GB2608405A

GB2608405A - Improved detection of conductive material

Info

Publication number: GB2608405A
Application number: GB2109439.6A
Authority: GB
Inventors: John Bowring Nicholas
Original assignee: Security Screening Tech Ltd
Current assignee: Security Screening Tech Ltd
Priority date: 2021-06-30
Filing date: 2021-06-30
Publication date: 2023-01-04
Also published as: GB2609307A; GB202209503D0; GB202109439D0; WO2023275541A1

Abstract

An apparatus for detecting conductive objects, e.g. objects of metal, carbon, silicon, conductive plastic-based materials in a fluid medium e.g. air or water comprises a transmitter coil 20 for transmitting a magnetic field arranged with respect to a predetermined detection zone (e.g. within a fluid medium). The coil, or a second coil, acts as a receiver coil for detecting a magnetic field induced in the conductive object(s). A power supply applies intermittent voltage pulses to the transmitter coil. A means 12 (e.g. an active component) impedes the effect of a flyback pulse on the receiver coil at the end of each voltage pulse such that data acquisition can commence. The apparatus can distinguish conductive objects from other conductive objects classify conductive objects, and may involve training an artificial neural network. Concealed weapons e.g. knives may be detected, and also defects e.g. cracks in objects such as pipes and wires.

Description

Improved Detection of Conductive Material

Field of the Invention

The application relates to apparatus and methods for the detection of conductive objects, especially metal objects and objects of other conductive material e.g. carbon, silicon, 5 conductive plastic-based materials, and, in particular, conductive objects e.g. of metal or other conductive material having particular characteristics. The application also relates to apparatus and methods for distinguishing conductive objects from other conductive objects. The application also relates to apparatus for classifying conductive objects comprising a classification module, a method of manufacturing such a classification 10 module, a method of gathering data for such a classification module, and a method of training such a classification module, for example, method(s) and apparatus for training and/or using an artificial neural network.

Background

Known metal detection systems use very low frequency induction balance technology or 15 beat frequency oscillation technology or pulse induction technology. This application relates to pulse induction technology, but some of the apparatus, methods and techniques may be applicable to other detection systems.

US4868504 JOHNSON is directed to induction balance-type metal detectors with two distinct coils for metal detection in the ground. JOHNSON aims to recover power to the supply during the flyback pulse. A gating circuit is turned on after the end of the flyback pulse, with the intention to eliminate pure reactive and pure resistive components produced by magnetic and dissolved salt in the soil present, and to capture eddy currents in a metal object in later time slots. Thus, the timing gate opens after the flyback pulse ends. The characterisation of metal objects is not carried out nor contemplated.

U510181720 EMERY describes a dual polarity high voltage blocking circuit for the high voltage flyback pulse in a pulse induction metal detector. EMERY specifies a 'critically' damped coil, damped by a parallel resistor stating it provides 'the ability.. to sample the received signal earlier.. .due to the loading of the transmit coil' but gives no indication of purpose, effectiveness, or timescales. Furthermore, EMERY uses the flyback pulse to disconnect the receive circuitry from the transmit coil.

U55576624 CANDY describes a pulse induction metal detector for detecting metals in soils which uses combinations of signals detected during non-transmission of the magnetic field to eliminate certain contributions to the signal due to remnant magnetisation of ferrites etc. within a solid target volume. US2017299753 CANDY describes a metal detector for detecting a metallic target in soil using a predetermined synchronous demodulation function to assist in rendering it more insensitive to unwanted background contributions from the soil.

AU4752202 PALTOGLOU describes a method of distinguishing between ferromagnetic and non-conductive materials in the ground using best fit applied to decay curves (exponential or power law) or varying the pulse length and monitoring the decay. The voltage in the receive coil is measured at two or more times after the end of the transmit pulse and the characteristic time decay of the receive signal is determined, usually by computing a ratio between the signal strength at two different times to provide a method of nulling the ambient surrounding matrix of materials, usually ground containing ferromagnetic material. Whilst differences are noted between conductive and ferromagnetic materials, there is no attempt to characterise conductive items in the way suggested by the present invention.

US2009045813 EMERY appears to be an inductive balance arrangement which uses high voltage flyback pulses occurring at the termination of each of a continuous train of high current pulses. Signals are intercepted from metallic objects in the ground using a receive coil tuned to a resonant frequency with a parallel capacitor. The ratio of the resistive and the reactive components of the metallic object are displayed as an indication as to the nature of the metallic object in the field of the search coil. There is no attempt to characterise conductive items in the way suggested by the present invention.

US2006006872 NELSON describes a pulse induction metal detector for soils using a coil that can be used in transmit and receive modes (once the transmitter magnetic fields have decayed enough). NELSON proposes a variable resistor used across the transmit and/or receive coil can be used to increase or decrease the kick-back voltage and so optimise sensitivity. US2006006874 NELSON describes a two-configuration pulse induction metal detector for mine detection in soils that can be switched between differential (two receiver coil) and summing (one receiver coil) configuration.

W003/034095 KEENE aims to identify the precise location of metal within a body using pulsed induction (PI) or continuous wave (cw) metal detection methods in gradiometer arrangements. KEENE mentions the use of cumulative signal over several pulses. The characterisation of the metal objects is not carried out nor contemplated.

US5552705 KELLER describes a non-obtrusive weapon detection system to determine a time constant of decay to define a decay curve (optionally displayed on a log graph to find the slope) and to compare this with those in a database. Keller explains that 'shapes of decay curves under observation [show]... the different metal objects have distinct decay transients that are measurable and wherein the time constant of each object is a direct measure of the electromagnetic scattering cross section of object's profile'.

KELLER explains 'Accurate determination of a time constant will be accomplished by post-processing after a cycle of excitation has been completed and an averaged signal resides within digital memory of the digital system One easy approach to determining time constant is by fitting the transformed data, that is, a signal in which the signal amplitude is converted to its logarithm, with a best-fit one-parameter linear correlation function. The one parameter in the fit is the slope, which is the parameter to be used in identifying the nature of the scatteree.

This is one of few documents that mentions metal detection about a person rather than in the ground. KELLER acknowledges different objects have distinct decay transients and uses a database of previously identified time constants to identify an object. TVV340908B HWANG is a 1996 document which appears to describe the same non-obtrusive weapon detection system as KELLER Lammed BIES describes an easy to build pulse induction metal detector with digital signal processing (DSP) for use in soil. https://wwvv.lammertbies.nl/electronics/pi-metal-detector (accessed May 2021). BIES describes different decay curves between ferromagnetic and non-ferromagnetic targets, and differences between objects of different materials. BIES indicates that a metal object in the field will decay at different rates and also notes, As with all metal detectors it is an educated guess and not a definite answer, because size, depth, surrounding targets and soil response may alter the signal in such a way that proper discrimination is not possible.' The present invention seeks to alleviate one or more problems remaining in the art. It remains difficult to distinguish different types of conductive objects, especially when these are carried about a person. Further, it is difficult to detect different characteristics of conductive objects when carried about a person. This is especially the case when two or more conductive objects such as a mobile phone, keys, or a knife may be present.

Statements of the Invention

In a first aspect the invention provides apparatus for detecting conductive (e.g. metal) objects (including for example detecting conductive parts of objects, such as defects in conductive objects, sharp edges, thin blades, corners, thin plates, narrow discontinuities etc.) in a fluid medium (such as air or liquid e.g. water) comprising: -a first coil as a transmitter coil (Tx) for transmitting a magnetic field, the transmitter coil (Tx) arranged with respect to a predetermined detection zone (e.g. within a fluid medium); -the first coil, or a second coil, as a receiver coil (Rx) for detecting a magnetic field induced in one or more conductive object(s) within the predetermined detection zone; -a power supply for applying intermittent voltage pulses to the transmitter coil (Tx); -a means (e.g. an active component) for impeding the effect of a flyback pulse (e.g. in the transmitter (Tx) coil) on the receiver (Rx) coil at the end of a (e.g. the or each or some or all) voltage pulse(s) such that data acquisition can commence (e.g. can commence earlier, for example earlier than if such a means were not provided).

Preferably, the means for impeding the effect of the flyback pulse is configured such that the recovery time R after a voltage pulse ends e.g. before useful data acquisition commences, is less than an eddy current decay time in at least one target object of interest.

