A device is a method for measuring an object exhibiting β activity.

Abstract

A method for characterizing an object, the object comprising at least one radionuclide emitting β radiation, the method comprising: a) placing a detector facing the object; b) detecting the radiation emitted by the object; c) from the radiation detected by the detector, forming an input spectrum, comprising a β component; d) applying an identification algorithm, associated with a radionuclide, to the input spectrum, the identification algorithm determining the presence of the radionuclide, to which the identification algorithm is associated, in the object; e) as a function of d) identifying the radionuclide contained in the object; f) applying a deconvolution algorithm to the input spectrum so as to estimate a contribution of each radionuclide, identified in e), in the input spectrum.

Description

Title of the invention: Device is a method for measuring an object exhibiting activity p Technical field

[0001] The technical field of the invention is the measurement of the activity [3 of an object by spectrometry. EARLIER ART

[0002] Knowledge of the radiological status of the processes and equipment of a nuclear facility is essential for establishing robust decommissioning scenarios and defining waste management, in particular its categorization and disposal. Non-destructive in-situ nuclear measurements, coupled with modeling techniques, make it possible to establish a radiological inventory of the equipment on stands, processes, and civil engineering structures.

[0003] Gamma spectrometry is one of the most commonly used passive non-destructive nuclear measurement techniques for obtaining qualitative and quantitative information on gamma-emitting radionuclides present in an object. Gamma spectrometry allows the acquisition of gamma spectra, on which characteristic peaks can be identified, corresponding to a signature that allows the radionuclides to be identified.

[0004] However, some radionuclides emit little or no gamma radiation. These are, for example, so-called pure [3] emitters; their in situ radiological characterization is made difficult by the very short range of electrons in dense media. An example of a pure [3]-emitting radionuclide is 90Sr. The identification of pure [3] emitters, and the quantification of their activity, is usually carried out in the laboratory through destructive analyses of samples taken in the field. However, these destructive laboratory measurements have certain drawbacks: questions about the representativeness of the samples taken, and the cost and time required for analysis.

[0005] The publication Vetter K, “In-situ quantification of gamma-ray and beta-only emitting radionuclides”, arXiv, Apr. 09, 2023, http: / / arxiv.org / abs / 2304.07632, hereinafter referred to as [Vetter], describes a spectrometric measurement in which a compact semiconductor detector, of the CdZnTe type, is used to acquire a spectrum. The device is positioned sufficiently close to an object to be characterized to detect the gamma-ray radiation generated by 137Cs, the internal conversion electrons emitted by Cs, and the [3] radiation emitted by Sr. From the spectrum, a component representative of the interactions occurring in a surface region of the detector is extracted, which contains most of the interactions. of electronic origin (internal conversion electrons, radiation [3]) as well as a component representing interactions occurring deep within the detector, which includes interactions produced by gamma photons. A spectrum is acquired by placing a screen between the detector and the object, so as to determine the interactions, in the surface zone, due to gamma photons. The spectra due to electrons and photons are then deconvolved, using a maximum likelihood algorithm, to quantify the radionuclide activity.

[0006] The advantage of the method described in the Vetter publication is that it can estimate the activity of pure [3] emitters on an object without taking a sample and without destructive analysis. However, this method requires determining the interaction depth in the detector, which is relatively complex. Separating the spectrum into a component representing the interactions of electronic origin and the interactions of photon origin can be tedious to implement, particularly in field applications outside of laboratory conditions. Furthermore, analyzing the spectra is relatively complex.

[0007] Another drawback is the use of a CdZnTe type detector: this type of detector is based on crystals whose volume is limited to a few cm3. Finally, the use of a double measurement, with and without a screen, constitutes another implementation constraint.

[0008] The inventors have developed an alternative method, having the same objective as the method described in publication [Vetter]. The inventors' method does not require consideration of the interaction depth in a detector. It is not limited to the use of a semiconductor detector and can advantageously be implemented on scintillator detectors. Description of the invention

[0009] A first object of the invention is a method for characterizing an object, the object comprising at least one radionuclide emitting radiation [3, the method comprising: - a) arrangement of a detector facing the object, the detector being configured to acquire a spectrum, representing a distribution of the energy released, in the detector, by the radiation emitted by the object; - b) detection of the radiation emitted by the object, by the detector, during an acquisition period, and acquisition of a spectrum of the detected radiation. - c) from the spectrum of the detected radiation, formation of an input spectrum, comprising a [3] component, which corresponds to a distribution of the energy released by the [3] radiation in the detector; - d) application of an identification algorithm, associated with a radionuclide, to the input spectrum, the identification algorithm being configured to determine the presence of the radionuclide, to which the identification algorithm is associated, in the object, step d) being repeated for different radionuclides, implementing different identification algorithms; - e) depending on d) identification of each radionuclide contained in the object; - f) application of an input spectrum deconvolution algorithm to estimate a contribution of each radionuclide, identified in e), in the input spectrum; - g) for each radionuclide identified in e), from the contribution of the radionuclide, in the input spectrum, estimated in f), estimation of an activity and / or depth along which the radionuclide extends in the object.

[0010] Steps d), f) and g) are implemented by a processing unit from the input spectrum. Step e) can be implemented by the processing unit.

[0011] According to one possibility, in step d), each identification algorithm is an artificial intelligence identification algorithm associated with each radionuclide, with at least two different radionuclides being respectively associated with two different identification algorithms.

[0012] Each identification algorithm can be a neural network.

[0013] According to one possibility, the deconvolution algorithm is based on a deconvolution database comprising at least one representative spectrum of each radionuclide identified in e).

[0014] According to one possibility: - the deconvolution database contains, for the same radionuclide, different spectra representing different distributions of the radionuclide in the object; - step g) involves determining the distribution of the radionuclide in the object.

[0015] According to one possibility: - The deconvolution database contains, for the same radionuclide, different spectra representative of different depths of radionuclides in the object, from a surface of the object facing the detector; - step g) involves determining the depth to which the radionuclide extends in the object.

[0016] At least one radionuclide to which an identification algorithm is associated can be a pure [3] emitter.

[0017] The detector may include an organic scintillator-type material to detect the radiation emitted by the object.

[0018] The detector may comprise a volume of semiconductor or inorganic scintillator less than 10 mm thick, disposed facing the object, the thickness being considered in a direction normal to the object.

[0019] According to one possibility, the object exhibits natural activity, and step c) comprises: - estimation of a spectrum of natural activity of the object; - subtraction of the object's natural activity spectrum from the acquired spectrum during step b) so as to form the input spectrum.

