METHOD FOR THE EXTRACTION OF DISCRETE NEUTRON COMPONENTS FROM PHOTONUCLEAR REACTIONS USING A TRAINED NEURAL NETWORK

Description

TECHNICAL FIELD

[0001] The technical field of the invention is that of the exploitation of neutron spectra using a neural network to extract useful information, for example in the context of the detection of dangerous, toxic or illicit substances contained in an object. PREVIOUS STATE OF THE ART

[0002] It is known that neutron and photon radiation sources can be used to detect, through specific nuclear reactions, the presence of certain light elements (namely nitrogen, carbon, oxygen, and chlorine) in various materials. This allows for the detection of dangerous, toxic, or illicit substances such as explosives, certain toxic gases (chlorine-containing gases), or narcotics, contained or concealed within an object being examined (which could, for example, be a container, package, or suitcase).

[0003] Among the known detection methods, we can mention the active neutron interrogation (INA) method, the so-called "tagged" photon method and the active photon interrogation (IPA) method.

[0004] The active neutron interrogation (INA) method utilizes neutron-induced reactions. These reactions are used to detect light elements (C, N, O, Cl) in materials, notably using the particle-associated particle (PAP) technique. This detection is based on measuring the gamma radiation produced by these reactions. However, accurately measuring gamma spectra is generally difficult due to experimental conditions, often characterized by significant background noise. Indeed, gamma radiation is widespread because it can be easily produced by a multitude of mechanisms (reactions with chemical elements other than those of interest, radioactive decay of radioisotopes created by activation or naturally present in materials, etc.), leading to complex spectra that make it difficult to separate the different contributions of the chemical elements.

[0005] The so-called "tagged" photon technique is based on the detection of neutrons coinciding with electrons from a bremsstrahlung radiation source. This technique allows the extraction of neutron components for each energy of the bremsstrahlung photons. However, it requires the implementation of complex methods with specialized high-frequency electronics for detecting coinciding particles, as well as heavy, cumbersome, and expensive equipment, particularly magnetic dipoles, which are not well-suited to the development of a compact detection device for homeland security applications.

[0006] Finally, the active photon interrogation (IPA) method is a detection technique based on the use of photon sources. It utilizes photonuclear reactions whose cross sections (probabilities) are well known for most nuclei of light elements such as nitrogen, carbon, oxygen, and chlorine.

[0007] Photonuclear reactions have the particularity of being threshold reactions, meaning that they can only take place if the initial photons have an energy greater than a certain value, which is specific to each nucleus.

[0008] In the case of nitrogen, for example, which is detected from neutrons produced by (y,n) photonuclear reactions, the minimum energy of the photons must be greater than 10 MeV.

[0009] In the case of chlorine detection by photon irradiation (PAI), current techniques are based on the creation, through (γ,ν) photonuclear reactions, of radioisotopes whose decay leads to gamma radiation emission. Spectrometric measurement of this gamma radiation allows for the identification of these radioisotopes and thus the tracing back to the elements responsible for their production. However, the PAI method coupled with gamma spectrometry is very complex to implement given the significant gamma noise generated by numerous mechanisms.

[0010] The photons used in active photon interrogation cannot be obtained using conventional methods based on radioactive sources, as these do not allow for sufficient energy levels. Therefore, it is necessary to have high-energy photon sources that are sufficiently intense to achieve a satisfactory signal-to-noise ratio. Furthermore, the source must be as compact as possible for deployment in the field, for example, in applications such as the detection of illicit materials.

[0011] Traditionally, the high-energy photons (greater than or equal to 6 MeV) used to implement the IPA method are bremsstrahlung photons, which are produced by an electron accelerator. As is known, in an electron accelerator, a sufficiently energetic electron beam (several MeV) strikes a target made of a material with a high atomic number Z, resulting in bremsstrahlung radiation. The disadvantage of using an electron accelerator lies in the fact that the resulting photon source (bremsstrahlung photons) is characterized by a continuous energy spectrum. However, only a few percent of the photons obtained by this method (i.e.,those with an energy greater than 6 MeV required to induce photofission reactions (y,f)) can be used, the vast majority of photons being the cause of unnecessary irradiation of the substances present in the object to be probed.