Preferably, a recovery time R after the voltage pulse is the duration of the flyback pulse e.g. until it diminishes to a negligible amount, and/or in which a recovery time R after the voltage pulse is a characteristic decay time (e.g. a decay time constant) of the flyback pulse and/or the time for the curve to fall by a factor from a maximum e.g. 2 or 3 or 4 or 10 or 100 or 1000 etc. times less than the maximum, and/or other characteristic time measurement or to a pre-set limit e.g. -10mV, 5mV, 2mV, 1mV.

The eddy current decay time of an object may be a characteristic decay time as measured on a decay curve typically after background subtraction e.g. an exponential fitted to the data, and/or may be the time for the decay to fall to a negligible amount, and/or the time for the curve to fall by a factor of the maximum e.g. 2 or 3 or 4 or 10 times less than the maximum, and/or other characteristic time measurement. In practice, the difference in the measured decay curves at the receiver coil when a conductive object with a small dimension characteristic -such as a sharp edge or thin blade or narrow discontinuity -is present and when an object is not present is very slight, but the decay curve of such an object can be seen using the methods and apparatus of the invention and in particular the digital signal processing techniques of the invention.

Preferably, the means for impeding the effect of a flyback pulse comprises a damping circuit (or damping component) associated with (e.g. parallel to) the transmitter coil (Tx), the damping circuit having at least one transient suppression component (e.g. an active transient suppression component).

Preferably, the means for impeding the effect of a flyback pulse may comprise at least one transient suppression component associated with (e.g. parallel to) the transmitter coil (Tx), The at least one transient suppression component may be external to the power supply or may be internal to the power supply. Optionally, both internal and external components may provide damping such as a discharge capability. As but one example one or more external critical damping resistors and one or more internal and/or external active transient suppression components (e.g. Zener diodes) may be provided.

The at least one transient suppression component may comprise at least one Zener diode, or a chain of two or more Zener diodes in series, e.g. the Zener diode or the chain of Zener diodes in parallel with a transmitter coil (Tx).

Preferably the at least one transient suppression component (e.g. the type and/or value and/or configuration of the at least one transient suppression component) is selected such that the recovery time R after the end of the volage pulse (e.g. which may be equivalent to a representative decay time of the flyback pulse (e.g. its decay time or duration or similar measure as explained elsewhere) is less than the expected time (e.g. its decay time or duration or similar measure as explained elsewhere) for eddy currents to decay in one or more target conductive objects of interest. In one or more example embodiments, the decay time constant of the flyback pulse e.g. which may be equivalent to the decay time constant of a damping circuit) is less than the decay time constant of eddy currents in one or more target objects (or parts of objects) of interest.

Preferably, the at least one transient suppression component comprises at least one Zener diode in reverse bias configuration (e.g. to an expected flyback pulse), the at least one Zener diode in parallel with the transmitter coil (Tx).

Preferably, the at least one transient suppression component comprises a series of two or 35 more Zener diodes in parallel with a transmitter coil (Tx). A reversed biased diode may also be provided which is forward biased when flyback occurs.

Preferably, the damping circuit comprises one or more damping resistors in parallel with the transmitter coil (Tx) e.g. one of which may be an adjustable damping resistor such as a trim potentiometer. Preferably, the trim potentiometer resistor should be of typically values between 1k and 10k (depending on the inductance of the transmit coil -with larger 5 values required for higher inductances) resistance to facilitate optimal/critical damping. The damping resistor may be of a reasonably high value, as the higher the value, the more quickly the power is dissipated. Nevertheless, it should be born in mind that if the value is set too high, 'overshoot' may occur. The resistor is therefore preferably adjusted to 'critically damp' the flyback (i.e. to reach steady state in the smallest time). For inductances (say) 10 of 800 pHenry, a value of approximately 1 kilo-ohm (k0) may be optimal.

Preferably, the means for impeding the flyback pulse is configured is configured to damp (e.g. promote discharge of) the flyback pulse in a time less than the time constant of a target object (where a target object is referred to this is to be understood to refer to both a target object and/or to a part of an object being targeted, e.g. a knife, or a sharp edge, or thin blade or narrow discontinuity in material structure on or in an object).

Preferably, the means for impeding the flyback pulse is configured to damp (e.g. promote discharge of) the flyback pulse in a time period selected from: 50ps, 45ps, 40ps, 35ps, 30ps, 25ps, 20ps, <3ps 5ps (ps = microsecond).

Preferably, the applied pulse has a predetermined limited voltage e.g. to facilitate a lower voltage in the flyback pulse and faster switch off. For example, 12V, 24V, C6V or 48V pulses might be used. Providing pulses of lower voltages can assist with rapid switch off, but this may be balanced against the requirement to extend the reach of the transmission coil signal, and/or with other arrangements of the system, for example a separate receiver coil is provided may allow higher voltage pulses to be used (which will develop higher currents and so magnetic fields. In certain arrangements a higher current may be appropriate to facilitate a higher magnetic field and this may be provided by a higher voltage pulse.

Preferably, a pick-off circuit is provided for measuring a signal from the receiver coil (e.g. at the end of a pulse, typically at the end of a recovery time R of a flyback pulse F, which may be before the flyback pulse has completely disappeared but it is sufficiently low that magnetic fields from decaying eddy currents in target objects of interest can be revealed using one or more of the technique(s), apparatus and method(s) of the invention.) The power supply may comprise a power MOSFET, and may optionally comprise an optoisolator to provide a voltage pulse to the transmitter coil (Tx).

Preferably, the transmitter coil (Tx) has a predetermined (e.g. preselected) limited inductance e.g. to facilitate lower voltage in the flyback pulse and faster switch off (for example, for a 20-turn coil of 33 cm diameter, the inductance is 270 micro-Henries. If the number of turns is doubled, the inductance goes up to 1 milli-henry 0.e. quadruples. Larger inductances can cause significant lengthening of the dead (recovery) time R. Preferably, the apparatus comprises a control unit (e.g. a microcontroller) for controlling one or more of: timing of delivery of voltage pulses, pulse spacing, pulse length, pulse mark-space ratio, pulse voltage, and timing of measurement of the response.

It will, however, be noted that the response may be measured over the whole time period using the pick-off circuit but only particular segments of the measured response may be used. This is seen more clearly in Figure 9 in which a response measured in a received coil (Rx) is seen. Here, a first time period P is that in which a pulse is transmitted from the transmitter coil (Tx) (e.g. into a fluid medium) and a second time period commences after the end of the first time period and comprises a recovery time R during which data measurements are not measured or may be measured but are ignored (e.g. by subtracting a comparable background response, and/or using a gate timing circuit) because of the significant flyback pulse response F picked up by the receiver coil (Rx), and a data acquisition time period D which commences after the end of the recovery time period R. Preferably, the apparatus comprises a control unit for controlling transmission of a pulse from the transmission coil (Tx) into a fluid medium during a first time period Ti; and also for controlling measurement of the induced signal in the receiver coil (Rx) during a second time period T2 after the end of the first time period and, optionally, for removing or ignoring a third time period T3 (e.g. the recovery time period) of the induced signal in the receiver coil (Rx) forming a first part of a second time period (T2), such that data is collected in respect of a data acquisition time period after the end of the third time period T3 (e.g. the recovery period R).

Preferably, the pulses are one or more of: square pulses, equi-spaced, of the same length, 35 of varied lengths, of constant length during a particular data acquisition sequence (e.g. they may be varied from one data acquisition sequence to the next).

Preferably, the receiver coil (Rx) is separate from the transmitter coil (Tx). Optionally, the receiver coil (Rx) is of a different size and/or shape and/or number of coils from the transmission coil (Tx) to assist in deconvoluting the response of the receiver coil (Rx) from the flyback pulse existing in the transmitter coil (Tx) at pulse switch off. Optionally, the receiver (Rx) and transmission coils (Tx) are generally or substantially planar. Optionally, the receiver (Rx) and transmission coils (Tx) lie generally or substantially in the same plane (or planes e.g. when these are not entirely planer, each coil having a small depth). Optionally, the transmission coil (Tx) and receiver coil (Rx) are overlapping. Optionally, the transmission coil (Tx) and receiver coil (Rx) are concentric.

