Systems and methods for identifying isotopes

Abstract

The present invention relates to a method for identifying isotopes, and more particularly to a computer-implemented method and corresponding system for determining the presence of a radioactive source.

Description

[Technical field]

[0001] The present invention relates to a method for identifying isotopes, and more particularly to a computer-implemented method and corresponding system for determining the presence of a radioactive source. [Background technology]

[0002] During the past decade, there has been an increasing need to shield nuclear material or radioactive sources and prevent threats to the safety of the general public. The deployment of inspection systems and portable monitoring services using passive gamma and neutron radiation detectors provides a useful layer of protection to public safety. However, the goal of improving the detection efficiency and accuracy of such detectors requires the system to meet the conflicting demands of having high sensitivity for detecting weak radioactive sources while minimizing susceptibility to false alarms, which can be time consuming and expensive.

[0003] There are several problems faced by developers in the radiation detector industry. First, environmental noise and signal attenuation can hinder detection accuracy. For example, when a detector operates on a street, walls and materials within a building will significantly attenuate radiation, thus affecting the detector's ability to identify radioactive sources. Furthermore, this problem must be compensated for in real time for live detection. Other factors such as calibration shifts, masking sources including Naturally Occurring Radiation Materials (NORM) and medical isotopes, as well as a wide variety of source strengths, can complicate and hinder detection even further.

[0004] Various methods are used to detect and identify radioactive sources. The simplest method is to only consider the gross counts of gamma rays from a radioactive source by ignoring the spectral information and to alert the user when the total exceeds a count threshold. However, there are more advanced methods, such as several learning algorithms, that can incorporate and use the entire radiation spectrum information. Very recent machine learning methods work exclusively with spectral data, for example simulated in [1], as input to the detection algorithm. However, these spectra cannot fully reproduce the complex characteristics that appear in real-world situations, which can lead to low detection accuracy in systems deployed with these methods.

[0005] The present invention has been devised to alleviate or overcome at least some of the problems set forth above. [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] M. Kamuda and C.J. Sullivan, "An automated isotope identification and quantification algorithm for isotope mixtures in low-resolution gamma-ray spectra," Radiation Physics and Chemistry, vol. 155, pp. 281-286, 2019 Summary of the Invention [Problem to be solved by the invention]

[0007] A system and method for identifying isotopes is provided. [Means for solving the problem]

[0008] According to a first aspect of the present invention there is provided a method for isotope identification performed by a computing device, the method comprising the steps of collecting a plurality of spectral data from a spectroscopic apparatus, generating, by a processor of the computing device, a plurality of data sets based on the plurality of spectral data, and determining, by the processor, a plurality of isotope probabilities, each isotope probability of the plurality of isotope probabilities corresponding to a respective data set of the plurality of data sets, the isotope probabilities indicating a probability that a corresponding data set of the plurality of data sets corresponds to a particular isotope, the method further comprising identifying, by the processor, an isotope based on the plurality of isotope probabilities.

[0009] The spectroscopic device can be any device suitable for collecting real-world isotope spectra. For example, the spectroscopic device can be a scintillation counter configured to detect and measure the intensity and / or energy of ionizing radiation emitted by the isotope and incident on the scintillation counter. Optionally, the spectral data is filtered to match the characteristic resolution of the spectroscopic device to reduce artifacts produced by statistical noise.

[0010] Those skilled in the art will appreciate that the term "isotope probability" refers to the probability that a data set contains a particular isotope, e.g. 137 It will be understood that this means the probability corresponding to Cs. Gamma rays with different energies emitted by the isotopes during nuclear decay are detected by a spectroscopic device. The spectroscopic device provides a plurality of spectral data to a computing device, where the spectral data can be gamma ray energies detected by the spectroscopic device. Machine learning based techniques are used to classify or identify the isotopes associated with the gamma rays. The invention thus provides a method for identifying and classifying individual isotopes. The method also finds use in multi-label cases, where an isotopic mixture comprising multiple isotopes becomes classified. The method is advantageous as it is robust to noise and classifies isotopes with improved accuracy.

