Background

Existing neutron detectors fall into two main categories. At one end are simple detectors such as those used to monitor radiation exposure, which are capable of providing only a count of the number of neutrons incident on the detector. However, such detectors cannot provide any information as to an energy distribution of the incident neutrons, or as to a direction from which the neutrons came. Moreover, they can be prone to false alarms and can require time-consuming verification of the measurements.

At the other end are complex detectors which rely on the use of intensive offline calculations to determine a point of origin of individual neutrons. Such detectors find utility in many applications, including neutrino detectors, as described in US

2014/ 0306117 for example. Neutron detectors are also commonly used in applications such as neutron imaging, but these detectors are large and complex, and require significant processing resources to analyse the behaviour of individual neutrons.

There is consequently a need for a portable and easy to use, but direction sensitive, neutron detector for use in applications such as nuclear security. Any such detector must be sufficiently simple and cost-effective to be assembled and operated in the field. In particular, it is desirable to have a portable and easy to use neutron detector, which limits the number of false alarms whilst providing, in the case of an alarm, sufficient information to allow appropriate countermeasures to be taken.

Such a detector also has applications in the medical sector, and in areas of public safety such as radiation protection where, for example, a more accurate indication of radiation dose per body organ could be given than with conventional radiation monitors. Summary

In a first aspect, a method is provided as defined in the appended independent method claim. In a second aspect, a neutron detector is provided as defined in the appended independent apparatus claim. In the following description, a method for determining a direction to a neutron source external to a neutron detector comprising an n x m array of detector units, where n and m are at least 2, is described. The method comprises: measuring a number of neutron capture events at each detector unit, each detector unit being optically isolated from the other detector units in the array and each detector unit comprising a scintillation element arranged to absorb a neutron and output photons in a neutron capture event; determining a distribution of neutron capture events across the array of detector units; and interpreting the determined distribution to infer a direction to the neutron source. The output photons can be optical, or low energy, photons, i.e. photons having an energy of less than or equal to tooeV, and a wavelength greater than or equal to tonm. In contrast to known approaches, which analyse the trajectory of each individual incoming neutron, the present invention makes use of the aggregate total of neutron capture events. The distribution of neutron capture events across the detector follows a probabilistic model; individual neutrons might be captured by the scintillation element of any individual detector unit, so counting individual capture events is not a useful indicator of direction to a neutron source. However, the overall, aggregated, distribution of neutron capture event does provide information regarding the direction of the neutron source. This approach obviates the need for complex analysis of individual neutron capture events and, as such, computational load is reduced and a simple and easy to use neutron detector is provided, which can be deployed as a portable, field-based, detector.

The use of an at least 2 x 2 array of detector units, i.e. an array having at least two detector units in a first dimension of the detector (such as the width) and at least two detector units in a second dimension of the detector (such as the depth) facilitates effective determination of a direction to the neutron source by providing a distribution of neutron capture events in two dimensions. The distribution of neutron capture events is dependent upon the incidence direction of the incoming neutrons on the array; by aggregating the number of neutron capture events across the array, the direction to the external neutron source can be determined.

Optionally, the neutron detector comprises an n x m array of detector units, where one or both of n and m is 3, optionally where one or both of n and m is more than 3. For example, the detector may comprise: a 2x3 array of detector units, a 3x3 array of detector units, a 2x4 array of detector units, a 3x4 array of detector units, etc.— n and m may be any suitable number, depending on the specific application of the detector. Optionally, the method further comprises identifying a type of neutron source based on the determined distribution. The distribution of neutron capture events is affected not only by the direction of the neutron source relative to the detector, but by the energy distribution, or spectrum, of the incoming neutrons (since higher energy neutrons will, on average, travel further through the detector before being absorbed by a scintillator element). The energy spectrum of the incoming neutrons is correlated with the type of neutron source - i.e. whether the source is shielded or unshielded, and what specific neutron emitter the source is. Analysis of the determined distribution of neutron capture events can thus advantageously facilitate identification of the type of neutron source, which can enable appropriate protocols to be enacted or necessary

countermeasures to be taken. The effectiveness of the neutron detector for security related applications is therefore increased.