Preferably, the receiver coil (Rx) is smaller than the transmitter coil (Tx). This has been found to produce a faster response. The advantages of a separate receiver are as follows -more turns are possible on the receiver as there is no flyback current being dissipated in the coil -this increases sensitivity. In other words, this facilitates an arrangement with fewer turns on the transmitter and more turns on the receiver, giving a shorter dead/flyback time and more sensitivity overall. As the distances between the transmit and receive coil are increased, the flyback dead time (recovery time R) is reduced and hence the sensitivity to smaller objects (or objects with characteristics of smaller dimensions (thinner blade, shaper corner, less metal) (which have a shorter-lived response) is improved.

Preferably, the receiver coil (Rx) is of a different size and/or shape and/or number of coils from the transmission coil (Tx).

Preferably, the means for impeding comprises or further comprises a compensation coil (Cx) for creating a magnetic-free region within the transmitter coil (Tx), e.g. between the transmitter coil (Tx) and the receiver coil (Rx). Again, the benefits of a field free region is that the receiver coil can be used to commence data gathering immediately after the transmit pulse ends, improving sensitivity and detection distance, since there is no charge stored in the receiver coil as it sits within a field free region it does not need to discharge before (useful) measurements can commence.

Preferably, the transmitter coil (Tx), receiver coil (Rx) and the compensation coil (Cx) (where provided) are generally or substantially co-planar (e.g. all lying generally or substantially within the same plane) and the compensation coil (Cx) lies within the 35 transmitter coil (Tx) and the receiver coil (Rx) lies within the compensation coil (Cx).

It will be apparent to those skilled in the art that by placing such apparatus within zones through which humans pass, concealed weapons with sharp edges or thin bladed objects or indeed even mobile phones carried about a human person can be detected. Mobile phones are generally larger overall so have a larger smallest (conductive) dimension with metallic content than say the smallest (conductive) dimension of a knife e.g. its tip.

In a further aspect the invention provides a method for detecting conductive objects (including for example detecting conductive parts of objects, such as defects in conductive objects, sharp edges, thin blades, corners, thin plates, narrow discontinuities etc.) (e.g. in a fluid medium such as air or liquid e.g. water), in a pulse induction detection apparatus such as that in claims 1 to 19 or, in an apparatus comprising at least a first coil, the first coil acting as a transmitter coil (Tx), and the first, or a second coil, acting as a receiver coil (Rx), the method comprising: applying a voltage pulse to the transmitter coil (Tx) for transmitting a magnetic pulse from the transmitter coil (Tx) (e.g. into a fluid medium, e.g. during a first time period Ti) into a predetermined detection zone (e.g. at a predetermined distance or within a predetermined range of distances from the transmitter coil (Tx) and/or from the receiver coil Rx); impeding the effect of a flyback pulse at the end of a (the or each or some or all) voltage pulse (e.g. in the transmission coil (Tx)) on the receiver coil (Rx) whereby data acquisition can commence (e.g. can commence earlier than would otherwise be the case).

Preferably, the step of impeding comprises one or more of: -impeding the effect of the flyback pulse such that a or the recovery time R after a voltage pulse ends is less than an eddy current decay time in at least one target object of interest using at least one active component e.g at least one transient suppression component; and, -using a damping circuit having at least one active component e.g. at least one transient suppression component e.g. one or more Zener diodes, e.g. in parallel with the transmitter coil (Tx)and, optionally, one or more damping resistors in parallel with the transmitter coil (Tx); Using a receiver coil (generally separated from the transmit coil) with more turns on the receive coil and fewer turns on the transmit coil Preferably, the step of impeding comprises damping (e.g. discharging) a flyback pulse such that the duration of the flyback pulse is less than the expected time for the eddy current to decay in one or more target conductive objects (or parts of objects) of interest.

The switch off time may be the time from the end of application of a voltage pulse to the time the pulse has disappeared e.g. decayed to a negligible amount from the transmitter coil (Tx) (e.g. as seen in the response coil where two coils are provided), or it may be a characteristic decay time associated with the end portion of the pulse e.g. a characteristic decay time constant, or a combination of both, or other similar time parameter.

Preferably, the method comprises, in a predetermined detection zone, transmitting a magnetic field from the transmitter coil (Tx) and determining a background response (B(t)) on a receiver coil (Rx) when no target is present in a detection zone; (e.g. a background response situated within a fluid medium such as air or water or other liquid which, unlike the ground, can be thought of as homogenous within the context of the present proposed measurements), introducing a target into the detection zone; transmitting a magnetic field from the transmitter coil (Tx) and detecting a response on the receiver coil (Rx) when a target is present in the detection zone; and, subtracting the background response (B(t)) from the detected response (D(t)).

Introducing a target into the detection zone e.g. in air typically involves introducing one or more objects carried by persons into the detection zone. Whereas in water this may involve redirecting a detection zone associated with the apparatus from a background location adjacent to a target object of interest to the target object of interest. The background zone adjacent to the target object of interest (e.g. an underwater pipe, can be considered homogenous in the context of the invention when no object is present, so a background zone may be to one side of an underwater pipe e.g. remote from the pipe in the context of this apparatus -so 2 to 5m away, to a detection zone containing the pipe).

Preferably, the method comprises determining an absolute amplitude of the response by subtracting a background response (B(t)) from a detected response (D(t)).

Preferably, the method comprises repeating the step of determining a background response (B(t)) and determining a summation of background responses (sum B(t)) or average background response (Av B(t)).

Preferably, the method comprises repeating the step of determining a detected response (D(t)) when a target is present and determining a summation of detected responses (sum D(t)) or an average detected response (Av D(t)) when a target is present.

Preferably, the method comprises subtracting a background response (B(t)), or an average or summed background response (Av B(t), sum B(t)), from a detection response (D(t)) when a target is present, or from an average or summed detection response when a target is present (sum D(t), Av (D(t)), to determine a background subtracted detected response (A(t)) e.g. an absolute amplitude of the detected response.

Preferably, the method comprises normalising the amplitude of a background subtracted detected response (norm A(t)) (e.g. to 1 and/or to the maximum amplitude of a response).

Preferably, the method comprises fitting one or more curves (e.g. exponentials or other first order, second order, or further order decay curves) to one or more of: a detected response 20 (D(t)), a background subtracted detection response (A(t)), a normalised background subtraction detection response (norm A(t)).

Preferably, the method comprises extracting one or more decay constants from the fit and using such to classify (e.g. distinguish or categorise or in sone circumstances identify) a 25 target object, and/or part of a target object and/or a characteristic of a target object (e.g. threat/ not threat, sharp knife, sharp point, or sharp discontinuity or narrow defect or crack).

One or more methods of the invention may comprise one or more of the following steps: controlling (e.g minimising or optimising for the circumstances) the voltage of the applied pulse - controlling (e.g. minimising or optimising for the circumstances) number of turns on the transmission coil (Tx); controlling (e.g. maximising or optimising for the circumstances) number of turns on the receiver coil (Rx); -controlling (e.g. minimising or optimising for the circumstances) diameter of the transmission wire; controlling diameter of the wire on the receiver coil (Rx); controlling (e.g. maximising or optimising for the circumstances) the number of repetitions of the detection response; - controlling (e.g. maximising or optimising for the circumstances) number of repetitions of the background response; -controlling (e.g. maximising or optimising for the circumstances) separation of the receiver coil (Rx) and transmission coils (Tx); -providing a compensation coil (Cx); - controlling (e.g. matching or optimising for the circumstances with relation to the transmitter coil) number of turns of the compensation coil (Cx); -controlling (e.g. matching or optimising for the circumstances with relation to the transmitter coil) the diameter of wire of the compensation coil (Cx); -controlling the distance to the target (average, medium, maximum, minimum) from at least one coil (Tx, Rx, Cx); - controlling the distance to the target from, for example, a centre of the transmission coil (Tx) or other fixed point on the transmission coil (Tx) coil; controlling the distance from compensation coil (Cx) to receiver coil (Rx) and/or transmission coil (Tx).

In a further aspect the invention provides a method of preparing training data for a neural 20 network model comprising using the apparatus as described herein and/or using the method(s) as described herein to gather training data, the training data comprising: a) one or more input components, the input components comprising one or more of a detection response (D(t)); a background subtracted (e.g. absolute) detection response (A(t)); a normalised, background subtracted (e.g. absolute) detection response (norm A(t)); 25 and b) output components, comprising one or more characteristic of objects being detected e.g. one or more of: nature of the object; number of objects, a mobile phone; several mobile phones; one key; several keys; a knife; a plurality of knives; characteristic information about the conductive object e.g. its sharpness and/or its depth and/or its width and/or its size; and c) optionally, using one or more output components selected from: threat; no threat; sharp; not sharp; defect; no defect; thin; not thin: blade; not blade, edge; not edge; knife; not knife; discontinuity; no discontinuity.