[0020] The detector may include a removable screen, configured to be interposed between the detector and the object, the method comprising: - acquisition of a background spectrum, during which the screen is interposed between the detector and the object; - step c) involves subtracting the background spectrum from the spectrum acquired during step b) to form the input spectrum.

[0021] A second object of the invention is a detection device, comprising a detector, configured to acquire a spectrum of radiation [3 emitted by an object, the spectrum representing a distribution of the energy released, in the detector, during interactions of ionizing radiation in the detector, the device comprising a processing unit configured to implement steps d) to f) of a process according to the first object of the invention.

[0022] The detector may include an organic scintillator-type material to detect the radiation emitted by the object.

[0023] The detector may comprise a volume of semiconductor or inorganic scintillator with a thickness of less than 10 mm.

[0024] The detector may include a removable screen, configured to be interposed between the detector and the object.

[0025] The invention will be better understood upon reading the description of the exemplary embodiments presented later in this description, in connection with the figures listed below. FIGURES

[0026] Fig. 1 schematically illustrates a measuring device enabling implementation of the invention.

[0027] Figure 2A shows a spectrum of 137Cs measured in the laboratory. Unless otherwise specified, for each spectrum described in this application, the x-axis represents the energy (unit MeV) and the y-axis represents the number of interactions detected.

[0028] Fig. 2B shows a γ spectrum of 137Cs measured in the laboratory.

[0029] Fig. 2C shows a spectrum of 90Sr measured in the laboratory.

[0030] Fig. 3 shows a [3y] spectrum of an object containing 137Cs and 90Sr.

[0031] Figure 4 schematically represents a modeled configuration.

[0032] Figures 5A and 5B show modeled spectra, as well as contributions from 137 90 artificial radionuclides (Cs and Sr) and contributions from natural radionuclides in two different configurations.

[0033] Figure 6 schematically illustrates the main steps of a process according to the invention.

[0034] Figure [Fig.7A] schematically represents another modeled configuration.

[0035] Fig. 7B shows examples of learning spectra.

[0036] Fig. 8A shows a modeled spectrum of an object containing 90Sr.

[0037] Figure 8B represents the probability of the presence of different radionuclides in the spectrum of Figure 8A. This is the output of an identification algorithm.

[0038] Fig. 9A shows a modeled spectrum of an object containing 14C and 36C1.

[0039] Figure 9B represents a probability of presence of different radionuclides in the spectrum of [Fig.9A].

[0040] Fig. 1OA shows modeled spectra for the same 90Sr activity distributed over different depths, according to an activity gradient following an exponential shape.

[0041] Fig.1OB shows the modeled spectra of Fig.1OA normalized by their respective integrals.

[0042] Fig.1OC shows a ratio between each spectrum of Fig.1OB and a spectrum representative of a homogeneous activity of 90Sr, normalized by its integral.

[0043] The [Fig.lOD] is the equivalent of the [Fig.lOC] for an activity of 137Cs distributed according to different depths.

[0044] The [Fig. 11] is a comparison of a modeled spectrum of 90Sr activity, distributed over a depth of 40 mm, according to an exponential gradient, and a spectrum of homogeneous 90Sr activity.

[0045] Figures 12A to 12D show the progressive adjustment of spectra resulting from the deconvolution algorithm with respect to a measured spectrum, as a function of the iterations.

[0046] Figure 13 schematically illustrates a sample taken transversely to a channel in a graphite-moderated nuclear reactor.

[0047] Figures 14A to 14D represent probabilities of presence of 137Cs and 90Sr in graphite samples: each probability was established by implementing 100 different identification neural networks for each sample.

[0048] Figure 15 shows an example of spectral deconvolution implementation to determine the contributions of different radionuclides in a measured spectrum. In Figure 12, the x-axis corresponds to each channel.

[0049] Fig. 10A shows a spectrum measured on a graphite sample.

[0050] Figure [Fig. 10B] shows a spectrum measured by placing a screen between the detector and the graphite sample.

[0051] Fig. 16C shows a spectrum obtained by subtracting the spectra shown in Figures 16A and 16B. PRESENTATION OF SPECIFIC IMPLEMENTATION METHODS

[0052] Figure 1 represents a measuring device for measuring the activity of an object 2. The device includes a scintillator detector 10, comprising a scintillator material 11, preferably organic, preferably based on polyvinyltoluene (PVT), as described in the publication Venara J. et al., "Design and development of a portable 3-spectrometer for 90Sr activity measurements in contaminated matrices," Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detetors and Associated Equipment, vol. 953, p. 163081, Feb. 2020. Light pulses are formed through interactions between ionizing radiation and the scintillator material. These light pulses are converted into electrical pulses by one or more photodetectors 12. The electrical pulses are then processed by a spectrometry circuit 13.The spectrometry circuit 13 is configured to form an amplitude histogram of the pulses detected by the organic scintillator during an acquisition period.

[0053] By implementing an energy calibration function, resulting from an energy calibration, it is common practice to establish a correspondence between the pulse amplitudes and energy values. When an ionizing particle deposits all its energy in the scintillator, the amplitude of the pulse it generates corresponds to the particle's energy before interaction in the detector material. The detector's energy calibration, that is, the correspondence between the pulse amplitudes and the energy, was performed using a 207Bi and / or 137Cs source, which emit electrons at discrete energy values ​​by internal conversion. When the source used for calibration emits gamma radiation, It is possible to exploit certain characteristic energies, for example the energy corresponding to the Compton front or the energy of photoelectric peaks.

[0054] The use of an organic scintillator detector is suitable for performing charged particle spectrometry, of type [3. An organic scintillator is also sensitive to ionizing photons of type X or y. However, the materials forming an organic scintillator have a low atomic number, which makes them less conducive to the formation of photoelectric interactions, compared to inorganic scintillators or semiconductor detectors.

[0055] The scintillator detector is covered with a thin, optically sealed envelope 14, for example an aluminized PET (Polyethylene Terephthalate) film 18 pm thick, to ensure isolation from ambient light. A thin layer minimizes the probability of interaction with gamma radiation. In the following text, the term [3] particle refers to a [3] particle.

[0056] The thickness e of the scintillator material is, for example, 4 mm. The diameter of the scintillator material in this example is 76 mm. An organic scintillator material has the advantage of being relatively insensitive to gamma radiation due to its low atomic number. Furthermore, this type of scintillator limits the backscattering of particles [3]. Another advantage is the stability of the scintillator material's response to temperature variations. The response, in terms of the light intensity produced, when exposed to the same radiation, is stable from 0° to 50°C, which is suitable for field use.

