Systems and Methods for Classification of Radiation Energy Spectra
20240312572 ยท 2024-09-19
Inventors
- Steven Wayne PAULY (Knoxville, TN, US)
- William Mark RICHARDSON (Knoxville, TN, US)
- Phillip John BOONE (Lenoir City, TN, US)
Cpc classification
G01V5/20
PHYSICS
International classification
G01V5/26
PHYSICS
G01T1/167
PHYSICS
G01T1/29
PHYSICS
Abstract
A method of determining if a radiation spectra comprises radiation corresponding to naturally occurring radiation material sources (NORM), the method including: gathering the radiation spectra; normalizing the gathered spectra by using one or more predefined normalization techniques; energy scaling the normalized spectra by using one or more energy scaling techniques; and determining materials corresponding to each energy reading in the scaled energy spectra for categorizing the determined materials into one or more predefined categories by using one or more classification algorithms.
Claims
1. A radiation portal monitor comprising a plurality of detectors in data communication with a processing system wherein the processing system is configured to determine if a radiation spectrum comprises radiation corresponding to naturally occurring radiation material sources (NORM) by: generating a database of training spectra, each of the training spectra in the database representing a NORM material; gathering the radiation spectra; normalizing the gathered spectra by using one or more predefined normalization techniques; energy scaling the normalized spectra by using one or more energy scaling techniques; fitting the normalized and scaled spectra to each of the training spectra in the database; determining a residual of fitting the normalized and scaled spectra with respect to each of the training spectra; and identifying the normalized and scaled spectra as a NORM material based on the residual and a threshold value.
2. The radiation portal monitor of claim 1 wherein the normalized and scaled spectra is determined as a NORM material if the residual is less than the threshold value.
3. The radiation portal monitor of claim 1 wherein the processor is further configured to identify the normalized scaled spectra as belonging to an other category if the residual is greater than the threshold value.
4. The radiation portal monitor of claim 1 wherein the processor is configured to determine the threshold value empirically based on the training spectra and the residual of fitting the normalized scaled spectra with a class of the training spectra.
5. The radiation portal monitor of claim 1 wherein gathering radiation spectra comprises gathering energy spectra radiated by an object being inspected in a radiation portal monitor by using one or more of the plurality of detectors.
6. The radiation portal monitor of claim 5 wherein the plurality of detectors are large-area low-resolution radiation detectors.
7. The radiation portal monitor of claim 5 wherein the radiation detectors are polyvinyltoluene (PVT) detectors.
8. The radiation portal monitor of claim 5 wherein the radiation detectors are gamma radiation detectors.
9. The radiation portal monitor of claim 5 being a passive system that does not include a radiation source.
10. The radiation portal monitor of claim 5 being a passive radiation portal monitor (RPM) used to detect presence of ionizing radiation sources passing through the RPM.
11. The radiation portal monitor of claim 10 wherein the RPM comprises a plurality of panels, wherein the panels are spaced so to enable a cargo container to pass between two panels.
12. The radiation portal monitor of claim 11 wherein each of the panels comprises a radiation detector.
13. The radiation portal monitor of claim 1 wherein energy scaling the normalized spectra comprises applying predefined weights to the gathered energy spectra to obtain a weighted matrix of said spectra.
14. The radiation portal monitor of claim 1 wherein the one or more energy scaling techniques comprise the Spectrum Dose Index (SDI) technique.
15. The radiation portal monitor of claim 14 wherein the SDI technique adds more weight to higher energies in the normalized spectra for aiding in determination of NORM materials.
16. The radiation portal monitor of claim 1 wherein one or more training spectra used to train Optimal Linear Associative Memory (OLAM) based artificial neural networks is used to populate the database of training spectra.
17. The radiation portal monitor of claim 16 wherein the training spectra for an OLAM based artificial neural network comprises one spectrum from each isotope of NORM materials.
18. The radiation portal monitor of claim 1 wherein an Optimal Linear Associative Memory (OLAM) neural network implemented within a field-programmable gate array (FPGA) integrated circuit is used to determine if the radiation spectra comprises radiation corresponding to NORM materials.
19. The radiation portal monitor of claim 1 wherein the normalized and scaled spectra is fitted to each of the training spectra in the database in a sequential fashion.