Preferably, in the method of preparing training data, one or more input components comprise one or more characteristics of the measuring apparatus and/or measuring arrangements.

Preferably, in the method, the characteristics of the apparatus and/or measuring arrangements comprises one or more of: voltage of the applied pulse(s); -number of turns on the transmission coil (Tx); number of turns on the receiver coil (Rx); one or more dimensions (e.g. width, diameter, length) and/or shape of at least one coil (Tx, Rx, Cx); diameter of the wire of the transmission coil; -diameter of the wire on the receiver coil (Rx); number of repetitions of the detection response;

number of repetitions of the background response;

- separation of the receiver coil (Rx) and transmission coil (Tx); - presence of a compensation coil (Cx); number of turns of the compensation coil (Cx); diameter of wire of the compensation coil (Cx); - distance to the target (average, medium, maximum, minimum) from at least one coil (Tx, Rx, Cx); distance to the target from, for example, a centre of the transmission coil (Tx) or other fixed point thereof; - distance from compensation coil (Cx) to receiver coil (Rx) and/or transmission coil (Tx).

Other such similar characteristics that can influence the magnitude and/or timing of the 25 response can be envisaged from the disclosure.

In a further aspect the invention provides a computer-implemented method of training a neural network comprising using the apparatus as described herein and/or the method as described herein and/or training data prepared as described herein, the method 30 comprising: providing data to a neural network comprising one or more of: a detection response (D(t)), an absolute detection response (A(t)); a normalised, absolute detection response (norm A(t)); and/or characteristics of the object being detected e.g. sharp / not sharp, threat / no threat.

In a further aspect the invention provides an apparatus comprising a neural network trained in classifying target objects using the apparatus as described herein, and/or using the method(s) as described herein, and/or using training data from the method(s) as described herein, and/or using the computer-implemented method as described herein, optionally wherein the neural network was trained using an absolute detection response (A(t)) or a normalised absolute detection response (norm A(t)) for classifying objects.

In a further aspect the invention provides a method of classifying conductive objects comprising as described herein e.g. using the apparatus of any of claims 1 to 19 or 35, 37 10 or 38 and/or using the method of any of claims 20 to 30 and/or training data prepared using claims 31 to 33, and/or the computer implemented method of claim 34 to classify objects.

In a further aspect the invention provides an apparatus as described herein concealed in a poster, floor tile, or any item of street furniture.

In a further aspect the invention provides an item of clothing comprising an apparatus as described herein.

In a further aspect of the invention, there is provided a method of determining if one or more objects are present comprising: fitting a first order exponential function to decay data, 20 and determining a bad fit to the data and/or fitting a second order exponential function to decay data and determining a good fit to the data.

In one aspect of the invention, the invention may be thought of as, in a pulse induction eddy current detection apparatus or method, controlling the recovery time of the flyback pulse to be less than the decay time of the eddy currents in one or more particular objects of interest 25 (e.g. in a knife or other sharp or thin bladed objects or features).

In another aspect, the invention may be thought of as determining the amplitude of the eddy current response from an object as well as determining the decay time of the eddy current response from an object.

In one or more embodiments, determining the amplitude of the response comprises 30 subtracting a background response determined in the absence of an object (preferably contemporaneously, so not too long before or after, or just to one side of a fixed object e.g. a pipe, which is possible in a fluid medium such as air or water).

In one or more embodiments, the background is determined contemporaneously e.g. within a time frame of up to 10 minutes, or up to 5 minutes, but preferably within up to 1 to 2 minutes.

Any feature of any embodiment in any aspect of the invention may be used in any other 5 embodiment in any other aspect of the invention as understood from reading this disclosure.

Brief Description of the Invention

The present invention will now be described, by way of example only, with reference to the following figures in which like reference numerals refer to like features.

Figure 1 is a schematic circuit diagram for a pulse inductance (PI) detection apparatus according to the invention.

Figures 2 to 5 are schematic drawings of example coil geometries.

Figure 2 illustrates a square transmitter coil (Tx) 80 x 80cm, 15 turns, 1mm diameter wire coil and a separate, non-overlapping, inwardly located, circular receiver coil (Rx) of 22cm 15 diameter, 100 turns, 0.5mm diameter wire.

Figure 3 illustrates a square transmitter (Tx) coil of 80 x 80cm, 15 turns, 1mm diameter wire and a separate, overlapping, rectangular receiving coil (Rx) of 100 x 50cm, 50 turns, and 0.5mm diameter wire. The geometry may be designed to maximise the separation between the transmit and receive coils (within certain limitations), so as to try to minimise the dead time (recovery time R), such that sensitivity is improved.

Figure 4 illustrates a single coil transmitter/receiver (single coil Tx/Rx) square coil embodiment of 80 x 80cm, 15 turns, 1mm diameter wire (no separate receiver coil).

Figure 5 illustrates an outer transmitter coil of 80 x 80cm, 15 turns, 1mm diameter wire, a reverse-wound intermediate compensation coil (Cx) of 60 x 60cm, 7 turns, 1mm diameter wire, wound in the opposite direction (clockwise or anticlockwise) and an inner, circular receiver coil (Rx) of 22cm diameter, 100 turns, 0.5mm diameter wire. The compensation coil (Cx) creates a field-free region within its inner perimeter for the receiver coil (Rx). The outer coil is perhaps 0.5-1 cm deep, although the depth is not particularly important given the separation between the outer and compensation coil. The number of turns on the compensation coil may be easily calculated as a fraction of the two diameters. The field free region may be optimised empirically, by making small adjustments to the compensation coil until the transmit field disappears at the receiver).

Figure 6 is a table showing detection distances determined by experiment (as examples of the nominal extent of the detection zone) for various sharp objects for various coil configurations, when body worn or when used in a poster configuration (as but one example, using the coils of Figure 2 with the receiver coil (Rx) within the transmitter (Tx) coil or indeed located outside the perimeter of the transmitter (Tx) coil at a distance to one side of it but in the same plane.

Figure 7 is a typical waveform (intensity on the y-axis, channel number on the x-axis) 10 acquired from a single coil pulse induction (PI) system via a receiver circuit after a single sweep (1024 points, 16-bit resolution).

Figure 8 is a typical response (intensity on y-axis, channel number on x-axis) from a single coil PI system receiver circuit after 200 data sets (sweeps) (e.g. a sweep such as that seen in Figure 2) have been collected and averaged showing much improved signal to noise.

Figure 9A is a scan from a PI apparatus of the invention (y-axis in millivolts (mV)), taken using a Vx10 probe (x-axis in microseconds, 50 microseconds per division) showing the pulse and data acquisition sequence.

Figure 9B is a table showing example coil configurations and experimentally determined recovery times R. Figure 9B shows data taken using a 2-coil (Tx, Rx) arrangement illustrated in Figure 3. The data was taken as follows: With 26 turns on both transmit and receive coils, a 10V pulse of duration 120 microseconds gives the fastest response of 15 microsecond (ps) dead time. If the pulse is extended to 200 microseconds, this lengthens the dead time to 16 microseconds. If the transmit voltage is doubled to 20V, this extends the dead time to 18 microseconds. If the turns on the receiver is approximately doubled to 46 then the dead time is 16.9 microseconds. If the pulse is then increased to 30V with 46 turns on the receiver coil, the dead time is 22 microseconds.

Figure 10 is an example of a scan (amplitude in arbitrary units versus time in ps) with a knife in the detection zone from a P1 system of the invention showing scans with flyback pulse recovery times R of 20ps (42) and 30ps (44). The improvement in the signal intensity is evident with the faster recovery time. The waveform is displayed after background removal.

Figure 11 shows a typical response (intensity on y-axis, channel number on x-axis), here single sweeps from a single coil PI apparatus of the invention, one of a background sweep with no objects present, one of a detection sweep after background removal with a mobile phone present around 0.75m away.

Figure 12 shows overlaid normalised responses (normalised intensity on y-axis, channel number on x-axis) here single sweeps from a single coil PI apparatus (here a single, circular coil of 16 turns of 32cm diameter) showing overlaid responses for ferrous (knife) and nonferrous (drinks can) items, each individually (on separate occasions) present within the detection zone. A background sweep has been subtracted from each of the detection sweeps.