In particular, this identification is facilitated by the above described neutron detector architecture; the 2 x 2 array of detector units enables utilisation of the shielding effect of detector units at the fore of the detector on other detector units within the detector to determine a distribution of neutron energy. For example, a detector unit in direct line of sight to the neutron source will, on average, have a higher rate of absorbed neutrons than the detector unit(s) behind it, because the detector unit in direct line of sight can shield the other detector unit(s) from the neutrons. Analysis of the overall distribution of neutron capture events across the array (or the distribution of the number of neutrons absorbed by the scintillator element of each detector unit) therefore enables a picture of the energy of the external neutron source to be determined. It is has been recognised that by‘segmenting’ the detector in this way, and considering neutron capture events within each individual detector unit (or segment) of the detector, a low cost and easy to use method for determining the direction and energy spectrum of incoming neutrons is provided. Preferably, the step of interpreting the determined distribution comprises using a machine learned model to interpret the determined distribution wherein the machine learned model. The machine learned model may be effective to at least partially compensate the determined distribution for background neutron radiation. Machine learned models are particularly beneficial for such a purpose, as they can efficiently and effectively identify irregular patterns representative of background radiation in unseen data. This ability to recognise patterns of radiation means that the compensation can be applied in real-time, enabling the provision of a portable neutron detector.

The machine learned model can be trained by providing a learning algorithm with a plurality of data sets, each data set representative of a distribution of neutron capture events due to a neutron source and a corresponding, known, direction to said neutron source. In this way, the learning algorithm is provided with a known target (or direction) is known, and the distribution representative of said target. The learning algorithm uses the training data (the known distribution and corresponding direction) to improve recognition of patterns of background radiation in a self-taught manner, thereby enabling compensation for said background radiation. Machine learned models have recognised benefits in pattern recognition applications.

The use of machine learning, in combination with the availability of more compact and efficient detectors (such as that described herein with regard to the second aspect), facilitates the full use of the data recorded by the detector to provide a high level of information, without the need for complex analysis tools. These hardware and software advances thus facilitate the provision of a portable detector, which can image for neutron radiation quickly and without the need for expensive surveys. The machine learned model can be implemented as a neural network, or in any other suitable manner.

Optionally, the neutron detector comprises an n x m x p array of detector units, where p is at least 2. Preferably, n, m andp each equal at least 3. The use of a 3D detector array in the method of the first aspect can provide improved direction determining capabilities, since the use of an extra dimension in the detector provides for detection in an additional direction of incidence. Detection of neutron capture events is therefore less sensitive to the orientation of the detector relative to the neutron source (for a 2D array, if the external neutron source is perpendicular to the plane of the detector the distribution of neutron capture events across the array may not be sufficiently representative of the direction to the neutron source to allow for said direction to be determined). In particular, 2 units are needed in each dimension of the detector to facilitate analysis of neutron capture distribution without regard to detector orientation, preferably 3 detection units are provided in each dimension to improve the direction determination by adding another possible direction of incidence. Preferably, each detector unit is the same size and n = m = p. This cubic arrangement is beneficial because the symmetrical arrangement of detector units provides, on average for a sufficiently large number of neutron capture events, the same dependence of neutron capture events to number of detector units in each dimension of the detector. As such, corrections may not need to be applied to the collected data in order to determine the direction to the neutron source. More preferably, n, m and p each equal 4 - this arrangement provides a suitable trade-off between precision in the determination of the neutron source direction and the overall cost of the detector (in terms of both the expense of the hardware and the computational load / resources required for operation).

In the following description, a neutron detector for detecting a neutron source external to a neutron detector is also described. The neutron detector comprises at least one photon detector and an array of detector units, where each detector unit is optically isolated from each of the other detector units. By providing moderator material within each optically isolated detector unit, a greater differentiation between the locations of neutron capture events for neutrons of different energy levels can be seen, facilitating improved determination of the energy spectrum of incoming neutrons. Each detector unit comprises: a transparent neutron moderator material having at least a first face, a second face and a third face; a scintillation element at least partially covering the first face of the transparent neutron moderator material and arranged to absorb incident neutrons and output photons; and first and second optical fibres arranged to receive photons from the scintillator material and provide the photons to the at least one photon detector, the first optical fibre arranged within a groove in the second face of the transparent neutron moderator material and the second optical fibre arranged within a groove in the third face of the transparent neutron moderator unit, the groves arranged such that the first and second optical fibres cross. The output photons can be optical photons.