[0011] Preferably, each dataset of the plurality of datasets comprises binned data, said dataset representing a functional association between a first bin of the binned data and a second bin of the binned data. In particular, the spectral data collected from the spectroscopic device is grouped according to a plurality of intervals (i.e., bins), and each interval is assigned a value. Each bin may, for example, be a particular energy range of gamma rays received at the spectroscopic device. An example bin interval is 10 3 ~10 4 eV. A bin count is associated with each bin, which indicates the number of gamma rays captured by the spectrometer within the bin interval.

[0012] In a preferred embodiment, the multiple data sets each comprise a respective bin ratio vector. Those skilled in the art will appreciate that the term "bin ratio vector" is a vector in which each entry is a direct proportion between a pair of bin counts for different bins. The use of a bin ratio vector is advantageous because it improves classification accuracy, especially when, for example, there is a large range of gamma rays detected. In particular, different spectra collected for the same isotope may have bins with significantly different values. The differences in values ​​may be due to variations in source strength, position, or background rate. The degree of similarity between the spectra may be affected by this partial matching. The bin ratio vector can capture the relative sizes of peaks in the spectra, which in turn improves the robustness of the method.

[0013] In some embodiments, generating the plurality of data sets comprises bin-sorting, by the processor, the plurality of spectral data into n spectral bins; generating, by the processor, a plurality of bin ratio vectors; and storing, by the processor, the respective bin ratio vectors in each dataset of the plurality of datasets.

[0014] Preferably, the step of generating the plurality of bin ratio vectors comprises the steps of: generating, by a processor, a histogram based on the n spectral bins; generating, by the processor, a histogram ratio matrix M based on the histogram, where the histogram ratio matrix M is a square matrix of length n; and extracting, by the processor, the plurality of bin ratio vectors from the histogram ratio matrix, where there are k bin ratio vectors, k=n-1. In particular, the histogram ratio matrix M is a square matrix of length n, where each element is a ratio between a pair of bin counts. Those skilled in the art will understand that the histogram ratio matrix M means a bin ratio matrix, where each element is a direct proportion between a pair of bin counts.

[0015] Preferably, each entry M of the histogram ratio matrix M i,j is the ith bin divided by the jth bin. Thus, the ratios between bin counts are represented in the histogram ratio matrix m.

[0016] Preferably, an entry of the yth bin ratio vector is a bin ratio of a first bin to a second bin, the second bin being y bins away from the first bin, such that each bin ratio vector comprises entries that are different ratios of the bin counts.

[0017] In an alternative embodiment, the bin data comprises a bin difference, i.e., the bin data comprises the difference between a first bin count and a second bin count. Preferably, the isotope probability is determined by using an ensemble, said ensemble comprising a plurality of classifications generated by a multi-class classifier, each classification corresponding to a respective data set. The term "multi-class classifier" will be understood by those skilled in the art to mean a classification model configured to classify a data set into one of a plurality of classifications. The classification indicates an isotope. The term "ensemble" will be understood by those skilled in the art to mean a collection of classification models. The multi-class classifier can be one or more selected from, but not limited to, a k-Nearest Neighbours (k-NN) classification method, a Logistic Regression (LR) classification method, a Linear Discriminant Analysis (LDA) classification method, a Support Vector Machine (SVM) classification method, an Artificial Neural Networks (ANN) classification method, and a Convolution Neural Network (CNN) classification method. Those skilled in the art will understand that any combination of classification methods may be used. Furthermore, one skilled in the art will appreciate that since k-NN, LR, LDA, and SVM functions only support binary classification, for the multi-label case, additional binary association methods are applied to transform them into multiple separate binary problems.

[0018] In some embodiments, the ensemble is an artificial neural network (ANN) ensemble, the ANN ensemble comprising a plurality of ANN classifications, each generated by an ANN, each ANN classification corresponding to a respective data set. Those skilled in the art will appreciate that the term "ANN classification" refers to an isotope classification or discrimination determined by an ANN. Thus, with an ANN ensemble, multiple ANNs are trained rather than a single ANN. The present invention is advantageous because it provides improved generalization capabilities. It is further advantageous because with the present invention, prediction variance is reduced.

[0019] In some embodiments, the isotope probability is determined based on a plurality of ANN classifications. Preferably, each ANN comprises one or more parameters, the parameters being one or more selected from a range of: a hidden layer comprising a certain number of neurons; a learning rate; and a regularization strength.

[0020] The parameters are preferably selected by applying a random search method. The multi-class classifier is preferably optimized.