[0057] The measuring device includes a processing unit 20, configured to implement spectrum processing steps described below. The processing unit 20 is programmed to execute instructions encoded in a memory, connected to the processing unit by wired or wireless link. The processing unit 20 may, in particular, include a microprocessor.

[0058] According to one variant, the detector may comprise a semiconductor material, suitable for spectrometry [3. It may for example be a silicon-type semiconductor, for example planar silicon.

[0059] The advantage of an organic scintillator material is its ability to be manufactured in different dimensions and shapes. When the object is a sample taken from it, the shape can be adapted to the shape of the sample.

[0060] According to one possibility, the detector 10 includes a removable screen 15, acting as a shutter, configured to be positioned: - in a closed position, between detector 10 and object to be measured 2; - or in an open position, freeing up the space between detector 10 and object 2, as shown in [Fig.1].

[0061] The presence of the removable screen 15 is not necessary.

[0062] In this example, the screen 15 is mobile in translation in a plane parallel to the detector material 11. The screen is for example made of aluminium, the thickness being for example equal to 4 mm.

[0063] The invention is based on a detection, by the detector 10, of [3 and possibly y particles emitted by the object 2. The object 2 is an object to be controlled, capable of exhibiting a mass or surface activity consisting of [3y emitting radionuclides, such as 137Cs, or pure [3 emitting radionuclides, such as 14C, 90Sr or 36C1.

[0064] In order to exhibit sufficient sensitivity to [3] particles, the detector is preferably placed at a short distance d from the object to be characterized. The distance d is preferably non-zero, and between a few mm and a few cm, for example 1 cm. This allows the formation of a [3y] spectrum, the short distance between the detector and the object favoring the contribution of [3] particles to the spectrum.

[0065] According to one possibility, the contribution of the y radiation in the [3y] spectrum is limited by interposing the screen 15 between the detector 10 and the object 2. This allows a y spectrum to be acquired. The y spectrum is then subtracted from the [3y] spectrum, which allows a [3] spectrum to be formed, which is considered to be representative only of the [3] radiation emitted by the object.

[0066] Thus, the invention is implemented either from a [3y] spectrum or from a [3] spectrum. Generally, the invention is implemented from a spectrum in which the contribution of [3] radiation is greater than 10%, or even 20% or 30% or more. By contribution, we mean the quantity of interactions, taken into account to form the spectrum, due to [3] radiation.

[0067] In the examples described below, the object is a concrete wall. An additional difficulty with a material such as concrete is the presence of natural activity. This activity, although low, can complicate the interpretation of measurements, particularly when addressing low levels of artificial activity, on the order of or less than 1 Bg / g. Among the natural radioelements potentially present in the object are, for example, potassium (K), as well as decay products of thorium (Th) and uranium-238 (238U). Such radioelements are, for example, present in concrete objects.

[0068] Another difficulty, well known in the field of nuclear measurement, is the potential presence of background noise resulting from artificial radioactivity present in the environment of the detector, outside of the object analyzed.

[0069] Figures 2A to 2C are spectra measured in the laboratory by placing known point sources of Cs or Sr in front of the detector, at a distance d of 10 mm. Fig. 2A shows a spectrum of the 137Cs source, formed from the [3 and y] radiation emitted by 137Cs in equilibrium with 137mBa. Fig. 2B shows a spectrum of the y radiation from 137Cs. Fig. 2B was obtained by placing a screen, as precisely described, between the 137Cs source and the detector, in order to absorb the [3] radiation emitted by the source. Comparing Figures 2A and 2B, we observe that a significant portion of the spectrum originates from the [3] radiation contribution. This is due to the short distance between the detector and the source.

[0070] Figure 2C shows the spectrum of the 90Sr source (in equilibrium with 90Y), the spectrum being entirely due to interactions of [3] particles in the detector. The activity of the standard source was 9 kBq, and the acquisition time was 10 minutes.

[0071] Fig. 3 shows a spectrum, measured in the laboratory, using the two standard point sources of Cs and Sr, with an activity of the order of 9 kBq.

[0072] The process, described below, aims to obtain, from a spectrum such as that shown in [Fig. 3]: - to identify the [3 or [3y] emitting radionuclides present in the analyzed object; - to estimate a distribution of each radionuclide identified in the object: surface or volume distribution, and possibly thickness, from the surface, in which each radionuclide is present; - and / or quantify the activity of each identified radionuclide.

[0073] The invention is particularly interesting if the presence of at least one pure [3]-emitting radionuclide is suspected in the analyzed object. The inventors have observed that the short range of electrons in the detector material, in this case the organic scintillator, leads to very different spectra depending on whether the activity is distributed superficially or deep within the analyzed object. The use of a low-density organic scintillator material can make it possible to obtain different spectra depending on the depth at which the activity is distributed within the object, particularly with regard to radionuclides emitting high-energy [3] particles.Also, although it is possible to use an inorganic scintillator material, such as Na₂, Csl, LaBr₃, or a semiconductor material (Si, Ge), using an organic scintillator material allows for better discrimination between spectra corresponding to activity distributed at different depths within the analyzed object. Using an organic scintillator results in a higher ratio between the [3] contribution and the y contribution in the measured spectra. Furthermore, an organic scintillator can be sized so that the surface area exposed to the object being monitored is large, compared to a scintillator made of an inorganic material. In addition, organic scintillators can be produced in a wide variety of shapes, both planar and non-planar, allowing for good adaptation to the geometry of the object being monitored.

[0074] Although preferable, the use of an organic scintillator is not an essential condition for implementing the invention. For example, a sufficiently thin semiconductor detector or inorganic scintillator can be used to limit sensitivity to gamma radiation. A semiconductor detector based on a silicon crystal or an inorganic scintillator can be used. Influence of natural activity

[0075] One of the intended applications of the invention is the low-level monitoring of concrete civil engineering structures to verify that activity levels meet predetermined targets. For example, this may involve achieving activity levels defined in so-called clearance levels, which are 1 Bq / g for 137Cs or 90Sr. At such levels, the natural activity of certain materials, such as concrete, can complicate the interpretation of the spectra, as previously mentioned.