This arrangement facilitates the provision of an efficient and compact detector suitable for use with the method of the first aspect. Optionally, the first and second optical fibres are orthogonal to one another. This can reduce the complexity of the analysis required to localise neutron capture events, facilitating the provision of a simpler and quicker detector device. Brief description of the drawings

The following description is with reference to the following Figures:

Figure lA shows a schematic diagram of a two-dimensional (2D) array of detector units, illustrative of the array of the second aspect and suitable for use with the method of the first aspect;

Figure lB shows a schematic diagram of a three-dimensional (3D) array of detector units, illustrative of the array of the second aspect and suitable for use with the method of the first aspect;

Figure 1C shows a schematic of the structure of detector units of the arrays of Figures lA and lB;

Figures 2A and 2B illustrate a neutron capture process;

Figure 2C illustrates detection of the neutron capture event of Figures 2A and 2B; Figure 3A shows a schematic plan view of a detector comprising a 3x3 array of detector units arranged with respect to a neutron source;

Figure 3B illustrates neutron capture events across the array of Figure 3A;

Figures 4A and 4B illustrate an exemplary distribution of neutron capture events, in different dimensions of the array of Figure 3A, for a low energy neutron source;

Figure 4C shows another exemplary distribution of the neutron capture events; Figures 5A and 5B illustrate an exemplary distribution of neutron capture events, in different dimensions of the array of Figure 3A, for a high energy neutron source;

Figure 5C shows another exemplary distribution of the neutron capture events; Figure 6 illustrates a method of determining a direction to an external neutron source;

Figures 7A and 8A illustrate exemplary distributions of neutron capture events characteristic of background neutron radiation;

Figure 7B illustrates an exemplary detection of a signal from a low energy neutron source relative to the background radiation of Figure 7A; and

Figure 8B illustrates an exemplary detection of a signal from a high energy neutron source relative to the background radiation of Figure 8A.

Detailed description

With reference to Figures lA and lB, a detector unit too comprises an array of detector units 102. Figure lA illustrates a 2D detector array, that is, an array having more than one detector unit in a depth and a width dimension and a single detector unit in a height dimension. In Figure lA the detector too comprises four detector units 102a to I02d. Figure lB illustrates a 3D detector too having more than one detector unit in each of a depth, width and height dimension. In Figure lB the detector too comprises eight detector units 102a to i02h (i02h not shown).

Figures lA and lB illustrate a planar and cubic arrangement of detector units 102, but any other suitable geometric arrangement of detector units 102 could be provided, depending on the particular purpose. When used in the method of the first aspect, described below with reference to Figure 6, the detector too should comprise at least an m x n array of detector units, where m and n are at least 2, as shown in Figures lA and lB.

With reference to Figure 1C, each detector unit 102 of the detector too comprises a transparent neutron moderator material 104 and a scintillator element 106 at least partially covering a face, or side, of the moderator material 104. Preferably the scintillator element 106 covers the entirety of one face or side of the moderator material 104, optionally it covers more than one side or face.

Each detector unit 102 also comprises at least two optical fibres 108, 110. The optical fibres are arranged within respective grooves in the moderator material 104. The groves are arranged such that the first and second optical fibres cross, in order to facilitate localisation of the detector unit in which a neutron capture event occurs. By arranging the crossing optical fibres 108, 110 within grooves in the moderator material 104, the fibres can be arranged more easily within the detector units 102a, 102b and a more compact detector can be provided. The grooves can be located on different faces of the transparent moderator material 104 such that the optical fibres 108, 110 run across the faces, or sides, of the moderator material 104, as illustrated in Figure 1C. Alternatively, the grooves can be located such that one or both of the optical fibres 108, 110 run along edges of the transparent moderator material 104, the edges being defined as the location at which two sides or faces join (or meet). In some embodiments, a third optical fibre is provided in each detector unit 102. This can provide additional information for localising neutron capture events within the detector array, which is of particular relevance when the detector too is a 3D array of detector units 102. Preferably, optical fibres 108 and 110 are arranged to be orthogonal to one another; this can facilitate an easier determination of the detector unit 102 in which a neutron capture event occurs, of particular relevance when the detector is used in the method of the first aspect, described below with reference to Figure 6. In some arrangements, optical fibre 108 is shared between two or more detector units arranged along a first direction of the array, as shown in Figure 1C, and optical fibre 110 is shared between two or more detector units arranged along a second direction of the array.