20. The radiation portal monitor of claim 1 wherein the NORM materials are one of: Ra-226, Th-232/Tl-208, K-40, and U-238.
Description
BRIEF DESCRIPTION OF THE DRAWINGS
[0035] The accompanying drawings illustrate various embodiments of systems, methods, and embodiments of various other aspects of the disclosure. Any person with ordinary skills in the art will appreciate that the illustrated element boundaries (e.g. boxes, groups of boxes, or other shapes) in the figures represent one example of the boundaries. It may be that in some examples one element may be designed as multiple elements or that multiple elements may be designed as one element. In some examples, an element shown as an internal component of one element may be implemented as an external component in another and vice versa. Furthermore, elements may not be drawn to scale. Non-limiting and non-exhaustive descriptions are described with reference to the following drawings. The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating principles.
[0036]
[0037]
[0038]
[0039]
[0040]
[0041]
[0042]
[0043]
[0044]
[0045]
[0046]
[0047]
[0048]
[0049]
[0050]
[0051]
DETAILED DESCRIPTION
[0052] In embodiments, the present specification provides a system and method for reducing false alarms in radiation-based inspection systems. In embodiments, the system and method of the present specification provide a quick real-time classification of detected energy spectra from radiation-based inspection systems into naturally occurring radiation material sources (NORM), or other categories and the percent match to each class. NORM sources comprise isotopes such as Ra-226. Th-232/Tl-208, K-40, and U-238.
[0053] The present specification is directed towards multiple embodiments. The following disclosure is provided in order to enable a person having ordinary skill in the art to practice the invention. Language used in this specification should not be interpreted as a general disavowal of any one specific embodiment or used to limit the claims beyond the meaning of the terms used therein. The general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the invention. Also, the terminology and phraseology used is for the purpose of describing exemplary embodiments and should not be considered limiting. Thus, the present invention is to be accorded the widest scope encompassing numerous alternatives, modifications and equivalents consistent with the principles and features disclosed. For purpose of clarity, details relating to technical material that is known in the technical fields related to the invention have not been described in detail so as not to unnecessarily obscure the present invention.
[0054] In the description and claims of the application, each of the words comprise, include, have, contain, and forms thereof, are not necessarily limited to members in a list with which the words may be associated. Thus, they are intended to be equivalent in meaning and be open-ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It should be noted herein that any feature or component described in association with a specific embodiment may be used and implemented with any other embodiment unless clearly indicated otherwise.
[0055] It must also be noted that as used herein and in the appended claims, the singular forms a, an, and the include plural references unless the context dictates otherwise. Although any systems and methods similar or equivalent to those described herein can be used in the practice or testing of embodiments of the present disclosure, the preferred, systems and methods are now described.
[0056] In various embodiments, the system includes at least one processor capable of processing programmatic instructions, has a memory capable of storing programmatic instructions, and employs software comprised of a plurality of programmatic instructions for performing the processes described herein.
[0057] In various embodiments, a computing device may be employed to receive and process data signals and image data and may include an input/output controller, at least one communication interface and a system memory. The system memory includes at least one random access memory (RAM) and at least one read-only memory (ROM). These elements are in communication with a central processing unit (CPU) to enable operation of the computing device. In various embodiments, the computing device may be a conventional standalone computer or alternatively, the functions of the computing device may be distributed across a network of multiple computer systems and architectures. In some embodiments, execution of a plurality of sequences of programmatic instructions or code, which are stored in one or more non-volatile memories, enable or cause the CPU of the computing device to perform or enable various functions, processes and algorithms, such as, for example, performing image reconstruction for display on a screen. In alternate embodiments, hard-wired circuitry may be used in place of, or in combination with, software instructions for implementation of the processes of systems and methods described in this application. Thus, the systems and methods described are not limited to any specific combination of hardware and software.
[0058]
[0059]
[0060] In an embodiment, the RPM 1221 may be used in conjunction with an active X-ray imaging system, particularly with application of blanking synchronization of the RPM system when X-rays are emitted. In an embodiment, the method of the present specification for suppressing nuisance alarms due to NORM materials may be implemented in RPM systems contained within an X-ray imaging system. U.S. Pat. No. 8,963,094 titled Composite gamma-neutron detection system and filed on Feb. 24, 2015 and U.S. Pat. No. 10,393,915, titled Integrated primary and special nuclear material alarm resolution, and filed on Aug. 27, 2019, assigned to the Applicant of the present specification are herein incorporated by reference in their entirety.