[0021] In some embodiments, the multi-class classifier is optimized through backpropagation, where the multi-class classifier is configured to minimize a cost function with respect to the network weights.

[0022] The cost function is preferably the cross-entropy loss function for single isotope classification. In some embodiments, the cost function is a mean squared error cost function for isotopic mixture classification.

[0023] In some embodiments, the cost function comprises a penalty term, said penalty term being an L2 regularization factor. In an alternative embodiment, the multi-class classifier is optimized according to a scaled conjugate gradient method.

[0024] The multi-class classifier is preferably trained until a threshold difference between the predicted output and the correct output is reached. Preferably, the isotopes are identified by ensemble prediction.

[0025] In some embodiments, the ensemble prediction comprises determining, for each dataset of the multiple datasets, the isotope having the highest probability and identifying the isotope with the greatest number of highest probabilities.

[0026] It will be understood that any feature described herein as suitable for incorporation into one or more aspects or embodiments of the present disclosure is intended to be generalizable across any and all aspects and embodiments of the present disclosure. Other aspects of the present disclosure may be understood by those skilled in the art in light of the description of the present disclosure, the claims, and the drawings. The foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to limit the claims.

[0027] One or more embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings. [Brief description of the drawings]

[0028] [Figure 1] Diagram of a system for identifying isotopes. [Diagram 2] FIG. 2 is a diagram of a method for identifying isotopes using the system of FIG. 1. [Diagram 3] A diagram of a histogram ratio matrix showing the extraction of the bin ratio vector. [Figure 4] Schematic diagram of the artificial neural network architecture. [Diagram 5] Diagram of how to train an artificial neural network model. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0029] 1 is a system 100 for identifying isotopes. The system 100 includes a computing device 102 in communication with a spectrometer 104. The spectrometer 104 is configured to receive gamma radiation from the isotope source and measure the spectral content (i.e., spectral data) from the isotope source. In this example, the spectrometer 104 is a scintillation counter 104 configured to detect and measure the intensity and / or energy of ionizing radiation emitted by the isotope and incident on the scintillation counter. For example, the scintillation counter 104 may be configured to detect and measure the intensity and / or energy of ionizing radiation emitted by the isotope and incident on the scintillation counter, such as, but not limited to, the isotope. 57 Co 60 Co, 67 Ga, and 131 The scintillation counter 104 may measure energies indicative of any one or more of the isotopes I and I. Those skilled in the art will appreciate that this list of isotopes is by way of example only, and that the scintillation counter 104 may measure energies indicative of any known isotope.

[0030] The computing device 102 is configured to receive the spectral data from the spectrometer 104 and perform an artificial neural network (ANN) ensemble model generation method 200 and an isotope identification method 500 for identifying isotopes.

[0031] Prior to execution of the isotope discrimination method 500, an artificial neural network (ANN) ensemble model is generated or pre-trained using the isotope discrimination system 100 according to the training method 200 shown in FIG.

[0032] In a first step 202, a spectrometer detects spectra from gamma rays emitted by a number of known isotopes over a period of time so that a spectral count can be determined. One skilled in the art will understand that the term "known isotopes" means isotopes that are known to a user. There may be any number of known isotopes used for this training process, such as 18 isotopes.

[0033] At step 204, the computing device 102 collects or receives training spectral data for each of a plurality of known isotopes from the spectrometer 104. The training spectral data corresponds to a detected spectrum of gamma rays received from the known isotopes. The spectral data represents energies associated with the gamma rays received and measured by the spectrometer 104. Thus, a gamma ray spectrum is generated.

[0034] Optionally, the spectral data may be filtered to match the characteristic resolution of the spectrometer 104 to reduce artifacts generated by statistical noise. Further optionally, randomly generated gain shift noise may be added to the spectral data to simulate a realistic environment. The gain shift noise may be added by introducing small random variations in the linear term of the quadratic calibration curve of the spectrometer 104. Further optionally, a random background noise pattern may be added to the spectral data. A plurality of different randomly generated background noise patterns may be mixed in random proportions to generate a plurality of background spectra.

[0035] In step 206, the training spectral data is binned into n spectral bins for each of the plurality of known isotopes. That is, each of the training spectral data is grouped according to their energy. For example, for n=1024, 1024 bins can be selected, each corresponding to a particular energy range. 7 For the measured energy range of eV, each bin is approximately 10 3 eV range. 0~3×10 6 For the measured energy range of eV, each bin has a size of 2.93x10 3 It is more preferable that the spectral data span the 1024 eV range. Thus, each spectral data point is assigned to one of 1024 bins that correspond to the energy being measured.