[0012] Ideally, such a source should be based on the use of monoenergetic photons obtained, for example, from a low-energy proton source, in order to identify specific signatures (or fingerprints) composed of discrete neutron components in the spectra of neutrons emitted in (γ,n) reactions induced by these photons on light nuclei (C, N, O, Cl). Discrete neutron components will be the signature of the presence of certain light elements; the position of these discrete neutron components will tell us about the nature of the light elements in question, and the height of these discrete neutron components will tell us about their relative quantity.

[0013] An example of active photon interrogation is described in MCFEE JE ET AL: "Photoneutron spectroscopy using monoenergetic gamma rays for bulk explosives detection", NUCLEAR INSTRUMENTS & METHODS IN PHYSICS RESEARCH (2013).

[0014] However, previous work by the inventors aimed at studying the implementation of a mono-energetic photon source has shown many limitations related, in particular, to the intensity that such a source could have and its isotropy.

[0015] In the context of the present invention, the inventors focused on IPA detection and sought to design a method for exploiting neutron spectra from photonuclear reactions involving multiple photon energies. In other words, the inventors sought to exploit neutron spectra obtained by irradiation with bremsstrahlung photons or with multiple lines (i.e., at least two lines). DESCRIPTION OF THE INVENTION

[0016] This goal is achieved in particular through the use of a trained neural network.

[0017] The invention thus relates to a method for extracting discrete neutron components, resulting from photonuclear reactions between photons and at least one chemical element to be detected, from a neutron spectrum resulting from photonuclear reactions obtained by irradiation, of a material comprising said at least one chemical element to be detected, with a photon source, at least one energy of the photons of the source being greater than the photonuclear reaction threshold of said chemical element to be detected, using a multilayer and multichannel neural network having an architecture with a convolution stage and a deconvolution stage; in which, if said at least one chemical element to be detected is chosen from carbon, nitrogen, oxygen or chlorine and if the photon source is a first source of photons at several energies, the process includes: a preliminary step of training the neural network by supervised learning, which is carried out by iteration from: ▪ first neutron spectra, from photo-nuclear reactions obtained by irradiation of the material with bremsstrahlung photons;and ▪ second neutron spectra, each of the second neutron spectra being derived from photonuclear reactions induced by irradiation of the material with photons of the same energy, which is greater than the photonuclear reaction threshold of the chemical element to be detected, and less than or equal to the maximum energy of the photons from said first multi-energy photon source, the first and second spectra being fictitious spectra established by a Monte Carlo simulation, the first spectra being provided as input to the neural network, and the second spectra being used to calculate a cost function intended to adjust weights and biases of the neural network; and a prediction step, using the trained neural network, comprising: ▪ the provision, as input to the trained neural network, of the neutron spectrum derived from photonuclear reactions induced by irradiation of the material by said first photon source;and ▪ if the irradiated material includes, in its composition, at least one chemical element chosen from carbon, nitrogen, oxygen or chlorine, the obtaining, at the output of the trained neural network and for at least one specific energy of interest of said first photon source, of a predicted neutron spectrum, comprising discrete neutron components resulting from photo-nuclear reactions of said chemical element chosen with said specific energy of interest;or wherein, if the material to be irradiated comprises said at least one chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, and at least one other different chemical element, capable of emitting neutrons and which is not carbon, nitrogen, oxygen or chlorine, and if the photon source is a second single-energy or multi-energy (line or Bremsstrahlung) photon source, the process comprises: a preliminary step of training the neural network by supervised learning, which is carried out by iteration from: ▪ first neutron spectra, from photonuclear reactions obtained by irradiation of the material with Bremsstrahlung, line or single-energy photons;and ▪ second neutron spectra, each of the second neutron spectra being derived from photonuclear reactions induced by irradiation of the material with photons of energy greater than the photonuclear reaction threshold of said at least one chemical element to be detected and greater than the photonuclear reaction threshold of said other different chemical element; the first and second spectra being fictitious spectra established by a Monte Carlo simulation, the first spectra being provided as input to the neural network, and the second spectra being used to calculate a cost function intended to adjust weights and biases of the neural network; and a prediction step, using the trained neural network, comprising: ▪ the provision, as input to the trained neural network, of the neutron spectrum derived from photonuclear reactions induced by irradiation of the material by said second photon source;and ▪ if the irradiated material contains, in its composition, at least one chemical element chosen from carbon, nitrogen, oxygen or chlorine, obtaining, at the output of the trained neural network and for at least a specific energy of interest from said second photon source, a predicted neutron spectrum, comprising discrete neutron components resulting from photonuclear reactions of said chemical element to be detected with said specific energy of interest. ;