[0061] The classification systems and methods of the present specification enables reduction of nuisance alarms in passive radiation portal monitors wherein the classification information may be used to determine whether to notify an operator of an alarm based on the presence of a potential radiation-based threat. In embodiments, the classification systems and methods of the present specification may be used to reduce nuisance alarms in any passive radiation detection system, such as, but not limited to, portals, conveyers, handhelds, scrap steel scanning systems, and area monitors that are configured to generate an alarm upon detection of gamma radiation and are used to measure energy spectra emanating from radioactive substances.
[0062]
[0063] Normalization module 1104 normalizes the gathered energy spectra by using one or more predefined normalization techniques, as further described below.
[0064] Scaling module 1106 uses at least one predefined technique to energy scale the normalized energy spectra. In an embodiment, but not limited to such embodiment, scaling module 1106 applies predefined weights to the gathered energy spectra to form a weighted matrix of said spectra by using a Spectrum Dose Index (SDI) technique, which helps to boost the high energy counts from isotopes that emit higher energy gamma radiation in the energy spectra. In embodiments, SDI is a simple scaling that is used to compensate for increases in detected gamma energy.
[0065] In an embodiment, the SDI technique comprises determining a number of counts in a channel of a histogram of the energy spectrum, multiplying the determined counts by the channel number, and summing up all of the multiplied counts. By way of background, as used in the specification, a channel corresponds to a physically defined region of a detector that generates a single electrical signal corresponding to the number of photons received at that physical locationalso known as a count-over a predefined period of time. In embodiments, the scaling module 1106 enables scaling up of the higher energy isotopes from the object being irradiated by gamma rays, which may correspond to NORM sources. In embodiments, energy scaling causes the shape of the spectrum towards the higher energies to have more pronounced features. Energy scaling compensates for detector efficiency with respect to detector energy. As is known, a gamma ray detector becomes less efficient as the gamma energy increases. In embodiments, the efficiency of a detector is measured for a plurality of gamma energies and a resultant energy efficiency curve is used to boost low efficiency regions of the detector.
[0066] In embodiments, an x-axis of the histogram of energy spectrum depicts integer bins x, wherein x=1, 2, 3 . . . n, representing the amplitudes of the pulses collected from the gamma detector. The y-axis of the histogram of energy spectrum represents the quantity of events which occurred for each amplitude bin, and are depicted as a function of x (i.e. F(x)) where x ranges from 1 to n and F(x) ranges from 0 to the size of the variable chosen to represent these values (typically a 16-bit or 32-bit variable). In embodiments, the SDI scaling technique is used to scale the function F(x) by multiplying each F(x) by its bin number x, by using the formula:
Since the histogram of detector pulse amplitudes can be scaled so that the x-axis represents the gamma energy, and the energy response of gamma detectors becomes less efficient as the gamma energy increases, the SDI technique compensates for the decreased efficiency of the detector by boosting the number of events as the energy of detected rays increases. Thus, the use of the SDI technique adds more weight to higher energies in the spectrum which aid in determination of NORM sources. For example, if the efficiency of the detector is 20% at 2 MeV, then the counts received in the detector at this region are scaled up by dividing the counts received by 0.2.