[0076] The inventors have modeled the impact of natural radioactivity on the spectra measured by a device such as the one described in connection with [Fig. 1]. [Fig. 4] represents a modeled geometry, with the detector confined in a metallic envelope 16. In [Fig. 4], a homogeneous Sr and Cs activity over a thickness of 1 cm extending from the surface of the object, opposite the detector, has been modeled. The modeling was carried out using the MCNP (Monte Carlo N Particle) transport code. [Fig. 5A] shows a modeled spectrum [3], taking into account an activity of 90Sr and 137Cs of 1 Bg / g, as well as activities of 40K, 232Th and 238U of 0.5 Bg / g, 0.03 Bg / g and 0.02 Bg / g respectively. The [3] spectrum was modeled by considering only the interactions of [3] particles in the detector. Figure 5A shows the respective contributions of each radionuclide.The share of natural activity is 26% (spectrum “totalnat [3” on [Fig.5A]) of the total [3] spectrum. The value of 26% corresponds to the integral of the totalnat spectrum over the integral of the total spectrum.

[0077] Figure 5B shows a modeled γ spectrum, taking into account a 137Cs activity of 1 Bg / g, as well as 40K, 232Th, and 238U activities of 0.5 Bg / g, 0.03 Bg / g, and 0.02 Bg / g, respectively. The γ spectrum was modeled by considering only the interactions of γ particles in the detector. The natural activity accounts for 18% (the "total γ spectrum" in Figure 5B) of the total γ spectrum.

[0078] Table 1 shows, for different specific activities of 137Cs and 90Sr with 137 90 Cs activity = Sr activity, distributed homogeneously over a 1 cm thickness of concrete, the share of natural activity in the spectrum [3 and in the y spectrum. Bq / g (137Cs = 90Sr) Natural proportion [3 (%) Natural proportion y (%) 1 26 17.5 2 14.9 8.8 5 6.6 3.5 10 3.4 1.8 20 1.7 0.39 30 1.2 0.6 40 0.8 0.4

[0079] Table 1

[0080] The results presented in Table 1 show the proportion of natural radioactivity in the measured spectrum. When this proportion is deemed too high, for example, for low levels of artificial activity, the acquired spectrum can be corrected to eliminate the contribution of natural activity. This can be achieved by estimating the contribution of the object's natural activity to the measured spectrum (the [3y] spectrum or |3 spectrum). This natural activity, which is assumed to be homogeneous throughout the object, can result from: - either from an analysis of a sample taken from the object or from another object, considered to be representative; - either a measurement, for example a measurement by high-resolution gamma spectrometry, for example Germanium, on the object or on another object considered to be representative.

[0081] The contribution of natural activity in the spectrum is then subtracted, so as to have a [3y] spectrum or [3] spectrum in which the contribution of natural activity is considered negligible.

[0082] Figure 6 schematically illustrates the main steps of the invention.

[0083] Step 100: Positioning the device 10 facing the object to be monitored and acquiring a spectrum of the radiation emitted by the object. This can be a [3y spectrum when the object contains y-emitting radionuclides, or a [3 spectrum when the object contains only pure [3-emitting radionuclides.

[0084] Step 110: Acquisition of a y spectrum and correction of the y contribution in the acquired spectrum.

[0085] During step 110, which is optional, the screen 15 is placed between the object and the detector. This allows a spectrum representative of the y component of the [3y] spectrum acquired during step 110 to be acquired. A [3] spectrum is formed by subtraction, as described below, in connection with Figures 16A to 16C.

[0086] Step 110 is optional. It is implemented when the contribution of the y radiation in the acquired spectrum is too large.

[0087] Step 120: correction of natural activity

[0088] During step 120, a contribution of natural activity to the [3 or Py] spectrum resulting from step 110 or step 100 is estimated. This contribution is subtracted from the spectrum acquired in step 100 or from the spectrum resulting from step 110. Step 120 is optional. It is implemented when the contribution of natural activity to the acquired spectrum, or the spectrum resulting from step 110, is too large.

[0089] Following steps 100 to 120, we have an input spectrum Spin, which is either the spectrum acquired during step 100, or the spectrum formed following any corrections described in connection with steps 110 to 120. The input spectrum Spin forms an input data for the algorithms described in processing steps 130 and 140, implemented by the processing unit 20.

[0090] Step 130: Identification of radionuclides

[0091] An important aspect of the invention is to combine two successive steps of input spectrum analysis: a first identification step, so as to identify the radionuclides present in the object, based on the input spectrum, without quantification. The objective is to identify the radionuclides present from a pre-established list. Following this first step, and based on the identification carried out, a second step is undertaken, aimed at estimating the contributions of the identified radionuclides to the spectrum.

[0092] The identification step is implemented using identification algorithms, which are supervised learning artificial intelligence algorithms. Each identification algorithm is designed to identify the presence of a radionuclide i in the input spectrum. Each identification algorithm can be a neural network, for example, a convolutional neural network (CNN), or for example, a Bayesian convolutional neural network, associated with a radionuclide i. The index i refers to the radionuclide to which the neural network is associated. Such neural networks are common in the processing of structured data, such as images or histograms. A succession of convolutional layers allows the extraction of features from the input spectrum.The convolutional layers lead to multilayer perceptron-type layers, allowing the determination of a probability of presence of the radionuclide associated with the convolutional neural network CNN, based on the features extracted by the convolutional layers.

[0093] The output of each identification algorithm is a probability of the presence of the radionuclide, to which the identification algorithm is associated, in the object. Thus, there are as many identification algorithms as there are radionuclides in the list that could potentially be present in the object. The input spectrum Spin forms an input for each of these algorithms. For each radionuclide in the list, the output of each algorithm allows the radionuclide to be considered identified. or not identified, according to the input spectrum. Each identification algorithm has previously been trained, using modeled or acquired training spectra in the presence and absence of the radionuclide, to which it is associated, in the object, or in an object considered comparable.

[0094] The output of each CNNt identification algorithm is a probability Pj that the radionuclide i is present in the object examined. It is considered that above a certain threshold, for example 0.5, the radionuclide ' is present in the object. Conversely, below the threshold, the radionuclide ' is not present in the object.

[0095] Preferably, the convolutional neural network implements a Monte Carlo Dropout, which corresponds to the deactivation of certain neurons, according to a probability distribution or randomly, during the training phase and during the use of the neural network. Thus, the network can provide different results from several implementations based on the same input data. This makes it possible to obtain a measurement statistic.

[0096] Figure 7A schematically illustrates a detector model, on the basis of which, using the MCNP code, various training spectra were generated for different radionuclides, namely C, Cl, Sr, and Cs. The detector was assumed to be located 10 mm from the calibration source. Starting from four initial spectra, modeled by considering a surface distribution for each radionuclide, a training database was created by combining the spectra, weighted by various randomly defined parameters: number of radionuclides in the spectrum, proportion of each radionuclide, number of interactions considered in the spectrum, minimum energy, and maximum energy. This generated 500,000 training spectra with a number of interactions considered (number of counts) ranging from 1E3 to 1E7.