[0036] At step 208, a training histogram is generated for each of the plurality of known isotopes based on the n spectral bins. The training histogram comprises n entries and provides an indication of how much of the training spectral data falls within each bin. In other words, the training histogram provides a bin count associated with each bin. In this example, the training histogram comprises 1024 entries.

[0037] In step 210, a training histogram ratio matrix M is generated for each of the plurality of known isotopes. The training histogram ratio matrix M is a square matrix of length n. In this example, the training histogram ratio matrix M is a square matrix of length 1024. An entry M of the training histogram ratio matrix M is i,j is the bin count of the ith bin divided by the bin count of the jth bin:

[0038]

number

[0039] Those skilled in the art will appreciate that the ratios are reciprocal, so the upper and lower triangles of the matrix are identical and the matrix diagonal elements are constant. At step 212, a number of training bin ratio vectors are extracted from each of the training histogram ratio matrices M. This extraction step may be more easily understood with reference to Figure 3, which shows a training histogram ratio matrix 300 for a 1024-bin histogram.

[0040] For a training histogram ratio matrix M of n bins, the number of bin ratio vectors is n-1. Thus, the training histogram ratio matrix M of this example, which has 1024 bins, has 1023 bin ratio vectors. The entries of the yth training bin ratio vector are the bin ratios of a first bin and a second bin, where the second bin is y bins away from the first bin. In particular, the first training bin ratio vector 302 (having a length of 1023) contains the ratios between the bin counts of adjacent bins (i.e., only one bin away). The second training bin ratio vector 304 (having a length of 1022) contains the ratios between the bin counts of the second neighbors. The final (i.e., 1023rd) training bin ratio vector 306 contains the ratios between the last (i.e., 1024th bin count) and the first bin count, with a separation of 1023 bins. Advantageously, feature diversity is introduced through ratios measured at multiple interval scales between bins, thereby improving the robustness of the method.

[0041] Each bin ratio vector may be stored in a respective data set. In step 214, a first set of isotope predictions is determined for each first bin ratio vector 302 in the training histogram ratio matrix M. This prediction is accomplished by inputting the first bin ratio vectors 302 into a first artificial neural network (ANN) model, such as the ANN model shown in FIG.

[0042] The ANN architecture 400 as shown in Figure 4 comprises an input layer 402, a hidden layer 404, and an output layer 406. Each layer 402, 404, 406 comprises a number of artificial neurons (referred to herein as "nodes") connected through edges. The nodes and edges have associated weights configured to adjust the contribution of certain input data over other input data. The output from each layer 402, 404, 406 propagates to the next layer, where a function is applied to each output. An example function is a rectified linear function.

[0043] The ANN also comprises a number of parameters, including a certain number of neurons in the hidden layer, a learning rate, and a regularization strength. The input layer 402 is configured to receive the bin ratio vector. In particular, the input layer 402 comprises a number of input nodes 402A, 402B, 402C, each configured to receive a respective entry of the bin ratio vector. Thus, there are N input nodes, where N is the length of the input bin ratio vector. For example, at step 512, the input node 402A is configured to receive a first entry of the first bin ratio vector 302 of the histogram ratio matrix 300. The input node 402B is configured to receive a second entry of the first bin ratio vector 302 of the histogram ratio matrix 300. Finally, the input node 402C is configured to receive the 1023rd (i.e., last) entry of the first bin ratio vector 302 of the histogram ratio matrix 300.

[0044] The hidden layer 404 is configured to apply a nonlinear transformation to the inputs received through the input nodes 402A, 402B, 402C. The hidden layer 404 comprises a number of hidden layer nodes 404A, 404B, 404C, 404D, each of which is an ANN model. The hidden layer 404 may also comprise a number of layers.

[0045] In some examples, the hidden layer 404 has a sigmoid activation function, 1 / (1+e -x ), a sigmoidal activation layer (not shown) configured to apply a sigmoidal activation layer. A sigmoidal activation layer may be used for cases where a single isotope is to be identified. A sigmoidal activation layer may also be used for cases where there are multiple isotopes to be identified from a single radiation source.