Alternatively, each detector unit 102 can comprise at least two optical fibres which are separate and distinct from the optical fibres provided in the other detector units of the array. Incident neutrons from an external neutron source pass through the moderator material 104, which moderates the energy of the incident neutrons. This moderation process is essentially a reduction of the initial high kinetic energy of the incident neutron by transfer of energy to the moderator material. This process is also known as a neutron‘slowing down’ process, since the reduction in energy corresponds to a reduction in the speed of the neutron. The moderation process is designed to reduce the neutron energy to form a‘thermal neutron’, a free neutron with a kinetic energy of about 0.025 eV (speed of 2.2 km/s). The resulting thermal neutrons can then be absorbed or captured by a scintillator element. This process is described in more detail with reference to Figures 2A and 2B.

With reference to Figure 2A, a neutron 112 is incident to the detector unit 102. The neutron 112 can be from a neutron source to be detected, or can be background neutron radiation. As the neutron 112 passes through the moderator material 104, its energy is reduced by means of a number of elastic scattering collisions with nuclei of the moderator material 104. These scattering collisions are represented by the dotted path within the moderator material 104. During an elastic scattering (or neutron recoil) interaction, energy is transferred from the neutron to an atom of the moderator material 104 and nylon or photon may be omitted. After the scattering collision, a photon (y) may be released, as indicated in Figure 2A.

In the example described herein, the moderator material 104 is not formed of a scintillating material (i.e. is non-scintillating) and the photons released are not optical photons. The photons released during the neutron recoil or scattering events are therefore not detected via the optical fibres. In the present method, it is not necessary to detect these non-optical photons, since analysis of individual neutrons is not performed (in order to reduce analysis time and computational expense); the direction and type of a neutron source is instead determined from the distribution of neutron capture events alone. However, in some instances it may be useful to detect the photons generated in the moderator material, since these signals are indicative of the path of the neutron prior to the neutron capture event and can therefore provide additional information regarding the neutron source. For example, analysis of the path a neutron took prior to the neutron capture event can be useful where the neutron source is diffuse in nature. Therefore, in some arrangements the moderator material 104 is a suitable scintillator material. After sufficient collisions within the moderator material 104, neutrons arrive at an energy level of about 0.025 eV, thereby forming a thermal neutron 114. The resulting thermal neutron may then be absorbed or captured by a scintillator element. When the detector too comprises an array having more than one detector unit 102 in some, or all, of the dimensions, the successive layers of moderator material 104 act together to slow neutrons - neutrons with a higher kinetic energy will on average pass through more layers of moderator material 104 before being absorbed or captured by the scintillator element 106 than those with a lower kinetic energy. In this instance, the thermal neutron 114 is captured by the scintillation element 106 of the detector unit 102 in a neutron capture event 116, but depending on the energy of the incident neutron 112, the neutron may be expected to pass through one or more additional detector units before forming a thermal neutron 114 and being captured by a scintillation element 106.

With reference to Figure 2B, the neutron capture event 116 causes a cascade of optical photons (y) to be emitted from the scintillation element 106. These optical photons pass through the transparent moderator material 104 but, due to the optical isolation between individual detector units 102, are prevented from passing into another detector unit of the array. The photons are eventually captured by the optical fibres 108, 110 of the detector unit 102 and transported to the end of the optical fibres for detection by a photon detector.