[0097] Figure 7B schematically illustrates the training spectra used to parameterize the identification algorithms respectively associated with the different radionuclides. In Figure 7B, the proportions of each radionuclide are shown for each spectrum.

[0098] Figures 8A and 8B show a first example of the application of the identification algorithm. Figure 8A represents a measured spectrum of a 90Sr source. The spectrum was used 100 times by each identification algorithm defined for each of the radionuclides C, Cl, Sr, and Cs. Figure 8B shows the probabilities of presence defined for each radionuclide, in the form of box plots. 90Sr is consistently identified, while the median of the probabilities of presence for C, Cl, and Cs is always less than 0.5.

[0099] Figures 9A and 9B show a second example of the application of the identification algorithm. [Fig. 9A] represents a spectrum [3] of a mixture containing 5% of 14C and 95% of 36C1. The spectrum was used 100 times with each identification algorithm defined for each of the radionuclides C, Cl, Sr, Cs. [Fig. 9B] shows the probabilities of presence defined for each radionuclide, in the form of boxplots. 14C and 36C1 are consistently identified, while the probabilities of presence of 90% and 137% of 14C1 are consistently identified. Sret Cs are always less than 0.5.

[0100] The performance of the identification algorithm was evaluated using test spectra obtained by combining four spectra measured experimentally using sources of C, Cl, Sr, and Cs, respectively. 10,000 test spectra were generated by combining the four measured spectra and varying the following characteristics: relative proportions, activity, number of interactions considered in the spectrum, minimum energy, and maximum energy of the spectra. The 10,000 test spectra were then processed by the identification algorithm.

[0101] Tables 2, 3, 4, and 5 are confusion matrices. The first column represents the ground truth. The first row represents the result of the identification algorithm. 0 means radionuclide absent, 1 means radionuclide present. The matrix values ​​correspond to the detection rates assigned to the radionuclide. The cell corresponding to row 0 and column 1 corresponds to a false positive. The cell corresponding to row 1 and column 0 corresponds to a false negative. The cells corresponding to row 1 and column 1, as well as row 0 and column 0, correspond to correct detections: the value determined by the identification algorithm corresponds to the ground truth. 0 1 0 0.48 0.0025 1 0.016 0.5

[0102] Table 2 (14C) 0 1 0 0.46 0.026 1 0.025 0.49

[0103] Table 3 (36C1) 0 1 0 0.42 0.06 1 0.0088 0.51

[0104] Table 4 (90Sr) 0 1 0 0.43 0.043 1 0.021 0.51

[0105] Table 5 (137Cs)

[0106] The results presented in connection with Figures 8B, 9B, as well as the confusion matrices shown in Tables 2 to 5, attest to the reliability of the identification carried out by submitting a [3 or [3y] spectrum to different algorithms, each algorithm being parameterized to a radionuclide, so as to identify the radionuclide in the spectrum.

[0107] Depending on the output of each identification algorithm, the presence or absence of each radionuclide, to which an identification algorithm is associated, is determined in the object.

[0108] Step 140 Deconvolution

[0109] During this step, the spectrum is subjected to a deconvolution algorithm in order to extract the components respectively associated with each previously identified radionuclide. An important aspect of this step is that the deconvolution is not performed blindly, but on the basis of a priori assumption resulting from the identification step.

[0110] The deconvolution algorithm is based on a deconvolution database, comprising less a detector response, which corresponds to a modeled spectrum for each radionuclide identified by considering a known activity and a known distribution of the radionuclide in the object.

[0111] According to one possibility, the distribution of the radionuclide in the object is known: it can be considered homogeneous, for example, when the object is a sample analyzed in a laboratory, having undergone homogenization. When the measurement is carried out on an object that has been activated, the distribution of the radionuclide can be determined by modeling the neutron flux to which the object was exposed. When the object is made of a non-porous material, for example, a metal, the activity can be assumed to be surface-based.

[0112] When the object is made of a porous material, for example concrete, different distributions of activity can be considered. The activity may, for example, follow a decreasing gradient from the surface of the object. For example, the gradient may have the form of an exponential function decreasing with depth. This type of profile is typical of contamination migration. Siz denotes a depth in From the surface of the object, we can consider that the distribution of activity A(z), according to depth, follows the form:

[0113] A(z) = A(0)e^(l)

[0114] A(0) is the surface activity and is a shape factor of the exponential. 4, whose unit is the inverse of a unit of length, conditions the depth to which the activity is distributed in the object. If the depth zmax is defined as the depth to which the activity is -A- of the surface activity A(0), then:

[0115] inpoo) (2) zmax— ——

[0116] The definition of h or z™** allows us to define a volume in which the activity is assumed to be concentrated.

[0117] Preferably, the database includes, for each identified radionuclide, different modeled spectra, corresponding respectively to different distributions of the radionuclide within the object and to the activities of the identified radionuclides. This provides a database which, for different radionuclides, includes spectra representative of different activity distributions within the object. For example, taking into account an exponential gradient, as described in (1) or (2), the database includes, for different nuclides, modeled spectra corresponding to different parameters Å or znmx.

[0118] The deconvolution algorithm is implemented with representative spectra corresponding to the radionuclides identified in identification step 130. It is understood that the prior identification of the radionuclides makes it possible to select, from the deconvolution database, the modeled spectra corresponding to each identified radionuclide. The deconvolution can then be implemented with a limited number of modeled spectra, restricted to only the identified radionuclides. This avoids deconvolution errors, in particular false positives, i.e., considering a radionuclide as present when it is not. Working with a limited number of radionuclides makes it possible to take into account different distribution profiles for each identified radionuclide, for example, different activity depths zma\ based on a gradient following a decreasing exponential form as described in (1).Thus, deconvolution allows us not only to estimate the activity of each selected radionuclide, but also the depth to which it extends within the object. It should be noted that the depth associated with one radionuclide may differ from the depth associated with another radionuclide.

[0119] The advantage of combining the identification step with the deconvolution step is that, during the deconvolution step, only representative spectra are selected. identified radionuclides. This allows for the consideration of different distributions for each radionuclide. Without radionuclide selection, deconvolution, taking into account different distributions, would be more risky due to the excessive number of spectra to consider.

[0120] The deconvolution database may include modeled spectra for different activity depths, as well as spectra obtained by interpolation between the modeled spectra, for example between two modeled activity depths.