[0067]
[0068] Referring to
[0069] In various embodiments, in order to achieve the OLAM based implementation of the present specification, it is required that the gathered energy spectra are scaled to correspond to the training data and coefficients derived offline. Since the collected energy spectrum may contain more or less data than the predefined training data, the collected energy spectra must be re-scaled to a size corresponding to the training data in order to apply the training data coefficients to the collected energy spectra. In an embodiment, re-scaling is achieved by using mathematical operations such as multiplication or division. In embodiments, the gathered energy spectra include a plurality of energy bins comprising radiation of similar energies, wherein each energy bin is assigned a bin number. In embodiments, the energy bins are of a constant energy size such as, but not limited to, 3 keV per energy bin or channel. The energy bins are formed by the analog-to-digital conversion of the pulse heights of pulses received from a radiation detector. For example, if the pulse heights have 10 bits of resolution, then 1024 energy bins are formed. These correspond to the voltage of the pulses received from the radiation detector, which are proportional to the current from the detector, which are, in turn, proportional to the energy deposited in the detector by the gamma photons that interact within the detector. In an embodiment, after calibration, the energy bins may be energy scaled to an energy level of 3 keV, for example, in the case of the radiation detector being a high-resolution detector. In embodiments, for a PVT detector, having a lower resolution, the pulse heights of pulses received from the detector may be of 8 bits of resolution which results in 256 energy bins being formed. If the bins are calibrated to cover 3 MeV (3000 keV) of energy, the energy bins may represent 3000 keV/256=11.7 keV energy per bin.
[0070] In an embodiment, once the collected energy spectrum has been re-scaled to a comparable size as the training data by applying multiplication or division to each bin of the energy spectrum, a coefficient array for each category (NORM or OTHER) is obtained by obtaining a dot product of an array of the training data and the collected normalized energy spectrum. The dot product is a sum of the bin-by-bin multiplication of the coefficient array and the energy spectrum. This embodiment requires intensive mathematical processing and hence is a less preferred embodiment.
[0071] In an embodiment, the normalization module is configured to enable implementation of Optimal Linear Associative Memory (OLAM) within a field-programmable gate array (FPGA) integrated circuit. In a preferred embodiment, the normalization module 1104 is configured to normalize the gathered energy spectra by counting the number of counts in the energy spectra collected into an energy histogram. In embodiments, a predefined number of counts are required to be collected for the classification module 1108. For example, if 1000 counts are required for making a classification decision, while determining coefficients for operation of the OLAM classifier offline, each class of the energy spectra is normalized to contain a total of 1000 counts. So, in case of an energy spectrum lasting for a duration of 10 minutes and comprising a million counts, each energy bin of the corresponding histogram is scaled so that the sum of all the energy bins equals 1000.
[0072] In embodiments, a coefficient is determined for each energy category from the normalized spectra using the OLAM algorithm. In embodiments, the coefficients for each category (NORM or OTHER) is accumulated per count until a total of a 1000 counts have occurred, which is when it is determined if the 1000 counts correspond a NORM or OTHER type of radiation source. As a count occurs, the amplitude of the count corresponds to the bin within the energy spectrum to which the count belongs, which is 1 while all other bins are 0. In an embodiment, the coefficient for that bin is added to an accumulator and once the number of counts reaches the predefined number of counts (in this example, 1000), the processing is completed and the accumulator for each energy class is used to determine a winning energy class, which is used to further determine if the energy class belongs a NORM or OTHER type of radiation source.
[0073] In an embodiment, the SDI technique comprises multiplying the number of counts in a particular energy bin by the bin number. For example, if there are two counts in an energy bin having a bin number of 200, then the SDI scaling method would include multiplying the two counts by 200 which is equal to 400. In an embodiment, a scaling technique comprising using an inverse of the energy efficiency curve of the detector is used to boost the energy bins having lower energy. For example, if the energy represented by bin number 200 is only 20% efficient, then this bin is scaled by dividing the counts within it by 0.2 in order to boost the energy efficiency of said energy bin. In other embodiments, various other suitable scaling techniques may be used to scale the gathered energy spectra, without limiting the scope of the present specification.
[0074] The classification module 1108 provides two categories, specifically NORM and other, to an OLAM algorithm training set. As is known, when used for isotope identification, the OLAM algorithm always presents, as an answer, results in only one of the categories or isotopes present in the training sets used to train OLAM based artificial neural networks. Thus, the OLAM algorithm is capable of identifying only the one or more isotopes present in its training sets. If the OLAM network is trained by using each known NORM source and the fact that NORM sources typically have higher energy features than other sources of radiation, and if the trained OLAM network is presented with a new isotope that is not present in the training sets, then the OLAM algorithm does not recognize the new isotope and uses a least squares residual technique to classify said isotope as other material by default. Hence if the OLAM algorithm does not recognize a particular energy spectrum obtained from the scaling module 1106, said energy spectrum is categorized as belonging to an other isotope/material. In other embodiments, classification module 1108 uses various other classification techniques to categorize the energy spectra for identifying NORM isotopes. In other embodiments, other neural networks such as, but not limited to a k-NN (k nearest neighbor) network may be used to categorize the energy scaled spectra for determining energy sources belonging to the NORM category.