The photon detector can be a silicon photomultiplier (SiPM) or any other suitable photon detector. The photon detector converts the optical signal transported by the optical fibres 108, 110 into an electrical pulse, which can then be recorded with dedicated electronics, such a controller. The controller can be part of the photon detector or a separate component electrically connected to the photon detector. With reference to Figure 2C, it is recognised that each optical photon that passes through the optical fibres 108, 110 can be detected by the photon detector, including any single optical photons (y) which may be generated (for example, by collisions of the incident neutron within a scintillating moderator material). However, these individual photon events are considered too short to be considered a neutron capture event 116 and therefore do not trigger an increase in a count of neutron capture events (although as discussed above, in some instances analysis of this individual photon events can be useful in providing additional information about the neutron source). In contrast to the individual photon events, the scintillation signal associated with a neutron capture event comprises a large number of optical photons, resulting in a bright signal with a characteristic time decay signature (see n of Figure 2C). Detection of this type of signal at the photon detector triggers an increase in the number of neutron capture events counted for the detector unit 102 in which the neutron capture event 116 occurred. It is the arrangement of the optical fibres 108, 110 which enables the location of the detector unit 102 in which the neutron capture event 116 occurred to be determined. As can be seen in Figure 1C, first optical fibre 108 is arranged along a first face of the moderator material 104 and second optical fibre 110 is arranged along a second face of the moderator material. The optical fibres 108, 110 are arranged to cross each other - since photons from the neutron capture event 116 will be captured by both optical fibres, the resulting photon detection of photons from each optical fibre can be used to pinpoint the detector unit 102 in which the neutron capture event 116 took place.

The moderator material 104 can comprise any suitable moderator material such as, for example, a plastic, a thermoplastic or a liquid. For example, a plastic such as polystyrene or polyvinyl toluene, PVT, a liquid such as water, or a thermoplastic such as poly(methyl methacrylate), PMMA, can be used.

The scintillator element 106 should be sensitive to thermal neutrons. For example, the scintillator element 106 can comprise a neutron absorber material and a scintillator, or scintillation material. The scintillator may comprise, for example, ⁶ LiF and ZnS, but any other suitable scintillation material can be used to form the scintillator element 106. The detector units 102 can be optically isolated from each other by means of reflective layers between each detector unit 102. For example, a reflective paint, such as a paint comprising Ti0 ₂ , maybe used on the faces of the detector unit 102, or the detector units maybe wrapped in, or shielded from one another by, a reflective material such as Tyvek™ (DuPont) or PTFE. As discussed, for the method of the first aspect, described with reference to Figure 6, no analysis of the particular characteristics of the signal shown in Figure 2C needs to be performed. For example, the energy of the individual photons detected not considered, nor are the energy or time characteristics of the photon pulse (scintillation signal) analysed. Instead, the photon pulse simply acts as a trigger indicating a neutron capture event 116 has occurred; it is the distribution of counts of such events across the array which forms the basis of the determination of the direction to the neutron source.

Such a distribution of counts of neutron capture events is now described with reference to Figures 3A and 3B. Figure 3A shows a schematic plan view of a detector 300.

Detector 300 comprises nine detector units 302a to 3021. In this example detector 300 is a square 2D array extending in a first direction, x, and a second direction, y.

However, detector 300 could also be a 3D cubic array, or any other suitable shape, whether symmetric or asymmetric. A neutron source 318 external to the detector 300 emits neutrons (n) 312. These neutrons 312 enter the detector 300 and are captured by scintillator elements of detector units 302a and 302b in neutron capture events 316a and 316b, as illustrated in Figure 3B. Other neutrons emitted by source 318 may be captured by detector units 302a to 3021 to form a distribution of neutron capture events.

Illustrative distributions of neutron capture events along the x direction of array 300 are shown in Figures 4A (low energy source) and 5A (high energy source). Similarly, illustrative distributions of neutron capture events along the y direction of array 300 are shown in Figures 4B (low energy source) and 5B (high energy source). Figures 4A, 4B, 5A and 5B have the same scale. For explanatory purposes, it is assumed that there is no background radiation.

For a low energy source, more neutrons are expected to be captured at the side of the detector 300 incident to the incoming neutrons 312 because many of the neutrons may not have sufficient energy to pass through the moderator material of multiple detector units 302 before being captured. In this way, the first detector units shield the detector units at the opposite side of the detector. However, for a high energy source, many of the neutrons are expected to pass through the first detector units and instead be captured in the middle detector units. In this way, it is the middle detector units which shield the detector units at the opposite side of the detector.

This shielding provides an indication of the energy of the neutron source, as can be seen from Figures 4C and 5C, and the overall aggregated distribution therefore provides information regarding both the energy of, and direction to, the neutron source. The overall distribution of the neutron capture events provides information as to possible line of sight to the neutron source. Moreover, because the number of neutron capture events 316 is expected to decrease as a function of distance from the neutron source 318 due to the shielding of the earlier detector units 302, analysis of the distribution can enable determination of the energy of the neutron source 318. Since the energy of the incoming neutrons affects the distribution, analysis of the distribution can facilitate identification of the type of neutron source; in other words, there is a mapping between the distribution of neutron capture events and the energy and direction of incoming neutrons. This analysis is described in more detail with reference to Figure 6. With reference to Figure 6, a method 600 of determining a direction to an external neutron source, such as neutron source 318, is provided. At step 610, a neutron detector is provided. Method 600 can be performed using detector too described above, where the detector comprises at least a 2 x 2 array of detector units 102.