[0121] To perform the deconvolution, the inventors implemented a method based on defining a likelihood function and maximizing it. The activity distributions of each estimated radionuclide correspond to the distributions that maximize the likelihood function. The likelihood function can be maximized using a MLEM-type algorithm, as described in the prior art. The deconvolution can be performed using another method, for example, regression, by implementing a supervised learning algorithm, such as a neural network. In this case, the output of the neural network corresponds to the contribution of each radionuclide to the input spectrum.

[0122] When implementing an MLEM-type method, the spectrum is deconvolved in several iterations, adjusting, with each iteration, the spectra in the deconvolution database corresponding to the different identified radionuclides, whose combination comes as close as possible to the input spectrum. The iterations continue until a convergence criterion is reached, which can be a minimization of a cost function representing the difference between the input spectrum and the spectrum obtained by combining the spectra in the deconvolution database for the identified radionuclides. The cost function can be calculated over all the energies of the spectrum or over predetermined regions of interest. The regions of interest are, for example, determined a priori, based on the variability of the detector response with respect to depth, and this for each radionuclide: cf.Figures 10C and 10D are described below.

[0123] The inventors modeled different spectra corresponding to different activity depths zma\ between 0.1 and 500 mm, with an exponential gradient as defined by (1) and (2), the activity being IBg / g and consisting solely of 90Sr. The object modeled was a concrete wall.

[0124] Fig. 1OA shows the different modeled spectra. Fig. 1OB shows the modeled spectra normalized by the integral of each spectrum. A spectrum showing a uniform distribution of activity over the entire thickness of the wall, the thickness being 500 mm, was also modeled. The spectrum corresponding to the uniform activity was normalized by its integral. Fig. 1OC represents the normalized spectra of [Fig.1OB] divided by the spectrum corresponding to the uniform activity normalized by its integral.

[0125] It is observed that as the activity depth increases, the detector response tends more towards that corresponding to a uniform profile. In Figure 10C, two dashed lines have been drawn, corresponding to a deviation of ±5% from the uniform profile. This deviation corresponds to the minimum acceptable deviation to ensure a good evaluation of the activity distribution. Up to zmax = 40 mm, the profiles shown in Figure 10C deviate from the band by ±5% corresponding to homogeneously distributed activity. Thus, it is considered that for 90Sr, the shape of a spectrum [3] allows discrimination of the maximum depth "max" of activity in the object, up to zmax ≈ 40 mm, with the assumption of a predetermined decreasing activity gradient. Figure

[11] represents a comparison of modeled spectra taking into account: - an exponential activity gradient decreasing according to a maximum depth znwx of 40 mm; - an activity distributed evenly over the entire wall.

[0126] The two spectra overlap, which confirms the conclusion resulting from [Fig.1OC].

[0127] Figure 10D is equivalent to Figure 10C, taking into account Cs activity instead of Sr. Figure 10D was obtained by modeling [3y] spectra. In Figure 10D, two dashed lines were drawn, corresponding to a deviation of ±5% from the uniform profile. Up to zmax = 500 mm, the profiles shown in Figure 10D deviate from the band by ±5% corresponding to a homogeneously distributed Cs activity. Thus, it is considered that for Cs, the shape of a [3y] spectrum allows discrimination of the maximum depth 'mx, at least equal to 500 mm, of the activity in the object, with the assumption of a predetermined decreasing activity gradient.

[0128] Figures 10C and 10D allow the definition of spectral regions of interest for calculating the cost function mentioned above. For example, regions where there is a high degree of variability with depth can be taken into account.

[0129] Figures 12A to 12D show the adjustment, as iterations progress, of spectra resulting from the MLEM algorithm (solid lines) and of a spectrum resulting from a measurement (dashed lines). The measured spectrum corresponds to a spectrum composed of 92% Cs and 8% Sr, with a surface distribution. The measured spectrum comprises 10,000 counts, i.e., 10,000 detected pulses. Figures 12A to 12D correspond respectively to 1, 10, 100, and 10,000 iterations. It can be observed that as and As iterations progress, the spectrum reconstructed by the MLEM algorithm gets closer to the measured spectrum.

[0130] Table 6 shows, for each iteration, the respective percentages determined, based on each spectrum reconstructed by MLEM, iteratively. The second line shows the actual percentages. We observe that the percentages get closer to the actual values ​​with each iteration. C-14 Cl-36 Sr-90 Cs-137 Measurement 0.000 0.000 0.080 0.920 Iteration 1 0.033 0.280 0.361 0.325 Iteration 10 0.037 0.195 0.098 0.905 Iteration 100 0.002 0.013 0.080 0.905 Iteration 1000 0.000 0.000 0.080 0.919

[0131] The inventors applied an MLEM-type deconvolution algorithm for different zmax depths of 90Sr and 137Cs, taking into account a gradient on i Decreasing exponential, and for different Sr and Cs activity ratios. The The configurations are shown schematically in table 7. Configuration zmax 90Sr (mm) zmax 137qs 90Sr / 137Cs (Bq) 1 1 0.5 1000 / 3000 2 5 10 1000 / 333 3 50 5 1000 / 10000 4 400 200 10000 / 1000

[0132] Table 7

[0133] Spectra [3, y and [3y] were modeled for each of these configurations, using MCNP, and then MLEM deconvolution was applied. The deconvolution results are reported in Tables 8 (configuration 1), 9 (configuration 2), 10 (configuration 3) and 11 (configuration 4).

[0134] On each table, zmax (unit mm), the standard deviation relating to the determination of (mm), the activity A(137Cs) or A(90Sr), unit Bq and the standard deviation relating to the determination of this activity (unit Bq) have been reported. Config. 1 90Sr 137Cs zmax o(zmax^ A(90Sr) o(A90Sr) zmax ^-max^ A(137Cs) o(137Cs) P 1.1 0.39 1034.87 76.96 0.5 0.02 2989.58 71.72 Py 1.1 0.38 1017.54 77.22 0.5 0.02 2987.52 60.01 Y 0.7 0.46 3010.69 5.86

[0135] [Tables8] Config. 2 90Sr 137Cs zniax A(90Sr) o(A90Sr) zmax oçzmax^ A(137Cs) o(137Cs) P 4.82 0.00 1011.15 0.62 6.72 5.44 1691.02 1346.24 Py 5.71 0.74 1083.10 86.16 9.03 1.91 326.12 23.68 Y 12.6 8.27 343.86 24.27

[0136] [Tableaux9] Config. 3 90Sr 137Cs ztmx oC™) A(90Sr) o(A90Sr) zmax oçzmax^ A(137Cs) o(137Cs) P 18.38 4.84 1498.93 846.04 4.82 0.00 9680.85 7.22 Py 8.71 12.28 2954.70 1458.95 4.82 0.00 9790.05 0.03 Y 4.53 2.97 9976.34 252.19