[0018] In the first case where at least one chemical element to be detected is chosen from carbon, nitrogen, oxygen or chlorine and if the photon source is a first multi-energy photon source (hereafter referred to as the "first case"), then discrete neutron components can be extracted from a neutron spectrum produced by a bremsstrahlung or line photon source, as if the neutron spectrum had been produced by irradiation with a single-energy photon source (i.e., of a single energy or mono-energetic).

[0019] In the second case where the material to be irradiated comprises said at least one chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, and at least one other different chemical element, capable of emitting neutrons and which is not carbon, nitrogen, oxygen or chlorine, and if the photon source is a second single-energy or multi-energy (line or bremsstrahlung) photon source (hereinafter referred to as the "second case"), discrete neutron components created by said at least one light element can be extracted from a neutron spectrum produced by irradiation of a complex assembly (material comprising at least one light element and object in which it is located) using a photon source (whether the photons are bremsstrahlung, multi-line or single-energy).

[0020] In both scenarios, only the training stage differs. The spectra provided for calculating the cost function are neutron spectra from mono-energetic irradiations in the first case, and spectra from irradiation of single elements in the second case. A "single element" is defined as the irradiated material consisting of only one element (or, in other words, a pure element), meaning that the element is not mixed with other different elements within the material.

[0021] As an illustration, in the second case, during the training phase, the neural network is provided with neutron spectra obtained from the irradiation of various samples or substances with Bremsstrahlung photons as network input. Learning for neural networks consists of calculating parameters such that the network's outputs, for the examples used during training, are as close as possible to the "desired" outputs (here: neutron spectra obtained from irradiation with monoenergetic photons). During training, the goal is therefore to minimize the gap between the network's actual responses and the "desired" responses by modifying the parameters in successive steps. The network will then be "shown" these "desired" responses for each neutron spectrum obtained from the irradiation of various samples or substances with Bremsstrahlung photons, and provided as network input.These "desired" responses do not actually constitute the input data per se, as are the neutron spectra resulting from irradiation with Bremsstrahlung photons.

[0022] According to an advantageous embodiment of the invention, the specific energy of interest is greater than or equal to 17 MeV.

[0023] In fact, during the prediction stage, a neutron spectrum can be obtained for a specific energy of interest, or several spectra for two or more different source energies. However, to minimize computational costs, it is preferable to predict a neutron spectrum for a single photon energy from the source, chosen according to the chemical element to be detected. In the context of explosives detection, where nitrogen detection is the focus, the chosen energy is 17 MeV, which is the most suitable for nitrogen detection.

[0024] The invention also relates to a method for detecting a chemical element chosen from carbon, nitrogen, oxygen, or chlorine, the method comprising: the irradiation, by a source of photons at several energies (the photons can therefore be bremsstrahlung or line), of a material comprising said chemical element; the detection of the neutrons emitted by the irradiated material and the acquisition of a corresponding neutron spectrum; the extraction, from the acquired neutron spectrum, of the discrete neutron components due to said chemical element to be detected, by implementation of the process for extracting discrete neutron components as described above, in the case where the photon source is the first source; comparison of the extracted discrete neutron components with a library of discrete neutron components associated with carbon, nitrogen, oxygen and chlorine, and identification of a match, whereby the presence of said chemical element is deduced.