Alternatively, method 600 can be performed using any neutron detector comprising an n x m array of detector units, where n and m are at least 2, each detector unit being optically isolated from the other detector units in the array and each detector unit comprising a scintillation element arranged to absorb a neutron and output photons in a neutron capture event. Preferably, the detector provided at step 610 is a 3D array comprising at least a 2 x 2 x 2 array of detector units. Such an arrangement provides the greatest flexibility in determining a direction to an external neutron source, since the three dimensions of the detector enable three directions of incidence of incoming neutrons to be detected, meaning that orientation of the detector with respect to the neutron source is immaterial. However, the method 600 can be performed using a 2 x 2 array, provided that the detector is orientated appropriately with respect to the neutron source; for example, the detector may be rotated and readings taken at each orientation to enable determination of the direction of the neutron source.

At step 620 of method 600, the number of neutron capture events occurring at each detector unit of the provided detector is measured, or counted. This measurement or counting can be performed in accordance with the approach described above with respect to Figures 2A to 2C. Alternatively, any other suitable method of counting a number of, or measuring a count of, neutron capture events in each detector unit can be used.

Once the number of neutron capture events at each detector unit has been measured, a distribution of the number of neutron capture events across the array of detector units is determined at step 630 by aggregating the total number of counts within each detector unit. For a 2D array, such a distribution can be represented as illustrated in Figures 4 or 5, for example. For a 3D array, the distribution of neutron capture events, or number of neutron capture events, can be represented as a heat map or as a matrix of data values, for example.

At step 640, a direction to the neutron source is determined based on the determined distribution of neutron capture events. In other words, the spatial distribution of neutron capture events is transformed into direction information indicating a direction to the neutron source. By logically breaking up the detection of neutron capture events into blocks and then aggregating the results, a quicker and easier method of direction determination is provided, which obviates many of the issues with known detector arrangements and the complexity of analysis of individual neutron trajectories. It is the spatial correlation of the neutron capture events to the direction of the neutron source which facilitates this treatment of the data.

Optionally, method 600 can comprise an additional step 650 of identifying a type of neutron source based on the determined distribution. Identifying a type of neutron source can comprise comparing the distribution of neutron capture events to distributions characteristic of different types of neutron sources. In particular, the distribution of neutron capture events can be considered to represent a‘fingerprint’ of the neutron source, indicating whether the source is shielded or un-shielded, and the type of source. Comparison of this‘fingerprint’ to known distributions of neutron capture events, for example through the use of a look up table or other suitable method of comparison, can allow for identification of the neutron source and thus for appropriate countermeasures to be taken. Security is thus improved.

Either or both of steps 640 and 650 can preferably comprise a step of applying a machine learned model to the determined distribution in order to compensate the distribution for background neutron radiation (step 660). When determining a direction to the neutron source based on the determined distribution (step 640) comprises the step of applying a machine learned model for compensating the determined distribution for background neutron radiation, the optional step of identifying a type of neutron source based on the determined distribution (step 650) preferably comprises comparing the compensated distribution to distributions characteristic of different types of neutron sources.

A machine learning process is particularly beneficial in this application as it facilitates automatic compensation for background neutron radiation due to the ability of machine learning algorithms (such as neural nets) to approximate non-linear distributions (like that of the number of neutron capture events across the detector) and determine a corresponding source; the background radiation is one feature of the input to the machine learning algorithm or process which can be recognised and compensated for during the determination of a source. In particular, machine learning processes, implemented for example on a neural network, can automatically distinguish between signals due to background patterns and signals resulting from a source. As such, the machine learned model can be effective to compensate for background signals and may be automatically adjusted over time to take into account changing

environmental factors due, for example, to a moving source and/ or moving detector. This is of particular importance for applications where a portable detector is desirable.