[0137] [TableauxlO] Config. 4 90Sr 137Cs zmax A(90Sr) o(A90Sr) zmax ^zmax^ A(137C s) o(137Cs) P 918.85 0.00 14787.14 6.90 538.95 194.86 393.17 61.54 Py 525.37 212.66 10673.22 1919.11 198.52 4.09 997.47 6.62 Y 199.25 72.04 992.20 118.84

[0138] The results presented in Tables 8 to 11 demonstrate the reliability of the algorithm, particularly when considering the [3y] spectrum for 137Cs and the [3] spectrum for 90Sr. Configuration 4 (zmax = 400 mm for 90Sr) corresponds to an activity depth outside the maximum depth of 40 mm defined for 90Sr, as described in relation to [Fig. 100]. Using the [3] spectrum to quantify the activity depth or the activity of 137Cs can lead to an error when the Sr activity is greater, by at least a factor of 3, than the Cs activity, which is the case for configurations 2 and 4.

[0139] Thus, it seems that taking into account a [3y] spectrum is optimal in the case of an Sr and Cs mixture, knowing that these two radionuclides, which are fission products, are frequently encountered together in spent fuel processing facilities, or in cases of radioactive pollution related to spent fuel.

[0140] The tests carried out in connection confirm the ability of the invention to quantify the activity of the identified radionuclides and to estimate the depth of activity of each of these radionuclides.

[0141] Comparison without implementation of the identification algorithm.

[0142] The spectra shown in Figures 8A and 9A were used as input spectra for the deconvolution algorithm, without prior identification of the radionuclides, i.e., without implementing the identification algorithm. For the spectrum corresponding to an activity of 90Sr (see [Fig. 8A]), the implementation of the MLEM algorithm found an activity percentage of 0% 14C, 4% 36Cl, 90% 90Sr, and 6% 137Cs. For the spectrum corresponding to an activity of 14C and 36C1 (see [Fig. 9A]), the implementation of the MLEM algorithm found an activity percentage of 5% 14C, 90% 36C1, 0% 90Sr, and 5% 137Cs. Experimental trial

[0143] Steps 110 to 140 were carried out on graphite samples S taken by coring transversely from horizontal channels CH intended for loading and unloading fuel from the core of a gas-cooled, graphite-moderated reactor. Each sample extended between a face Fl, referred to as the "channel face," adjacent to the fuel channel and oriented towards the interior of the fuel channel, and an opposite face F2, referred to as the "core face," oriented towards the graphite moderator of the reactor. Figure 13 schematically illustrates a channel CH and the location of a core hole from which a sample S is extracted.

[0144] Each graphite sample was cylindrical: 15 mm in diameter and 20 mm thick. The detector was positioned at a distance of 25 mm from one face of each sample.

[0145] Training spectra and a deconvolution database were created, taking into account the main radionuclides likely to be measured: - activation products: 14C, 36C1, 60Co, 133Ba, 152Eu, 154Eu; 137 90 - possible fission products: Cs and Sr.

[0146] The objective of the analysis was to verify the presence of fission products, which indicates the likely occurrence of cladding rupture. To assess the potential contamination by fission products, four identification neural networks were parameterized, addressing distinct minimum activity levels of the Cs and Sr fission products, respectively. The minimum activities were 0.1 Bq.g*, 1 Bq.g*, 10 Bq.g*, and 100 Bq.g*, respectively. The neural networks are subsequently referenced A, B, C, and D, respectively.

[0147] The 4 gratings were trained using 50,000 modeled spectra, with 50% of the spectra considered to show fission products, and 50% of the spectra being considered without fission products. For spectra containing fission products, the Cs / Sr ratio was randomly chosen between 0.25 and 4. A distribution depth was randomly chosen such that it was between 0.1 mm and 5 mm. The shallow depth is justified by the fact that the contamination by fission products is assumed to be dry. The presence of gamma background noise was also modeled, established from a reference measurement, considered representative of the background noise at the time of the measurements. For each training spectrum, a measurement time was defined. A random contribution of background noise was added to the modeled spectra, taking into account the measurement time, in order to simulate a statistical fluctuation of the background noise. The activity of the activation products (14C, 36C1,60Co, 133Ba, 152Eu, 154Eu) was established based on graphite activation calculations.

[0148] 80% of the spectra were used for training. 10% of the spectra were used for validation, to adjust the model. 10% of the spectra were used for testing.

[0149] Several samples taken along two different fuel channels were analyzed. Each spectrum, measured on a graphite sample, was analyzed one hundred times using the four neural networks. On each sample, an analysis was performed on the core-side face and the channel-side face.

[0150] Figures 14A to 14D represent the outputs of the four neural networks for different samples, as a function of distance from a mid-section of the reactor, respectively from spectra measured either on the core face (face F2) ([Fig. 14A] and 14C) or on the channel face (face F1): [Fig. 14B] and 14D. The y-axis corresponds to the probability of contamination, between 0 and 1, calculated by taking into account the 100 analyses for each neural network. Figures 14A to 14D represent the probability of the presence of 137Cs and 90Sr resulting from the 100 outputs of the neural networks for which a significant probability is obtained. The arrows identify the neural network for which the significant probability is obtained.

[0151] The x-axis corresponds to the position of the sample, unit cm, relative to an origin corresponding to the midpoint of the reactor. Figures 14A and 14B show the probabilities obtained as a function of the outputs of the four sample neural networks, used 100 times, for samples taken from a first channel, whose reference was 36-17C. Figures 14C and 14D show the probabilities obtained as a function of the outputs of the four neural networks for samples taken from a second channel, whose reference was 19-13C.

[0152] It is observed that only 1 sample taken from the 36-17C channel exhibits 1^-7 activity Cs+ Sr probable, but weak. On this channel, neural networks C and D conclude that there is no activity.

[0153] On the 19-13C channel, most analyses of the spectra measured on the channel face conclude that there is Cs+ Sr activity.

[0154] On a sample corresponding to the 15 cm contour of channel 36.17C, channel face. The spectrum was deconvolved, taking into account the outputs of the identification neural networks, and without taking into account the identification neural networks. Table 12 shows the results with the deconvolution algorithm running without and with consideration of the presence of Cs and Sr.