[0046] The output layer 406 is configured to provide an isotope prediction associated with each of the bin ratio vector entries following application of the hidden layer. In particular, the output layer 406 includes a plurality of output nodes 406A, 406B, 406C, each configured to output a respective isotope prediction. For example, output node 406A may be associated with the first entry of the first bin ratio vector 302 of the histogram ratio matrix 300, 241 The output node 406B may output a first isotope prediction, such as Am. The output node 406B may output a second entry of the first bin ratio vector 304 of the histogram ratio matrix 300, such as Am. 137 The output node 406B may output a second isotope prediction, such as Cs, associated with the 1023rd (or final) entry of the first bin ratio vector 306 of the histogram ratio matrix 300. 152 A third isotope prediction may be output, such as Eu.

[0047] In the case of a single isotope classification, the output node applies a softmax function. For multi-label problems, the output layer 406 has a sigmoid activation function, 1 / (1+e -x ), where x is a weighted sum of the inputs that have been passed through a sigmoid activation function, which then serves as the input to the next layer. Thus, by applying a sigmoid activation function, the output layer 406 may limit the output to values ​​between 0 and 1. In this case, a threshold may be applied, where if the output layer results in a value greater than the threshold, a particular isotope may be identified. For example, if the threshold is 0.5, an output layer value greater than 0.5 may indicate that the classification associated with the output layer value is to be output.

[0048] Returning to method 200, in step 216, a cost function for the first set of isotope predictions is evaluated. For single isotope classification, the cost function is the cross-entropy loss function:

[0049]

number

[0050] Here, K is the total number of output nodes 406, and t i isotope classification, and

[0051]

number

[0052] is the predicted classification provided by each output node 406. In this case, K is 1023 for the first bin ratio vector. For multiple isotope classification, the cost function is set to penalize higher values ​​of network weights and prevent overfitting, L 2 The mean squared error function with regularization is:

[0053]

number

[0054] Here w j is the weight, λ is the regularization strength, t i is the correct isotope classification,

[0055]

number

[0056] is the predicted classification provided by each output node 406. At step 218, a second set of isotope predictions is determined for each of the second bin ratio vectors 304 in the training histogram ratio matrix M. This prediction is accomplished similarly to the prediction of step 512, except using a second ANN model. In this case, the second bin ratio vector 304 has a length of 1022, so the second ANN model has 1022 input and output nodes. The parameters of the second ANN model are configured to minimize the cost function determined at step 514. Thus, a new ANN model is added to the ANN ensemble, including the second bin ratio vectors associated with each of the plurality of known isotopes.

[0057] Steps 216 and 218 are iterated, with the output of step 218 being used for the evaluation of the cost function in step 216 of the subsequent iteration. In addition, each subsequent iteration of step 516 utilizes subsequent bin ratio vectors for each of the multiple known isotopes.

[0058] Steps 216 and 218 may be repeated until a threshold is reached. The threshold may be an output difference threshold. For example, the predicted classification

[0059]

number

[0060] If the difference between the correct isotope classification and the power difference threshold is less than or equal to the power difference threshold, the method 500 may end. An example power difference threshold is 5×10 -2 The threshold may provide an indication as to whether the ANN model meets the performance metric.

[0061] Each subsequent ANN is then trained through backpropagation to minimize a cost function with respect to the node weights. For faster convergence, a Scaled Conjugate Gradient (SCG) method may be utilized.

[0062] Referring now to FIG. 5, a method 200 for distinguishing isotopes using the system 100 of FIG. 1 is shown. In a first step 502, the computing device 102 collects or receives spectral data from the spectrometer 104. The spectral data is the energy associated with the gamma rays received and measured by the spectrometer 104. Thus, a gamma ray spectrum may be generated.

[0063] Optionally, the spectral data may be filtered to match the characteristic resolution of spectrometer 104 to reduce artifacts produced by statistical noise. In step 504, the spectral data is binned into n spectral bins, i.e., each of the spectral data is grouped according to their energy. For example, for n=1024, 1024 bins can be selected, each corresponding to a particular energy range. 4 ~10 7 For the measured energy range of eV, each bin is roughly 10 3 The energy range can range from 1000 to 1000 eV. Thus, each spectral data point is assigned to one of 1024 bins that correspond to the energy being measured.

