(only for class1) o Experiment o o # of spectra per class 100 600 # of spectra per class 110 600 Total # of spectra 1,600 9,600 Total # of spectra 1,760 9,600

A Type 1 dictionary is a dictionary trained only with the simulated spectrum. However, only class #1 used measurement data.

Type 2 is a dictionary trained by adding 10 measurement data for each label to the simulated data used for Type 1 dictionary learning.

The performance of Type 1 and 2 dictionary learned using these training data was evaluated with test data consisting of 200 spectra for each label in source environments at different locations (5, 10, and 20 cm).

As described above, the present invention proposes a method of converting a spectrum signal obtained from a radiation detector mounted on an unmanned device into a rare nuclide identification signal that is easy to transmit and store, going one step further from the existing nuclide identification algorithm.

By using this, it becomes easy to apply the nuclide identification algorithm in an environment using MCU (micro controller unit), which has significantly poorer CPU and memory performance compared to PC.

It can be used not only for organic scintillation detectors but also for energy spectrum-based radiation detectors such as inorganic scintillators and semiconductor detectors. , γ) and the optimization process of the number of effective elements (T) of the sparse vector can improve the accuracy of nuclide discrimination.

4 shows a vector value sparsely expressed using a measured spectrum and a dictionary in which it is learned.

Here, (a) shows the measurement data for each class, and (b) shows the sparse vector calculated by Type 2.

Unlike the measurement spectrum, which has a value for each energy window, only the number of effective elements (T) set in the pre-learning stage has a non-zero value.

Such a sparse signal is more advantageous than the original signal in bit loss occurring in the transmission process, and the storage space occupied in the memory is also significantly less.

5 shows the accuracy of nuclide discrimination according to the change in measurement time under various source location conditions.

Here, (a) shows the accuracy of Type 1, and (b) shows the accuracy of Type 2.

As in (a), the sparse representation using the dictionary (Type 1) learned only from the simulated spectrum shows an accuracy of over 92%.

It was confirmed that even for sources that are difficult to measure due to the short half-life and the Nuclear Non-Proliferation Treaty, it is possible to classify nuclides with high accuracy by learning a dictionary through computational simulation.

As shown in (b), the learned dictionary (Type 2) including the measurement spectrum showed more than 95% accuracy when the measurement time was more than 2 seconds, and showed high accuracy like the existing neural network-based nuclide identification algorithm.

In the above, preferred embodiments according to the present invention have been shown and described. However, the present invention is not limited to the above-described embodiments, and without departing from the gist of the present invention appended in the claims, any person skilled in the art to which the invention pertains will be able to implement various modifications. .

10: radiation detector 20: classifier
100: training data preparation unit 200: pre-learning unit
300: nuclide classification unit

Claims (6)

a training data preparation step of designating a label for each element included in a gamma spectrum obtained from a target;
A pre-learning step of learning a dictionary through singular vector deposition (SVD) for each labeled element; and
Obtaining a sparse vector through a sparse representation of the learned dictionary, and discriminating a nuclide as a label corresponding to the term having the largest element based on the result of the dot product of the sparse vector with a classifier. A method for discriminating polynuclides through sparse expression of a gamma spectrum, characterized in that
According to claim 1,
The nuclide classification step uses the sparse expression to extract distinguishable features from a gamma spectrum corresponding to each class among a plurality of classes.
According to claim 1,
The sparse representation is a method for determining polynuclides through sparse representation of a gamma spectrum, characterized in that the input signal is expressed as a linear combination of base elements constituting the dictionary.
According to claim 1,
Wherein the classifier performs nuclide discrimination with respect to each component of the sparse vector of the input signal.
a training data preparation unit for designating a label for each element included in a gamma spectrum obtained from a target by a radiation detector;
a dictionary learning unit for learning a dictionary through singular vector deposition (SVD) for each labeled element; and
A sparse vector is obtained through a sparse representation of the dictionary learned in the dictionary learning unit, and a nuclide is determined as a label corresponding to the term having the largest element based on the result of the dot product of the sparse vector with a classifier. Multinuclide discrimination system through sparse expression of gamma spectrum, characterized in that it includes a classification unit.
6. The method of claim 5,
The nuclide classification unit
a sparse vector generator for obtaining a sparse vector through a sparse representation of the dictionary learned by the dictionary learning unit; and
and a nuclide discrimination unit for discriminating a nuclide as a label corresponding to a term having the largest element based on a result of dot product of the sparse vector with a classifier.

Images