Figure 13 shows two output waveforms, one for a knife and one for a mobile phone. Each output waveform is the result of an average of 200 sweeps of a detection scan, from which an average of 200 sweeps of a background have been subtracted, and the outputs normalised to +1.

Figure 14 shows i) an output scan when only a knife is present in the detection zone; ii) an output scan when only a mobile phone is present; and iii) an output scan when both a knife and a mobile phone are present in the detection zone. A second order function consisting of two exponentials is necessary to fit the data. If a single exponential is used, the fit shows larger errors (which the inventor has appreciated indicates that two objects rather than one

might be present in the field.

Figure 15 is an example of a feed-forward neural network of the sort that can be trained to classify the objects, e.g. in the detection zone.

Figure 16 is the threat output of 550 data sets (1 to 550) used in training a feed-forward backpropagation neural network. These results were used as part of the training to provide supervision, to train the feed-forward backpropagation neural network. Generally the resultant network is validated both on the original data set and a second unseen data set. It is possible to 'overtrain' networks such that the network classifies the original data very well but does a poor job of classifying new data.

Figure 17A shows a method of detecting responses according to the invention e.g. 30 responses indicative of sharp, conductive objects. Figure 17B shows a method of classifying objects e.g. using fitting or a trained ANN.

Figure 18A shows a method for gathering data and a method for training an artificial neural network (ANN) using the gathered data (which may include input and/or output components). Figure 18B shows a method of using an ANN.

Figures 17A and 17B, and 18A and 18B include various steps used in the one or more of 5 the apparatus and method(s) of the invention.

Figure 19 shows typical responses for background 90, a knife 92 and a mobile phone response 94 after processing by a microprocessor.

Figure 20 shows two analogue responses, one for a background when no object is present and one when as knife is present over two pulse cycles before background subtraction, 10 and before signal averaging.

Detailed Description of the Invention

It will be understood by those skilled in the art that any dimensions, and relative orientations such as lower and higher, above and below, and any directions, such as vertical, horizontal, upper, lower, axial, radial, longitudinal, tangential, etc., and any components and component values, and any shapes referred to in this application are within expected structural tolerances and limits for the technical field and the apparatus and methods described, and these should be interpreted with this in mind.

This application describes pulse induction technology for metal detection in a fluid medium and, in particular, for detection of metal with particular characteristics such as sharp edges or points. The apparatus and methods of the invention may have wider application e.g. to the detection and identification of flaws, cracks, or defects in metal such as pipes, parts for engine, parts for vehicles, parts for buildings and so on.

Conductive objects, or parts of the conductive objects (hereafter referred to as conductive objects), have physical characteristics associated with very short-lived eddy currents.

Typically, at the termination of application of a voltage pulse to a transmitter coil (Tx), a very large voltage transient, known as a flyback pulse, typically of a very high voltage e.g. 10 to 20 or even 100 times or even more of the pulse itself, occurs. This flyback decays rapidly but can, nevertheless, still obscure very short-lived eddy currents. The inventor has appreciated that particular types of conductive objects of interest such as sharp edges, sharp points and sharp discontinuities such as cracks, defects etc. are particularly affected by the longevity of this flyback pulse but can be distinguished using the apparatus and methods of the invention.

It will be understood by those skilled in the art reading this disclosure that the voltage pulse 5 applied to the transmitter (Tx) coil can be seen as a response in the receiver (Rx) coil. Where two separate transmission (Tx) and receiver (Rx) coils are provided, the time for switch off of the transmission (Tx) coil can be seen in the response received in the receiver (Rx) coil (see for example Figure 9). Where a single coil as Tx/Rx is provided, this means that the single coil cannot be used to gather a received signal until the current developed 10 due to the applied voltage pulse has sufficiently dissipated.

The switch off time may be the time from the end of application of a voltage pulse to the time the pulse has disappeared e.g. decayed to a negligible amount from the transmitter coil (Tx) (e.g. as seen in the response coil where two coils are provided), or it may be a characteristic decay time associated with the end portion of the pulse e.g. a characteristic decay time constant, or a combination of both or other similar time related measure.

It has been appreciated by the inventor and it will be understood from this disclosure that impeding the effect of the flyback pulse in the transmitter coil Tx on the receiver coil Rx such that very fast decaying eddy currents can be seen more swiftly in the Rx coil, is important, and that this can be achieved in one or more various ways. In one or more embodiments this is achieved by adding one or more active transient suppression devices, such as one or a chain of Zener diodes, e.g. to a damping circuit, preferably across the Tx coil.

Careful selection and control of the inductance of the transmitter coil (Tx) may also facilitate faster suppression of the flyback pulse and so a faster transmission coil (Tx) switch off time. Careful selection and/or control of the voltage of the intermittent voltage pulses can also facilitate a faster suppression of the flyback pulse and so a faster transmission coil (Tx) switch off time. Further the relative configuration and separation of the Tx and Rx coils and the presence of a field free region can also impede the flyback pulse resulting in faster commencement of (useful) data acquisition, in other words, in one example, the signal at the Rx coil has reduced sufficiently so as not to swamp the detecting circuitry, which includes significant amplification to find the tiny very fast decaying eddy currents. The time taken for the current developed in the Tx coil to dissipate will depend on the maximum current developed, for example, which in turn will depend on many characteristics of the Tx coil as well as the applied voltage pulse.

The inductance of the transmitter coil (Tx) and/or the voltage applied to it (to develop a suitable current) are preferably sufficient to access a detection zone e.g. to induce eddy currents in objects within a zone of interest at a pre-selected (e.g. limited) distance or range 5 of distances from a transmitter coil (Tx), but not so large that the resultant transient flyback pulse (on switch off of the applied voltage pulse) endures for a very long time, unduly overlapping with and swamping fast decay eddy currents in objects of interest (such as sharp edges, sharp points, sharp discontinuities such as cracks or defects and so on). In other words, the inductance of the transmitter coil (Tx) and the applied voltage must be 10 large enough to provide a signal into the detection zone but not so large as to remain unduly long following switch off.

In Figure 1, apparatus 10 comprises, in this example embodiment, a single transmitter Tx/receiver Rx coil 20, a pulse generator circuit 30, a measuring circuit 40 and a microcontroller 50. Microcontroller 50 may be provided by one or more microcontrollers and/or computing devices such as an ASIC, PIC controller, PC, laptop and so on. Pulse generator circuit 30 comprises, in this example embodiment, a power mosfet 26 for providing a voltage pulse to Tx/Rx coil 20. Microcontroller 50 provides pulse control 22 to an optional optical isolator 24 and so to a power generating device such as power mosfet 26. A power input line 32 provides power to apparatus 10. Measuring circuit 40 comprises a pick-off circuit 28 which may be in the form of one or two stage low-noise high-gain preamplifiers with voltage clipping which delivers an output to an analogue to digital port on a microcontroller. The VCC components on the power input line 32 may comprise linear volt, low-noise voltage regulators as would be understood by someone skilled in the art. Typically, the power input line delivers 12 volts. A damping circuit 12 is provided across the outputs of single Tx/Rx coil 20.

Damping circuit 12, together with power mosfet 26, provides a coil driver circuit for coil(s) 20. Damping circuit 12 comprises a damping resistor R1 (14) for critically damping across coil 20. Such a critically damped resistor R1 is known in the art. Damping circuit 12, however, also provides a trim potentiometer resistor R2 (18) for coil matching. Also in parallel to coil 20 is a diode 17 which is reverse biased during the transmit pulse but forward biased as the current collapses post-pulse, allowing the power to be dissipated in the Zener diodes as described below.

Damping circuit 12 here comprises damping resistors and further comprises an active transient suppression component comprising one or more (reverse bias) Zener diodes. In 35 this case this is external to the power mosfet. In other embodiments the active transient suppression component can provided alone or may indeed be internal to the power mosfet 26 (and so strictly speaking no separate damping circuit is required) The inventor has appreciated that the very fast switch off of the transmitter coil Tx allows previously unseen very fast switch-off eddy currents from sharp conductive objects to be 5 observed. By providing a transient suppression component across (here a single) Tx/Rx coil 20, the flyback pulse across coil 20 is damped very quickly indeed.