[0064] In step 506, a histogram is generated based on the n spectral bins. The histogram comprises n entries and provides an indication of how much spectral data falls within each bin. In other words, the histogram provides a bin count associated with each bin. In this example, the histogram comprises 1024 entries.

[0065] In step 508, a histogram ratio matrix M is generated. The histogram ratio matrix M is a square matrix of length n. In this example, the histogram ratio matrix M is a square matrix of length 1024. An entry M of the histogram ratio matrix M is i,j is the bin count of the ith bin divided by the bin count of the jth bin:

[0066]

number

[0067] Those skilled in the art will appreciate that the ratios are reciprocal, so the upper and lower triangles of the matrix are identical and the matrix diagonal elements are constant. At step 510, a number of bin ratio vectors are extracted from the histogram ratio matrix M. This extraction step may be more easily understood with reference to Figure 3, which shows a histogram ratio matrix 300 for a 1024-bin histogram.

[0068] For a histogram ratio matrix M of n bins, the number of bin ratio vectors is n-1. Thus, the histogram ratio matrix M of this example, with 1024 bins, has 1023 bin ratio vectors. The entries of the yth bin ratio vector are the bin ratios of a first bin and a second bin, where the second bin is y bins away from the first bin. In particular, the first bin ratio vector 302 (having a length of 1023) contains the ratios between the bin counts of adjacent bins (i.e., one bin away). The second bin ratio vector 304 (having a length of 1022) contains the ratios between the bin counts of the second neighbors. The final (i.e., 1023rd) bin ratio vector 306 contains the ratios between the last (i.e., 1024th bin count) and the first bin count, with a separation of 1023 bins. Advantageously, feature diversity is introduced through ratios measured at multiple interval scales between bins, thereby improving the robustness of the method.

[0069] Each bin ratio vector may be stored in a respective data set. At step 512, an isotope prediction is determined for each of the bin ratio vectors. In this example, an artificial neural network (ANN) model, such as that shown in FIG. 4, is used for each of the bin ratio vectors. The isotope predictions may be in the form of isotope probabilities, such that the ANN model provides the probability that a radioactive source is a particular isotope. Thus, for this example, an isotope prediction associated with each bin ratio vector of the histogram ratio matrix 300 is generated by the method 200. Further details regarding the ANN model and architecture are discussed in connection with FIGS. 4 and 5.

[0070] In the final step 514 of method 200, a final isotope prediction is determined. The final isotope prediction is determined through an ensemble prediction step. In particular, the isotope predictions associated with each of the bin ratio vectors, as determined in step 512 of method 500, are combined. In this example, the output of step 514 is a majority output of the isotope predictions associated with each of the bin ratio vectors. That is, the largest number of isotope predictions, or the isotopes with the highest cumulative probabilities, will be output by step 514.

[0071] In this example, the isotope 57 Co was output by the 879 output nodes of the output layer 406. Thus, the final isotope prediction is 57 Co. The description set forth herein may be directed to particular implementations. It should be understood that the discussion set forth herein is presented for the purpose of enabling one of ordinary skill in the art to make and use any subject matter defined by the subject matter of the claims herein.

[0072] The subject matter of the claims is not limited to the implementations and figures shown herein, but is intended to include modifications of those implementations, including portions of the implementations, and combinations of elements of different implementations according to the claims. In the development of any such implementation, such as in any engineering or design project, it should be understood that numerous implementation-specific decisions must be made to achieve developer-specific goals, such as compliance with system-related and business-related constraints that may vary from one implementation to another. Moreover, it should be understood that such a development effort may be complex and time-consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill in the art having the benefit of the present invention.

[0073] Reference has been made in detail to various implementations, examples of which are illustrated in the accompanying drawings and figures. In the detailed description, numerous specific details are set forth in order to provide a thorough understanding of the inventions presented herein. However, the inventions presented herein may be practiced without these specific details. In some other instances, well-known methods, procedures, components, circuits, and networks have not been described in detail so as not to unnecessarily obscure the details of the embodiments.

[0074] In addition, terms such as first, second, etc. may be used herein to describe various elements, but it should be understood that these elements should not be limited by these terms. These terms are used only to distinguish one element from another element. For example, a first element can be called a second element, and similarly, a second element can be called a first element. A first element and a second element are both elements, but should not be considered to be the same element.