[0025] The invention also relates to a method for detecting a material to be detected contained in an object to be probed, the material to be detected comprising at least one chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, and the object to be probed comprising at least one other different chemical element, capable of emitting neutrons and which is not carbon, nitrogen, oxygen or chlorine, the method comprising: irradiation, by a photon source (which may be a Bremsstrahlung, line or single-energy photon source), of the object to be probed and the material to be detected that it contains; detection of the neutrons emitted by the object to be probed and the material to be detected and acquisition of a corresponding neutron spectrum; extraction, from the acquired neutron spectrum, of the discrete neutron components due to said at least one chemical element, by implementation of the process for extracting discrete neutron components as described above in the case where the photon source is the second source; comparison of the extracted discrete neutron components with a library of discrete neutron components associated with materials comprising said at least one first chemical element, and identification of a match, whereby the presence of said material to be detected is deduced.

[0026] In the various aspects of the invention, the use of a neural network makes it possible to do away with the need to use a mono-energetic source in favor of using a more conventional, more intense and easily accessible source such as, for example, a source based on a linear electron accelerator generating a continuous energy spectrum of photons (bremsstrahlung radiation) which have the advantage of being strongly oriented forwards (giving a directional source).

[0027] In the context of this invention, we refer to neutron components as "discrete." This means that the neutron components form peak structures of varying widths, clearly distinct from the continuum of neutron spectra. The full width at half maximum (FWHM) of these peaks will depend, in particular, on the energy resolution of the spectra. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Other aspects, objectives, advantages, and features of the invention will become clearer upon reading the following detailed description of preferred embodiments thereof, given by way of non-limiting example, and made with reference to the accompanying drawings in which: there figure 1 represents, schematically, an assembly to illustrate one embodiment of the invention; the figures 2a to 2c represent respectively a bremsstrahlung photon spectrum, as obtained after target 2 of the figure 1 , a spectrum of neutrons produced by the irradiation of an explosive material with bremsstrahlung photons ( figure 2b ), as obtained by detector 5 of the figure 1 , and a predicted neutron spectrum for irradiation of this same explosive material with 17 MeV photons ( figure 2c), as obtained at the output of a trained neural network 6 according to the invention, with identification of the light elements present in its composition; the figures 3a and 3b schematically represent a multilayer and multichannel neural network ( figure 3a ) and a diagram to describe an architecture of this neural network, with a convolution stage and a deconvolution stage ( figure 3b ) ; there figure 4 represents a triangle of the normalized proportions of nitrogen, carbon and oxygen of several substances or products, with an identification of the groups falling into the category of illicit materials. DETAILED DESCRIPTION OF SPECIFIC METHODS OF IMPLEMENTATION

[0029] According to a first aspect of the invention, the process according to the invention makes it possible to extract neutron components from photonuclear reactions for each energy of photons composing a source whose photons have a continuous energy spectrum (bremsstrahlung) or a multi-energy spectrum (linear). The process according to the invention makes it possible to replace the existing "tagged" photon method with the use of a multilayer and multichannel neural network, thereby eliminating the need for specialized electronics and magnetic dipoles required by the "tagged" photon method. The process according to the invention also eliminates the need to use a single-energy source.Obtaining neutron spectra for each photon energy is achieved here in post-processing, without additional physical resources beyond those classically used in IPA detection, thus limiting the costs and bulk of the light element detection device used.

[0030] Thus, the use of a deep learning network, known as a "neural network" according to the invention, makes it possible to recover the neutron contributions in the total spectrum generated by photons from a bremsstrahlung or line source for the different energies of the source photons. The identification of light elements is therefore made possible within the various extracted neutron contributions.

[0031] According to a second aspect of the invention, the process according to the invention also makes it possible to extract the neutron components due to light elements from a complex matrix (which can be the material in which the light element(s) are located and / or the object in which this material is located).

[0032] Whatever the aspects of the invention, the specificity of the proposed approach consists in the learning and use of a multilayer and multichannel neural network trained so as to be able to predict the neutron components for each energy of bremsstrahlung or line photons, when provided with a neutron spectrum from the irradiation of materials by bremsstrahlung or line photons, or when seeking to analyze neutron spectra from any type of photon source (mono-energetic, line or bremsstrahlung) to extract the neutron contributions of the light elements of a complex matrix.