The machine learning process analyses data received from the detector for differences in the shape of the distribution of neutron capture events relative to the recognised patterns indicative of background distribution of radiation. This approach can reduce or minimise the risk of false alarms. Traditional approaches to neutron detection set an alarm level based on a multiple of a background neutron reading and an expected reading indicative of a neutron source. However, these traditional approaches take no account of variations in the background radiation due to environmental factors, or due to a change in detector location when a detector is used in a portable manner. By using machine learning to identify changes in the data representative of a changing background radiation, the method can provide increased sensitivity to neutron sources whilst simultaneously reducing the risk of false alarms.

An example of this compensation process is described with reference to Figures A and 7B (low energy source) and Figures 8A and 8B (high energy source). Figures 7A and 8A show examples of a distribution of neutron capture events characteristic of background neutron radiation. This background neutron radiation can be considered

representative of background radiation at a time , where there is no external neutron source. The number of neutron events in each detector unit is determined, the detector units being arrayed along a dimension of the detector, x. The number of detector units arranged in the x direction of the detector array and the number of neutron capture events is purely exemplary and is for illustration of the underlying principles only.

At a time t ₂ , later than time , the detector is brought within sufficient proximity to a neutron source 318 for detection of neutrons emitted by said source to occur (or a neutron source 318 is brought within proximity of the detector). Due to the presence of these additional neutrons, arriving from a single location, the number of neutron capture events in some of the detector units of the detector increases. As described above with respect to Figure 4, for a low energy neutron source the number of neutron capture events is expected to increase most in those detector units nearest the neutron source, since these detector units are expected to generally shield the aft detector units from the incoming radiation. This can be seen in Figure 7B, where grey indicates the count of neutron capture events characteristic of the background radiation (as in Figure 7A) and black indicates the increased count due to the proximity of a low energy neutron source.

As described above with respect to Figure 5, for a high energy neutron source the number of neutron capture events is not expected to increase significantly in the detector units in direct line of sight to the neutron source, since high energy neutrons will be more likely to pass through these detector units before being slowed sufficiently to be absorbed by the scintillation elements of the detector units. As such, a higher count of neutron capture events is expected in detector units further into the detector, with a lower count of events in those within direct line of sight. This can be seen in Figure 8B, where grey indicates the count of neutron capture events characteristic of the background radiation (as in Figure 8A) and black indicates the increased count due to the proximity of a high energy neutron source.

As can be seen from Figures 7B and 8B, it is not the absolute count of neutron capture events which is relevant to determining the direction to the external neutron source, but rather the aggregated difference in the number of neutron capture events between arrays and with respect to the background neutron radiation, preferably as recognised by a machine learned model. In other words, it is the asymmetry between neutron capture event counts in different detector units which is considered, rather than the absolute variation in the count. Machine learning processes are particularly beneficial for such a pattern recognition application, since they can facilitate the detection of a new signal on top of, or above, a background noise level, even when the background noise level is constantly varying. Moreover, the use of machine learning in method 600 has additional benefits as compared to traditional approaches of neutron detection. In particular, the detection efficiency of neutron capture events is much higher than the detection efficiency of neutron paths in approaches where individual neutron trajectory is considered event by event, such that fewer neutrons need to be incident upon the detector in order to generate a useful signal with the present method. Therefore, aggregating the neutron capture events distributed over the detector array in the manner claimed can provide a more efficient and more robust detection process, where the presence of a source can be determined more quickly than with conventional approaches. The machine learned model described herein can be trained or developed by providing a learning algorithm with training data sets. These training data sets preferably comprise a plurality of distributions of neutron capture events across a detector, and a plurality of directions to a neutron source, each corresponding to one of the

distributions. The learning algorithm can then iteratively develop a model which recognises patterns characteristic of background radiation, and patterns indicative of a particular neutron source direction, in order to provide the machine learned model.

As such, method 600 of the present invention adopts an approach in which neutron capture events are considered in less resolution than with previous approaches, but allows for the information to be processed much quicker. Not only does this approach make the method more robust, but it facilitates the periodic, rather than continuous, use of the method. In this way the energy use of the detector is minimised, which is of particular relevance for portable devices (which may be battery operated).

It is noted herein that while the above describes various examples, these descriptions should not be viewed in a limiting sense. Rather, there are several variations and modifications, which may be made without departing from the scope of the present invention as defined in the appended claims.