[0075] The terminology used in the description of the invention set forth herein is for the purpose of describing a particular implementation and is not intended to limit the invention set forth herein. As used in the description of the invention set forth herein and in the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates a different interpretation. The term "and / or" as used herein refers to and includes any and all possible combinations of one or more of the associated listed items. The terms "includes", "including", "comprises" and / or "comprising" as used herein specify the presence of stated features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

Claims

1. A method for isotope identification performed by a computing device, The process involves collecting multiple spectral data from a spectrometer, The process involves generating multiple datasets based on the multiple spectral data using the processor of the computing device, The process of determining multiple isotope probabilities using the processor, wherein each of the multiple isotope probabilities corresponds to each of the multiple datasets, and Equipped with, The isotope probability indicates the probability that the corresponding dataset in the plurality of datasets corresponds to a specific isotope. The isotope identification method further, A method comprising the step of identifying an isotope based on the multiple isotope probabilities using the aforementioned processor.

2. The method according to claim 1, wherein each of the plurality of datasets comprises binned data, and the dataset represents a functional relationship between a first bin of the binned data and a second bin of the binned data.

3. The method according to claim 2, wherein each of the plurality of datasets includes its respective bin ratio vector.

4. The process of generating the aforementioned multiple datasets is as follows: The processor performs the process of sorting the plurality of spectral data into n spectral bins, The processor performs the steps of generating multiple bin ratio vectors, The processor performs the steps of storing the respective bin ratio vectors in each of the multiple datasets. The method according to claim 3, including the method described in claim 3.

5. The process of generating the plurality of bin ratio vectors is as follows: The process of generating a histogram based on the n spectral bins using the aforementioned processor, A step of generating a histogram ratio matrix M based on the histogram using the aforementioned processor, The aforementioned histogram ratio matrix M is a square matrix of length n, and the process is as follows: The process involves the processor extracting the plurality of bin ratio vectors from the histogram ratio matrix, There are k bin ratio vectors, and k = n-1, and the process and The method according to claim 4, including the method described in claim 4.

6. Each entry M in the histogram ratio matrix M ｉ，ｊ The method according to claim 5, wherein is obtained by dividing the i-th bin by the j-th bin.

7. The method according to claim 6, wherein the entry in the y-th bin ratio vector is the bin ratio of the first bin to the second bin, and the second bin is y bins away from the first bin.

8. The method according to claim 2, wherein the bin-sorted data includes the difference between bins.

9. The method according to any one of claims 1 to 8, wherein the isotope probability is determined by using an ensemble, the ensemble comprising a plurality of classifications generated by a multiclass classifier, each classification corresponding to its respective dataset.

10. The method according to claim 9, wherein the ensemble is an artificial neural network (ANN) ensemble, the ANN ensemble comprises a plurality of ANN classifications, each of the plurality of ANN classifications is generated by an ANN, and each ANN classification corresponds to a respective dataset.

11. The method according to claim 10, wherein the isotope probability is determined based on the plurality of ANN classifications.

12. Each ANN comprises one or more parameters, and these parameters are A hidden layer containing multiple neurons, Learning rate and, Regularization strength and The method according to claim 9, wherein one or more are selected from the range.

13. The method according to claim 12, wherein the parameters are selected by applying a random search method.

14. The method according to claim 9, wherein the multi-class classifier is optimized.

15. The method according to claim 14, wherein the multiclass classifier is optimized through backpropagation and is configured to minimize a cost function with respect to network weights.

16. The method according to claim 15, wherein the cost function is a cross-entropy loss function for a single isotope classification.

17. The method according to claim 16, wherein the cost function is a mean squared error cost function for isotopic mixture classification.

18. The method according to claim 17, wherein the cost function comprises a penalty term, and the penalty term is an L2 regularization element.

19. The method according to claim 14, wherein the multi-class classifier is optimized according to a scaling conjugate gradient method.

20. The method according to claim 14, wherein the multi-class classifier is trained until the threshold difference between the predicted output and the correct output reaches a threshold.

21. The method according to any one of claims 1 to 8, wherein the isotope is identified by ensemble prediction.

22. The aforementioned ensemble prediction is, A step of determining the isotope with the highest probability for each of the plurality of datasets, A step of identifying the isotope with the highest probability number, The method according to claim 21, including the method described in claim 21.

Images