[0033] Neural network 6 is multilayered and multichannel ( figure 3a ).

[0034] It is multilayered, because it has an input layer, an output layer and one or more intermediate layers, called hidden layers.

[0035] It is multichannel, as each layer contains several channels. Each channel represents an energy interval of the spectrum. Each energy interval of a neutron spectrum corresponds to an input node (7) and an output node (8). For example, if we want to reconstruct a neutron spectrum with energies ranging from 1 to 11 MeV, we can have 100 channels for each energy with a sampling increment (step size) of 0.1 MeV. The greater the number of channels, the smaller the sampling increment, and the more precise the resolution.

[0036] Neural network 6 has a two-part architecture. It is composed of a convolution stage and a deconvolution stage ( figure 3b ).

[0037] The first stage extracts features and patterns from the neutron spectra fed into input 7 of the lattice 6, reducing the data. The second stage reconstructs a spectrum from the features extracted by the first stage, while preserving the original data dimensions.

[0038] As is known, supervised learning of the neural network relies on the following series of steps, for n pairs of input-target values, n being an integer that depends on the energy resolution of the spectra and corresponds to the number of energy intervals in the spectra: 1- presentation to the network of one of the pairs "input spectrum - target spectrum" (hereinafter "input-target spectra") of the n intervals; 2- calculation of a prediction of the network for the expected target; 3- use of a cost function to calculate the difference between the prediction (output) of the network and the target spectrum; 4- use of the neural network learning algorithm to adjust the weights and biases of the network, so that the network produces better predictions at each presentation of a pair of input-target spectra.

[0039] It should be noted that steps 1 to 4 constitute a single iteration (or learning cycle).

[0040] Steps 1 through 4 are repeated for a number of iterations until the network begins to produce sufficiently reliable results (i.e., outputs that are close enough to the targets given the input values). The number of iterations required to train the neural network is unknown. a priori but is generally several hundred, or even a thousand times greater.

[0041] In the present case, for each pair of input-target spectra, the input is a first fictitious spectrum (by reference to the designation "first neutron spectra" used to define the invention), for example a neutron spectrum produced by bremsstrahlung photons and, for the target, a second fictitious spectrum (by reference to the designation "second neutron spectra" used to define the invention) produced by a single-energy source.Thus, as an example, the supervised learning of the neural network is carried out by iteration from neutron spectra from irradiation by bremsstrahlung photons ("first fictitious spectra") which are provided as input 7 of the network and from neutron spectra produced by photons at a single energy ("second fictitious spectra"), which are used to calculate a cost function which will allow the adjustment of the weights and biases of the network 6 so that its prediction is as close as possible to the spectra from the expected deconvolution.

[0042] For training, we use fictitious spectra generated by Monte Carlo simulation, which allows us to easily obtain a multitude of input-target pair data. Data augmentation techniques are applied to multiply the spectra by slightly and randomly varying the intensity of each channel in order to obtain a sufficient number of different spectra for network training. For example, we can use the Monte Carlo N-particle Transport Code (MCNP™) software, which allows us to model nuclear physics processes using the Monte Carlo method.

[0043] Once its training is complete, the weights and biases of the adjusted network are recorded and the neural network can be used for forecasting purposes without a target spectrum being provided to it.

[0044] For example, a neutron spectrum obtained by irradiation with a photon source at several energies is provided as input to the driven grating, which will deconvolve it and digitally extract the neutron components, for a specific energy of interest (we can also extract the neutron components for several energies of the source, or even for each energy of the photons of the source).