[0075] In embodiments, where the OLAM neural network is used to classify an unknown/gathered, energy scaled spectra into a NORM or an other class, the output of the OLAM algorithm is in the form of two numbers that correspond to the extent at which (how much) the unknown energy scaled spectra matches each of the two classes. For example, if the energy scaled spectra comprises a NORM source, the output of the OLAM network is approximately 100% for the NORM class and approximately 0% for the other class. In another example, if a scaled energy spectrum comprises a mix of isotopes comprising Am-241 and Ba-133 (which belong to the other class) and Ra-226 (which belongs to the NORM class) and each isotope contributes approximately 33% to the total counts in the scaled energy spectrum, then the output of the OLAM network is approximately 66% for the other class and 33% for the NORM class.
[0076]
[0077] At step 304, energy spectra generated by an object being inspected in a radiation inspection system is gathered using one or more radiation detectors of the system. At step 306, the gathered energy spectra are normalized by using one or more predefined normalization techniques. At step 308, the normalized spectra are energy scaled by using one or more predefined energy weighting techniques. In an embodiment, predefined weights are applied to the gathered energy spectra to form a weighted matrix of said spectra by using Spectrum Dose Index (SDI) technique. At step 310, the normalized and scaled spectra is fitted to each of the training spectra in the database. In an embodiment, the normalized and scaled spectra is fitted to each of the training spectra in the database in a sequential fashion. In an embodiment, where an OLAM based neural network is used for classification of isotopes, a set of coefficients for each class of the training spectra are applied sequentially to the gathered unknown scaled spectra. In another embodiment, the scaled spectra is fitted across all of the training sets simultaneously.
[0078] At step 312, a residual of fitting the normalized scaled spectra with respect to each of the training spectra classes is determined. At step 314 the normalized scaled spectra is identified as belonging to the class of training spectra that results in the smallest residual as determined at step 312. In an embodiment, the training spectra belong to a NORM category, and the normalized scaled spectra that results in a residual (r) value (as determined in step 312) being less than a predefined threshold value, is determined as belonging to the NORM category; while the normalized scaled spectra that results in a residual value (as determined in step 312) being greater than a predefined threshold value, is determined as belonging to the other category. In an embodiment, the threshold value is determined empirically after the OLAM neural network is trained and all the residual values are determined. In embodiments, the residual values of isotopes belonging to the NORM class must be lower than the residual values of all other isotopes in the normalized scaled spectra. For example, if with respect to a normalized scaled spectra the residual values for NORM sources range from 2.4 to 20.5 and the residual values of all other isotopes range from 27.2 to 115.3, the threshold value may be defined as 23.85 which is half way between 20.5 and 27.2. In other embodiments, the threshold value may be predefined using other suitable techniques.
[0079] In a training scenario, radiation spectra are acquired to generate an experimental class that represents a predefined number of radionuclides belonging to the Naturally Occurring Radioactive Materials (NORM) group. Table 1 (below) illustrates all training radionuclides acquired with each corresponding vector symbol.
TABLE-US-00001 TABLE 1 Radionuclide Vector Symbol Ra-226 Ra K-40 K Th-232 Th
[0080] For conducting the training experiment, a background spectrum (Bkg?R.sup.Nc) was first collected. Subsequently, a radiation spectrum was acquired for 60 seconds, wherein Nc=256 channels from a PVT detector.
[0081] Each acquired spectrum vector is represented as a vector v?R.sup.Nc with the i-th individual element being v[i]. The preprocessed spectrum {tilde over (v)}?R.sup.Ne was calculated considering ?i={1, 2, . . . , N.sub.C} by using the equation:
where normalization is performed using both
[0082]
[0083]
[0084]
[0085] During training the set of processed training spectrum may be written more compactly as the matrix:
consisting of the column vectors of spectrum where N.sub.t=3 is the number of training vectors.
Measured spectra for classification, denoted by:
were also acquired for 60 seconds from a PVT detector having N.sub.c=256 channels. Again, each measured spectrum was normalized in the same manner and denoted by {tilde over (y)}.