Typically, the transient suppression devices are provided by one or more Zener diodes, for example one or more and, typically, a chain of Zener diodes, across the transmit coil are provided with sufficient capability to dissipate the voltage of the flyback pulse across the Zener diode chain as quickly as possible to provide a faster inductor switch-off time. These may be internal or external to the power mosfet 26.

The Zener diode in reverse bias configuration throttles the flyback pulse by breaking down when hit with the flyback pulse, so it discharges the transient voltage of the flyback pulse very quickly. Example Zener diodes include two or more Zener diodes in a chain across the transmitter coil (Tx). In one example, 6 x 30V Zener diodes have been used, in another, a chain of 5 x1 00V, 100W Zener diodes have been used. In one example, a 400V MOSFET has been used, in another, a 600V MOSFET. In one example, a 1 kO, 2W resistor R1 and a 2 kW, 2k0 1010 adjustable resistor with a 2000 series resistor have been used. Various supply capacitors are used to allow larger currents to be supplied to the coil (not shown), for example of 2000 -8000 microfarads (e.g. 6600 microfarads).

It has been noted that the larger the voltage dissipated across the active transient suppression component e.g. Zener chain, the faster the inductor switch off time and the faster the data acquisition period can commence.

Various embodiments and improvements have enabled the reduction of the switching (turn-25 off time) by more than 10ps (although the turn-off time varies dependent on the inductance of the coil and the voltage across the coil). This is particularly advantageous for the detection of knives which have very short eddy current decay life times of generally less than 30ps, and even less than 20ps 30ps, or ps, or.10 ps) . Examples of resistors parallel to the transmitter coil (Tx) include a 1kW resistor or a 2W 30 2kohms adjustable resistor, optionally together with a 200 Ohm series resistor. This allows for more control of damping and improvement towards optimal damping of the flyback current at turn off time.

Various example coil configurations are shown in Figures 2 to 5. Figure 4 shows a single transmitter Tx / receiver Rx coil 20 that may be used in the apparatus of Figure 1. Figures 2 and 3 show separate but overlapping, at least in part, transmitter Tx coil(s) 20A and receiver coil(s) 20B. Whilst not shown, it will be understood that receiver coil Rx 20B may be placed in the same plane as transmitter coil Tx but to one side of it (so not overlapping) to further impede the effect of the flyback pulse from the transmitter coil received at the receiver Rx coil 20B. This is one of many examples of how to assist in impeding (here deconvolufing) the response seen in the receiver Rx coil 20B from the effect of the flyback pulse in the transmitter Tx coil.

Figure 5 shows an alternative arrangement in which three co-planar coils are provided, a Tx coil 20A, an Rx coil 20B, and a compensation coil Cx (20C) in between an outer Tx coil 20A and an inner Rx coil 20B. Compensation coil Cx (20C) is counter-wound to Tx coil 20A to provide a field-free region within a Cx coil 20C. The compensation coil Cx (200) is provided with the same pulse as the Tx coil 20A so that the overlapping region between the Tx coil 20A and the Cx coil 20C within which Rx coil 20B resides is generally or substantially a field-free region.

The table in Figure 6 shows the detection distance limits for a body-worn 30cm x 40cm single coil embodiment, and an 80cm x 80cm poster coil configuration with a smaller 22cm circular receive coil inside a larger Tx coil Several items were introduced individually into the detection zone and a distance from a centre of the receive coil was determined, or from the centre of the Tx/Rx single coil in the body-worn embodiment. Using one or more of the methods and techniques in the present invention, the maximum useful range of detection for various blades was determined. It is anticipated that further investigation of the relative arrangements of the transmitter and receiver coil and the use of the various digital signal processing techniques described in this disclosure will facilitate further extension of these limits of detection, albeit these are already of a useful order of magnitude.

Figure 7 shows the output from the pick-off circuit 28 seen, for example, on a display of a microcontroller or other suitable display device. The analogue signal is divided into 1024 channels. The voltage pulse P can be seen between approximately channel 200 and channel 300, and the flyback pulse F can be seen immediately after ceasing of the voltage pulse, the decay of the flyback pulse occurs relatively swiftly and appears to have an exponential shape. The reduction in signal to noise between Figure 8 and Figure 7 after 200 sweeps and signal averaging can be seen.

Figure 9A shows the output in volts versus time in microseconds over the time of the voltage 35 pulse P. The applied voltage pulse P is approximately -12V on the Tx coil (. This pulse is seen at the detecting pick off circuit as -70mV (shown on the graph) due to clipping and inversion -to prevent overloading the pick-off circuit)) The pulse P is followed by a flyback pulse F clipped at a measured level of 4V (after amplification by 1000 and inversion by one or more amplifiers in the pick off circuit 28 (seen as 400mV in Fig 9A as a divide by 10 5 probe was used). The duration R of the flyback pulse is the recovery or dead time when the flyback pulse is too large to be seen. Once the flyback pulse falls to a certain level, typically around 4mV (amplified to 4V by pick off circuit 28), data acquisition can commence and continues during a later data acquisition period D. In other words the flyback pulse has fallen to a sufficiently low level to be digitised (and so taken up by the pick off circuit and 10 A/D converter). Pick off circuit 26 may also be known as a pick off and amplifier circuit.

The amplification and clipping is to ensure that it fits within the input range of the ADC (analogue to digital converter) on the micro controller, using as much dynamic range as possible. The amplifier circuit clips the pulse itself to -0.7 (as it is presented to the micro controller) to ensure that a large negative going pulse does not damage the micro-controller. There is one stage of inversion, so the (say) 12V outgoing pulse across the coil presents as 0.7V to the ADC.

In Figure 9A, the transmitter pulse is on (at the transmitter coil Tx) for time period P (typically of pulse length of 90-120ps). The flyback pulse F immediately follows switch off of this pulse. The duration of the flyback pulse is to be reduced (e.g. minimised) as much as possible, as this determines the recovery (or dead) time period R when no measurements can be taken. The data acquisition period D follows the recovery period R and can commence when the flyback pulse has fallen below a predetermined level (e.g. 4mV). The repetition rate of pulses is typically once every millisecond (ms). It may be once every 500 microseconds, or even faster, to allow faster SIN (signal/noise) improvement. It typically collects several hundred data points along the decay curve during this time. An upper limit on the repetition rate is determined by the requirement to have the decay signal fall to (virtually) zero and stabilise before recommencing the sequence. This will depend on the various electronic components.

Various example configurations of coils and pulse configurations including: pulse duration, pulse voltage, and a coil configuration are shown in the table in Figure 9B with measured recovery times R. It can be seen that from this table that, somewhat counter-intuitively, using a higher voltage pulse to generate more eddy current in the target objects, which one might think would produce a higher signal, in effect simply extends the recovery time R (the dead time) resulting in a mask of very fast-falling eddy currents. Therefore, it has been appreciated by the inventor that carefully controlling those elements which affect recovery time R is particularly important for successful detection of sharp conductive objects.

There are a whole variety of elements which might be so controlled (many of which impede the effect of the flyback pulse on the Rx coil), including pulse voltage, pulse duration, pulse repetition rate, coil size, coil shape, coil configuration, coil orientation, relative coil shapes, and/or configurations, and/or orientations, and/or number of turns, and/or wire thickness, and/or separation, and/or distance to the detection zone, and/or inductance, and/or presence of a compensation coil, adjustments of the coil driving circuitry such as the introduction of active transient suppression devices, the use of an associated damping circuit 12, for example, by the introduction of active transient suppression devices e.g. in the form of a chain of Zener diodes 16, (or indeed by the use of a power mosfet 26 with an internal active transient suppression device-such as one or more Zener diodes) and so on. It is the discovery and realisation by the inventor that exceedingly fast-decaying eddy currents of the order of 10-20ps are swamped by the flyback pulse and, furthermore, that, if the flyback pulse is suitably impeded, these tiny, very fast decaying signals can be rendered visible using advanced signal processing techniques such as multiple sweeps and averaging signal to noise, regular measurement of background, indeed regular measurement of multiple sweeps of background and an average background and subsequent subtraction of such a background of a response when a potential target object is within the detection zone.

In Figure 10 response scans On the Rx coil) with a knife in the detection zone for recovery times R of 20ps (42) and 30ps (44) are shown demonstrating the increase in amplitude of the detected signal when the recovery time R is reduced from 30ps to 20ps.