[0045] Analysis of the predicted neutron spectra obtained from the neural network allows, for example, the detection of light elements in a material irradiated by bremsstrahlung photons. This detection is based on peaks that become much more easily identifiable and serve as a signature. Each peak in the predicted spectrum corresponds to an energy level of a residual nucleus produced by the (y,n) reaction, which is therefore an isotope of the initial nucleus. The energy levels specific to each residual nucleus allow the identification of the element present in the material to be detected.4

[0046] To illustrate the invention, we will implement the process according to the invention using a bremsstrahlung photon source to detect TNT (of formula C 7 H 5 N 3 O 6 ) contained in a wooden object (cellulose, of formula (C 6 H 10 O 5 )n) by inducing photonuclear reactions there.

[0047] By referring to the figure 1An example of a setup for implementing the invention is schematically illustrated. A linear electron accelerator 1 generates electron radiation (e-) which is directed towards a specific target 2 (for example, a thin gold target), producing photon radiation (γ) which is directed onto the object to be probed 3 and onto the material to be detected 4 that it contains. The interaction of the γ photons with the material to be detected 4 produces neutrons, which are detected by a detector 5 enabling neutron spectrometry. The detector 5 can, for example, be chosen from Bonner sphere spectrometers or scintillators.

[0048] The neutron spectrum obtained at the output of detector 5 is introduced, at input 7, into the trained neural network 6, and at the output 8 of this neural network, a predicted neutron spectrum is obtained in which the extracted discrete neutron compositions are present.

[0049] As an example, the object to be probed 3 may be a wooden or cardboard package containing, as the material to be detected 4, a substance that is an explosive material, for example TNT; the bremsstrahlung photon spectrum obtained after the target 2 is illustrated in the figure 2a ; the neutron spectrum obtained at the output of detector 5 by irradiation of the explosive material with bremsstrahlung photons is illustrated in the figure 2b The neutron spectrum predicted by the trained neural network 6 (and obtained at output 8 of neural network 6) is illustrated in the figure 2cThe predicted spectrum is in fact identical to a spectrum obtained for irradiation of the explosive material with 17 MeV photons, which is a particularly interesting energy in the context of nitrogen (N) detection for illicit materials; the different peaks (which correspond to the discrete neutron components extracted by the process) have been identified and assigned to the light elements 16< O, 14< N and 12< C.

[0050] The location and height of the spikes form a specific signature, like a fingerprint, that allows us to identify the presence of TNT. This makes it possible to determine that the package contains TNT.

[0051] In this example of implementation of the invention, a conventional source producing bremsstrahlung photons is used to induce photonuclear reactions in the material to be probed 3, but a single-energy or line source could have been used.

[0052] The invention finds application in the field of detection by the active photonic interrogation (IPA) method of illicit materials in containers and packages.

[0053] The detection of light elements can be achieved based on specific signatures in the spectra of neutrons extracted by the neural network.

[0054] It is also possible to determine the concentrations of light elements such as nitrogen, carbon, oxygen, or chlorine from the height of the discrete neutron components of these neutron spectra. As shown by the figure 4 , knowledge of the proportion of these light elements (C, N, O, Cl) makes it possible to know the nature of the product or substance concealed in the object being examined (package, container or other).

[0055] Nitrogen, for example, is present in most explosives and can serve as a marker to detect an explosive material of trinitrotoluene (TNT, chemical formula C7H5O6N3), cyclotrimethylenetrinitramine (RDX) or any other conventional explosive, such as nitroglycerin (C3H5O9N3), or C4 (C4H6O6N6).

[0056] Cocaine (C17H21NO4) can also be detected.

[0057] As for toxic gases, their presence can be suspected by the detection of chlorine, present for example in phosgene (COCl 2).