[0086] Conceptually, if measured spectrum {tilde over (y)} is reconstructed using a linear combination of training spectrum, the system may be represented by the matrix equation:
where ??.sup.N.sup.
[0087] A solution does not exist for this overdetermined problem since A is not full rank. A matrix is said to have full rank if its rank equals the largest possible for a matrix of the same dimensions, which is the lesser of the number of rows and columns. Hence, the linear least squares solution {circumflex over (?)} was used to solve for the coefficients using:
These coefficients may be used to reconstruct the measured spectrum, where the reconstructed vector is denoted by {tilde over (y)} by projecting the measured vector {tilde over (y)} back into the space spanned by the columns of the training matrix A according to:
As is known a residual is a measure of how well a line fits an individual data point. Hence, residuals from a fitted model are defined as the differences between the response data and the fit to the response data at each predictor value:
The resulting residuals for each channel from reconstruction consist of the vector: r?.sup.Nc, are an essential component of classification, and are defined by:
The residual vector calculation may be simplified into a single matrix by calculating the projection matrix P that projects the measured spectrum vector y into the null space of training matrix A according to:
where I is the identity matrix and
[0088] In an embodiment, the classification method does not attempt to classify multiple classes simultaneously because reconstruction may result in negative weights that do not make physical sense. Rather, when a measured spectrum is not used in training, the fit often results in a large residual and as a result, misclassification. Hence, in an embodiment, only spectra of a single class are trained at a time for a binary decision indicating whether that class is a member or not. When a measured spectrum of a different class is encountered, the residual is used as a basis of classification.
[0089] In order to obtain a size of the residual as compared to the total size of the measured spectrum, in an embodiment, the size of the residual vector is divided by the size of the measured spectrum by using a simple decimal fraction. This indicates proportionally how large the fitted vector's residual is compared to the original spectrum. The magnitude or size of each vector is calculated as the sum of the absolute values of each element or in compact notation the L.sup.1 Norm function denoted by ?v?.sub.1 for a measured vector v. Consequently, the relative residual scalar {circumflex over (r)} used for classification may be obtained by:
Since, the relative residual is large when using a measured vector y that is not a member of the trained class and small when it is a member of the class, in an embodiment, the present specification provides a simple threshold created empirically using experimental data, as described above with respect to step 312 of
[0090]
[0091] Table 2 shows values for the measured spectra of a plurality of other (non-NORM) materials along with their corresponding relative residual values. It can be seen in Table 2 that all of the illustrated relative residual values are significantly higher than the relative residual value for background, i.e. {circumflex over (r)}=16.9.
TABLE-US-00002 TABLE 2 Radionuclide Vector Symbol {circumflex over (r)} Am-241 Am-241 170.8 Co-57 Co-57 163.2 Ba-133 Ba-133 109.2 Cs-137 Cs-137 76.5 Co-60 Co-60 25.3
[0092]
[0093]
[0094] In an exemplary test case a plurality of detected radiation spectra along with background spectra was analyzed by using the categorization method of the present specification. The background spectra comprise NORM isotopes in a majority concentration as these naturally occurring isotopes produce a gamma radiation background. By using a classification algorithm such as OLAM, approximately 80% of the background was categorized to the NORM category, while 20% to the other category. On applying the same energy scaling and classifying algorithms to a Cs-137 spectrum collected from each of four PVT detectors deployed in a radiation inspection system, and to the sum of the four PVT detectors, approximately 80% of the spectra was categorized to the other category and 20% to the NORM category, since the Cs-137 spectra collected included background radiation as well. In various embodiments, the method of the present specification may be used to identify a plurality of different pairs of isotope classes, such as, but not limited to MEDICAL and OTHER; or special nuclear materials, i.e. SNM and OTHER.
[0095] The above examples are merely illustrative of the many applications of the systems, apparatuses, and methods of present specification. Although only a few embodiments of the present invention have been described herein, it should be understood that the present invention might be embodied in many other specific forms without departing from the spirit or scope of the invention. Therefore, the present examples and embodiments are to be considered as illustrative and not restrictive, and the invention may be modified within the scope of the appended claims.