Figure 10 shows clearly the increased signal response 42 seen in the presence of a knife when the recovery time of the flyback pulse has been controlled to be 20ps. In contrast, the response 44 when the same knife was present in the same location and subject to the same subsequent digital signal processing is between half and two thirds smaller when the recovery time R of the flyback pulse is increased to 30ps. It is thought that this increase in sensitivity (and so detectability of sharp objects at useful distances) has not been demonstrated previously. This illustrates the effectiveness (in increasing the magnitude of the response signal) of reducing the recovery time relative to the decay time of the object (here a knife) being detected and showing the additional data available for later analysis when the recovery is much shorter.

Figure 11 shows a typical background response 46 at the receiver coil with no object 35 present within the nominal detection zone of around 1m and a typical response 48 at the receiver coil with a mobile phone present within around 0.75m after background subtraction. The phone is detectable to around 1.4m with one example of the PI apparatus of the invention (a 16-turn coil of 32cm diameter). Here the waveform of the mobile phone is inverted because this was digitised prior to inversion.

Figure 12 shows overlaid normalised response data from a pulse induction system of the invention showing a knife response 52 when a ferrous, sharp, kitchen knife is present in the detection zone and a non-ferrous response 54 when an aluminium drink can is present (on a separate occasion) in the detection zone. Each of the detection sweeps (after background removal) have been normalised to 1 on the intensity scale for easier comparison. It has been observed by the inventor that almost invariably the response is shorter-lived for sharp objects (or thin bladed objects) such as knives than for blunter objects in which the smallest dimension is larger e.g. drinks cans, mobile phones and keys when the background has been subtracted and data is normalised. In other words different objects can be distinguished from one another by the apparatus and methods of the invention.

Figure 13 is similar to Figure 12 and shows comparable responses that have been normalised to 1 for a knife response, 56 when a knife is present in the detection zone, and a phone response 58 when a phone was present in the detection zone. This again shows the knife response is much shorter lived than that for a mobile phone. Even so, this difference is incredibly hard to observe without background subtraction, and preferably also data averaging, and preferably also normalisation. Thus, using the methods and apparatus described in this application enables different types of objects and combinations of objects to be distinguished from each other and indeed in many embodiments to be identified even when present in combination with other objects.

Figure 14 is similar to Figure 13, although this has not been normalised and the original response is shown before inversion. Here, a very swiftly-falling knife response 62 is shown after background subtraction alongside a phone response 64 when only a phone is present in the detection zone. Also shown is a phone and knife response 66 when both are present in the detection zone, again, multiple sweeps of the data and background subtraction have been used.

Respective single, exponential curves (solid line) fit well to the data (dots) for the knife and mobile phone responses (62 and 64) but not to the response data (66) when both a knife and a mobile phone are present at the same time in the detection zone. Single order exponentials fit well to the knife response 62 and the phone response 64. Time constants can be derived from these for each type of object. Although not shown, a single order exponential curve did not provide a good fit to the phone and knife response data 66. A double exponential (solid line) has been fitted (using least squares fit) to the scan data 66 when both a knife and a mobile phone are present, and the deviation (sum of square errors) between the fit and the data is now much smaller, indicating the higher order function of double exponential is a good fit. Time constants for each 'item' present can be derived from this double exponential, although this can have uncertainties. It is enough to know that a single exponential is a bad fit to indicate that two objects are present.

It can be determined if two or more objects are present by fitting a first order exponential function to decay data, and determining a bad fit to the data and/or fitting a second order 10 exponential function to decay data and determining a good fit to the data.

Selection and training of neural networks is quite well known and a full description will not be provided here. Nevertheless, Figure 15 shows examples of the structure of a feed-forward neural network with one hidden layer with three neurons and four input variables to provide a single output. In the present case, a supervised backpropagation feed-forward neural network is used, for example, the supervision may include output data information such as threat / no threat, as well as input information such as drinks can, mobile phone, knife and so on, and one or more combinations thereof (e.g. see information which may be used in training in step 170 in Fig 17B). Other neural network models can be used.

The training data used to train the neural network is represented by the output data in Figure 16. Figure 16 shows a set of results from a neural network analysis of the original 550 data sets used to train the network initially. The network is trained such that the output neuron in the output layer is trained to give 0 when no threat is present and 1 when a threat is present. The y-axis is the threat level, below 0.2 is no threat, and above 0.8 is threat, and in between is unclassified. The data has been re-fed into the now trained neural network and has correctly classified the data as shown in Figure 16.

In the figure, data sets 1-150 (70) represent responses for background when no object was present or when a (here) non-threat object such as mobile phone was present. For example, data points from data sets such as those seen in Figure 11 from which everything post flyback pulse within the data acquisition period D (as seen in Figure 9A) is fed into the input layer neurons e.g. as seen in Figure 15). Data sets 151-300 (72) represent responses when different types of knives were present. Data sets 301-360 (74) represent responses when no objects were present (background). Data sets 361-380 (76) were a large kitchen knife held further from the detector (Rx coil). Data sets 381-400 (78) were a mobile phone, here an iPhone ®, held further from the detector (Rx coil). Data set 401-420 (80) were those from a large folding knife. Data sets 421-480 (82) were for no object present or keys present. Data sets 481-550 (84) were for a knife and mobile phone, here an iPhone 0, together in close proximity. Figure 16 demonstrates that different types of objects and different combinations of objects can be classified (e.g. into threat/no threat categories), and indeed may be categorised with appropriate training and sufficient input data, and optionally using other variable limitations (e.g. using distance as part of the input data), to classify the type of object (e.g. a sharp such as a knife, a phone, a set of keys and so on) and so distinguish one type of object from another.

Figure 17A shows a method 100 of detecting responses indicative of sharp, conductive objects. Step 110 comprises measuring a background response (e.g. when a detection zone is empty). Step 115 comprises measuring a detection response (e.g. when the detection zone is occupied). Step 120 comprises, optionally, subtracting the background response from the detection response. Step 125 comprises, optionally, repeating steps 110 and/or step 115, and averaging (or summing) the result of the subtraction.

In addition or alternatively averaging can take place before the subtraction. Step 130 comprises, optionally, collecting and averaging a number of background responses to determine an average background response. Step 135 comprises, optionally, collecting and averaging a number of detection responses to determine an average detection response. Step 140 comprises, optionally, subtracting the averaged background response from the averaged detection response.

Step 145 comprises, optionally, normalising one or more sweeps or average (or summed) responses. Background subtraction facilitates the distinguishing of a tiny signal and determination of an absolute magnitude of signal. Normalisation (particularly after background subtraction) facilitates comparison of the signal shapes and timings relative to the same normalised peak signal response.

Whilst such techniques are not new, it is believed this is the first time such techniques have been applied to the vary fast decaying eddy current signals from conducting objects with particular characteristics (sharp edges or points or discontinuities and such like). As far as the inventor is aware, there had previously been no reason to apply such techniques to eddy current decay responses.

Figure 17B shows a method 200 of classifying the objects using fitting or a trained ANN. Here, step 150 comprises classifying by one or more of: using fitting e.g. by fitting one or more exponentials, optionally first or second order or high order exponentials, optionally to determine one or more time-constants, to characterise on object, even a previously unseen object. Optionally a trained artificial neural network (ANN) may be used to characterise an object, even a previously unseen object.

Step 155 comprises establishing a risk score based on the results of, or as part of, the classifying step. Step 160 comprises displaying the risk score -threat, no threat, unclassified/nul result.

Figure 18A is a method 300 for gathering data and training an artificial neural network. Here, step 165 comprises gathering and/or determining and/or collating one or more items of information. Step 170 comprises training an ANN using information relating to one or more of input and/or output components: such as one or more of: Threat/no threat Sharp/not sharp - Object present/not present Mobile phone present/not present Knife present/not present -Drinks can (and/or food can) present/not present - Object sub-type (e.g. type of knife, type of mobile phone, type and/or size of can etc.) Tx and Rx coils both present Single Tx/Rx coil -Cx coil present (Tx, Rx, Cx) Coil size, and/or configuration, and/or shape and/or orientation, and/or relative orientation and/or number of turns, and/or wire thickness, and/or separation, and/or distance to detection zone and/or inductance - Number of sweeps On background and/or detection responses) Tx Pulse amplitude (voltage and/or current applied to the transmission coil), and/or repetition rate.