Claims

1. Method for extracting discrete neutron components, from photo-nuclear reactions between photons and at least one chemical element to be detected, from a neutron spectrum from photo-nuclear reactions obtained by irradiating a material comprising said at least one chemical element to be detected with a photon source, at least one energy of the photons of the source being greater than the photo-nuclear reaction threshold of said chemical element to be detected, using a multilayer and multichannel neural network having an architecture with a convolution stage and a deconvolution stage; wherein, if said at least one chemical element to be detected is chosen from carbon, nitrogen, oxygen or chlorine and if the photon source is a first multi-energy photon source, the method comprises: - a preliminary step of training the neural network by supervised learning, which is performed by iteration using: ▪ first neutron spectra, from photo-nuclear reactions obtained by irradiating the material with bremsstrahlung photons; and ▪ second neutron spectra, each of the second neutron spectra being from photo-nuclear reactions induced by irradiating the material with photons of the same energy, which is greater than the photo-nuclear reaction threshold of the chemical element to be detected, and less than or equal to the maximum energy of the photons of said first multi-energy photon source, the first and second spectra being notional spectra established by a Monte-Carlo simulation, the first spectra being supplied at the input of the neural network, and the second spectra serving to compute a cost function intended to adjust the weights and biases of the neural network; and - a prediction step, using the trained neural network, comprising: ▪ supplying, at the input of the trained neural network, the neutron spectrum from photo-nuclear reactions induced by irradiating the material with said first photon source; and ▪ if the irradiated material comprises, in its composition, at least one chemical element chosen from carbon, nitrogen, oxygen, or chlorine, obtaining, at the output of the trained neural network and for at least one specific energy of interest of said first photon source, a predicted neutron spectrum, comprising discrete neutron components from photo-nuclear reactions of said chosen chemical element with said specific energy of interest; or wherein, if the material to be irradiated comprises said at least one chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, and at least one other different chemical element, capable of emitting neutrons and which is not carbon, nitrogen, oxygen or chlorine, and if the photon source is a second, single-energy or multi-energy photon source, the method comprises: - a preliminary step of training the neural network by supervised learning, which is performed by iteration using: ▪ first neutron spectra, from photo-nuclear reactions obtained by irradiating the material with bremsstrahlung, line or single-energy photons; and ▪ second neutron spectra, each of the second neutron spectra being from photo-nuclear reactions induced by irradiating the material with photons of an energy greater than the photo-nuclear reaction threshold of the chemical element to be detected and greater than the photo-nuclear reaction threshold of said other different chemical element; the first and second spectra being notional spectra established by a Monte-Carlo simulation, the first spectra being supplied at the input of the neural network, and the second spectra serving to compute a cost function intended to adjust the weights and biases of the neural network; and - a prediction step, using the trained neural network, comprising: ▪ supplying, at the input of the trained neural network, the neutron spectrum from photo-nuclear reactions induced by irradiating the material with said second photon source; and ▪ if the irradiated material comprises, in its composition, at least one chemical element chosen from carbon, nitrogen, oxygen or chlorine, obtaining, at the output of the trained neural network and for at least one specific energy of interest of said second photon source, a predicted neutron spectrum, comprising discrete neutron components from photo-nuclear reactions of said chemical element to be detected with said specific energy of interest.

2. Method according to claim 1, wherein the specific energy of interest is greater than or equal to 17 MeV.

3. Method for detecting a chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, the method comprising: - irradiating, with a multi-energy photon source, a material comprising said chemical element; - detecting the neutrons emitted by the irradiated material and acquiring a corresponding neutron spectrum; - extracting, from the neutron spectrum acquired, discrete neutron components due to said chemical element to be detected, by implementing the method according to claim 1, the photon source being the first source; - comparing the discrete neutron components extracted with a library of discrete neutron components associated with carbon, nitrogen, oxygen and chlorine, and identifying a correlation, whereby the presence of said chemical element is inferred.

4. Method for detecting a material to be detected contained in an object to be probed, the material to be detected comprising at least one chemical element to be detected chosen from carbon, nitrogen, oxygen or chlorine, and the object to be probed comprising at least one other different chemical element, capable of emitting neurons and which is not carbon, nitrogen, oxygen or chlorine, the method comprising: - irradiating, with a photon source, the object to be probed and the material to be detected contained therein; - detecting the neutrons emitted by the object to be probed and the material to be detected and acquiring a corresponding neutron spectrum; - extracting, from the neutron spectrum acquired, discrete neutron components due to said at least one chemical element to be detected, by implementing the method according to claim 1, the photon source being the second source; - comparing the discrete neutron components extracted with a library of discrete neutron components associated with materials comprising said at least one first chemical element, and identifying a correlation, whereby the presence of said material to be detected is inferred.