[0155] Table 12 shows the activity (A) evaluations of various radionuclides (Bq units) and their relative uncertainties (S), with the deconvolution algorithm running without (I) and with (II) prior implementation of the identification neural networks. In this sample, neural networks A and B, activated 100 times, concluded that Cs and Sr were present in 50% and 70% of cases, respectively (see [Fig. 14B]). The last column shows the relative difference between the activities estimated in (I) and (II). Taking into account 137Cs and 90Sr allows for adjustment of the estimated activity values ​​for the activation products. I II I / II-l A (Bq) 2(%) A (Bq) 2(%) (%) Activation C-14 415029.78 16% 372737.95 11% 11% Cl-36 107.52 15% 96.79 11% 11% Co-60 76.08 16% 68.26 12% 11% Sr-90 144.77 26% 128.48 22% 13% Ba-133 18.54 57% 16.61 53% 12% Cs-137 177.16 28% 158.85 24% 12% Eu-152 93.91 628% 50.71 1109% 85% Eu-154 167.73 16% 150.72 15% 11% Contamination Sr-90 - - 13.48 27% - Cs-137 - - 34.81 42% -

[0156] Table 12

[0157] The deconvolution algorithm was implemented on a sample of channel 19-13 C, canal face, at the 120 cm elevation. On this sample, neural networks A to D, 14”7 on activated 100 times, concluded that Cs and Sr were present in 100% of cases. (see figure 14D).

[0158] The deconvolution algorithm resulted in a minimization of the cost function for: OH 1 4 ”7 - Sr and Cs activities of 246+ 46% Bq and 296 ±31 Bq respectively. OH 1 4”7 - Sretde Cs activity depths of less than 200 pm and 500 pm respectively, which confirms the hypothesis of surface contamination.

[0159] The sample was characterized by high-resolution gamma spectrometry, considered a reference method. The measured 137Cs activity was 358 Bq, with a relative uncertainty of 30%.

[0160] Taking into account the margins of error, the [3y] spectrometry is in agreement with the reference method.

[0161] Fig. 15 represents the measured [3y] spectrum, as well as the contributions, in the Spin input spectrum, of 137Cs and 90Sr and activation products (act), these contributions resulting from the deconvolution algorithm.

[0162] As previously indicated, in the case of intense y-radiation, it is possible to implement the invention from a spectrum [3. For this, we measure: - the spectrum [3y of the object; - the spectrum y, by interposing the screen between the detector and the object.

[0163] The spectrum [3 is obtained by subtracting the spectrum [3y and the spectrum y, possibly taking into account a difference in the acquisition periods.

[0164] Figures 16A to 16C illustrate this possibility. Figure 16A shows a measured [3y] spectrum. Figure 16B shows a measured y spectrum. Figure 16C shows the [3] spectrum calculated by subtracting the [3y] spectrum from the y spectrum.

[0165] Although described in connection with a non-destructively acquired spectrum, the method can be generalized to the analysis of spectra exhibiting a non-negligible [3] component. These may be laboratory measurement methods, for example, liquid scintillation methods. In this case, the object is a sample placed in front of the [3] spectrometer.

Claims

Demands

1. A method for characterizing an object, the object comprising at least one radionuclide emitting [3] radiation, the method comprising: - a) arranging a detector (10) facing the object (2), the detector being configured to acquire a spectrum, representing a distribution of the energy released, in the detector, by the radiation emitted by the object; - b) detecting the radiation emitted by the object, by the detector, during an acquisition period, and acquiring a spectrum of the detected radiation; - c) from the spectrum of the radiation detected by the detector, forming an input spectrum, comprising a [3] component, which corresponds to a distribution of the energy released by the [3] radiation in the detector;- d) application of an identification algorithm, associated with a radionuclide, to the input spectrum, the identification algorithm being configured to determine the presence of the radionuclide, to which the identification algorithm is associated, in the object, step d) being repeated for different radionuclides, implementing different identification algorithms; - e) as a function of d) identification of each radionuclide contained in the object; - f) application of a deconvolution algorithm of the input spectrum so as to estimate a contribution of each radionuclide, identified in e), in the input spectrum; - g) for each radionuclide identified in e), from the contribution of the radionuclide, in the input spectrum, estimated in f), estimation of an activity and / or a depth to which the radionuclide extends in the object;steps d) to g) being implemented by a processing unit from the input spectrum.

2. A method according to claim 1, wherein in step d), each identification algorithm is an artificial intelligence identification algorithm associated with each radionuclide, at least two different radionuclides being respectively associated with two different identification algorithms.

3. A method according to claim 2, wherein each identification algorithm is a neural network.

4. A method according to any one of the preceding claims, wherein the deconvolution algorithm is based on a deconvolution database comprising at least one representative spectrum of each radionuclide identified in e).

5. A method according to claim 4, wherein - the deconvolution database includes, for the same radionuclide, different spectra representing different distributions of the radionuclide in the object; - step g) includes a determination of the distribution of the radionuclide in the object.

6. A method according to claim 5, wherein: - the deconvolution database includes, for the same radionuclide, different spectra representing different depths of the radionuclide in the object, from a surface of the object facing the detector; - step g) includes a determination of the depth to which the radionuclide extends in the object.

7. A method according to any one of the preceding claims, wherein at least one radionuclide associated with an identification algorithm is a pure [3] emitter.

8. A method according to any one of the preceding claims, wherein the detector comprises an organic scintillator-type material for detecting the radiation emitted by the object.

9. A method according to any one of claims 1 to 7, wherein the detector comprises a volume of semiconductor or inorganic scintillator of thickness less than 10 mm, disposed facing the object, the thickness being considered in a direction normal to the object.

10. A method according to any one of the preceding claims, wherein the object exhibits a natural activity, step c) comprising: - estimation of a spectrum of natural activity of the object; - subtraction of the natural activity spectrum of the object from the spectrum acquired during step b) in order to form the input spectrum.

11. A method according to any one of the preceding claims, wherein the detector includes a removable screen (15), configured to be interposed between the detector and the object, the method comprises: - acquisition of a background spectrum, during which the screen is interposed between the detector and the object; - step c) comprises subtracting the background spectrum from the spectrum acquired during step b) to form the input spectrum.

12. Detection device (1), comprising a detector (10), configured to acquire a spectrum of radiation [3] emitted by an object (2), the spectrum representing a distribution of the energy released, in the detector, during interactions of ionizing radiation in the detector, the device comprising a processing unit (20) configured to carry out steps d) to f) of a method according to any one of the preceding claims.

13. A detection device according to claim 12, wherein the detector comprises an organic scintillator-type material for detecting the radiation emitted by the object.

14. Device according to claim 12, wherein the detector comprises a volume of semiconductor or inorganic scintillator of thickness less than 10 mm.

15. Device according to any one of claims 12 to 14, wherein the detector includes a removable screen (15), configured to be interposed between the detector and the object.

Images