In particular the training data may include output components including one or more characteristics of the objects being detected e.g. sharp / not sharp, threat / no threat, discontinuity/ no discontinuity, defect/ no defect and so on.

Figure 188 shows a method 400 of using a trained ANN to determine a characteristic. Here step 175 comprises using a trained ANN to classify detection responses e.g. threat/no threat, knife/no knife. This may be followed by step 160, displaying a score, e.g. a risk score, or displaying a characteristic (e.g. threat/no threat, knife/no knife) or other result.

Figure 19 shows typical responses for a background 90, a knife 92 and a mobile phone response 94 individually) in the detection zone are shown after processing by a microprocessor. The knife and mobile phone responses have been subject to background removal and digital signal averaging e.g. over 200 sweeps. Here, the data has also been 5 subject to normalisation. Optionally, a new background is obtained by a button press (when it is known there is nothing of interest in the detection zone) or automatically. A detector such as a motion or infra-red detector can be used (e.g. a PR detector) to detect when the designated detection zone is free from a potential carrier (e.g. an intruder) and a background can be taken. Typically, this is taken every few minutes. The digital signal 10 processing averaging and the background data removal is automatic within, for example, microcontroller 50 (or within a further microcontroller or computing system).

Optionally, data is automatically analysed by a trained neural network e.g. within the or a separate microcontroller 50 (or within a further microcontroller or computing system) as threat or no threat.

Figure 20 shows two responses as an analogue output from the receiver coil over 2 pulses with first response 96 when no object or background response is present and a second response, a knife response 98, when a knife is present. Signal averaging e.g. over 200 sweeps has taken place in a microcontroller. The signal is clipped to 3.3V and -0.7V to avoid damaging the microcontroller. Horizontal units are in ps and vertical units are in 10mV. This illustrates the very tiny effect the presence of a knife has on the response. Without the understanding developed by the inventor, there would be no reason to consider that responses 96 and 98 were anything other than a variation in set up or ambient conditions. However, by following the methodologies and techniques outlined in this disclosure, the presence of a knife or, indeed, a mobile phone can be observed as shown in Figure 19.

In one or more embodiments, the present application provides apparatus and methods for the detection of conductive objects, especially metal objects and objects of other conductive material e.g. carbon, silicon, conductive plastic-based materials, and, in particular, conductive objects e.g. of metal or other conductive material having particular characteristics, including for particular shapes or other characteristics such as sharp metal objects like knives. The application also provides apparatus and methods for the detection and/or classification of particular characteristics of conductive objects e.g. the presence of one or more sharp point(s), sharp edge(s), narrow crack(s) or other sudden discontinuity etc. in the conductive material structure. The application also provides to apparatus for classifying conductive objects (e.g. comprising a classification module within a microcontroller 50 for fitting or comprising a trained ANN), a method of manufacturing a classification module (a trained ANN), a method of gathering data for such a classification module (a trained ANN), and a method of training an ANN for such a classification module.

In one or more embodiments a number of innovations contributed to successful 5 implementation: an improved e.g. optimised damping circuit to allow fast switch off. This works by improved e.g. optimal damping, optionally using a parallel resistor and preferably using one or more, preferably a series of Zener diodes and, optionally, a further diode to achieve a faster shut off. The Zener diodes in effect provide a fast-acting short circuit when hit with the flyback pulse in reverse bias. Simulations have been carried out to demonstrate this using PSPiCE. Further, preferably the noise added by the parallel resistor is removed by multiple sampling of the eddy current decay signals and preferably also multiple sampling of the background prior to background removal.

Rapid recovery time of the system between pulses is possible. Signals from knives etc. have usually decayed after a few tens of microseconds, so fast switch off allows a new transmit sequence (a new transmit pulse) to be started (with a resultant increase in the signal to noise ratio) -Fast data acguisifion (10 Mega samples per second) Summation or Averaging e.g. over 100+ (e.g. 200 or more) sweeps (over 700 channels in each decay cycle (1024 channels), e.g. discarding the first 324 channels). As the microcontroller generates the pulse and then samples, the system is fully deterministic so it makes it easy to choose a sync point for this digital signal processing.

The ability to subtract a background (itself having high signal to noise ratio being an average); being able to subtract a precise background means that even very small changes (i.e. a small metal object in the field) is detected, which means (along with the fast response) that the system is very sensitive to even tiny metallic items.

The additional sensitivity provided by the improved signal to noise in the present system from earlier measurements, and in one or more preferred embodiments repeated measurements, facilitated by the faster switch off (which itself facilitates the gathering of the response earlier) contributes to improved signal to noise in the measured responses. Nevertheless, the presence or not of objects of particular characteristics such as knives or sharps (such as needles) is not easily visible to the eye, and further analysis such as the classification using least squares fitting and/or a trained neural network is particularly helpful. It may be that these can be dispensed with, but these are present in preferred embodiments.

Thus the apparatus and methods of the invention and, in particular, those used for revealing very short-lived signals, characteristic of objects with particular characteristics under investigation, have been used to train a neural network, which has then been implemented to classify previously unseen objects.

In one or more embodiments, this very sensitive detection of the nature and characteristics of the metal objects being detected is possible because the background is not convoluted, as it is in soil or the ground, with unknown, unremovable objects such as ferrite particles, rusty nails and so on that cannot be removed to provide a suitable background.

This ability to detect a representative background and, typically, an actual background in the absence of the object itself but not in the absence of any other surrounding (although somewhat distant objects e.g. street furniture), facilitates improved sensitivity by the derivation of the amplitude of the response as well as a decay rate of the response. This enables the provision of not only responses to a classification system such as a fitting module and/or a neural network (both for the training phase and for the later execution phase) but also the provision of absolute and even normalised, absolute responses to these. This further facilitates discrimination between different types of objects and, in particular, objects that may be harmful such as knives and other sharps from objects that are not harmful such as mobile phones or keys, even when these are carried together.

In one or more embodiments, integration over the decay curve is carried out to provide an increase in signal. In other words, summing of the channels into a single number to provide another figure of merit. Again, this means that tiny objects can be detected some distance away (e.g. a knife at 1 metre or a sim card close to the detector).

Classification using a trained neural network; Classification seems to work when a knife and mobile phone are both in the field such that the signal is a convolution of two exponential decay rates. Classification has been found to be particularly successful, it is thought because a) of the ability to subtract a background (itself having high signal to noise ratio being an average) and/or b) the high signal to noise from the averaging of the decay signals.

In one or more embodiments, the whole curve is measured, which may include multiple simultaneous decays from two or more objects.

In one or more embodiments, the background is removed from the signal over the whole curve, which enables determination of the amplitude of the signal as well as the decay rate. 5 This may be used to give additional information to the neural network (during training and/or during later classification).

In one or more embodiments, the distance from a transmitter coil (Tx) and/or to a sensing (Rx) coil from a target object may be used. This may be used to give additional information to the neural network (during training and/or during later classification). This removes an ambiguity and improves classification. The provision of a predetermined detection zone also assists in this, and may be defined by a range of distances within which an object to be detected can be expected. Other ambiguities may be reduced or removed as described herein.

In one or more embodiments, responses can be normalised according to distance, which 15 makes the classification into threat / no threat more robust.

In the present invention, it is possible to measure the distance to target and so the depth can be approximately known e.g. within a detection zone next to a poster or over a scan floor pad. There is no soil response when measuring in air or water. Therefore, it is possible to get some indication of size (by the intensity of the response after background removal) and shape/conductivity (by the swiftness of the decay). It is thought this improves classification (and, indeed, it seems to work in practice).

Thus, in at least one embodiment, classification is thought to be more reliable because of the improved signal to noise, background removal, and (in further iterations) using the distance to target to normalise the signals according to range. At the moment, data is collected at different ranges and added to a large training set, but if range information were available, then the intensity of the signal will be more useful and remove some ambiguities (e.g. a small object close to the detector might get confused with a large object with a similar decay time some distance away).

In one or more embodiments, the apparatus of the invention uses a parallel resistor (and Zener diodes) for critical damping and fast discharge of the Tx coil. Further digital signal averaging techniques are used to reduce the noise and improve the sensitivity. The invention can pick up small metal items (such as mobile phone SIM cards, small knives), which has proved not to be possible with other systems. Further, in one or more embodiments, the apparatus also has increased sensitivity by reduction of the inductance of the transmit/receive coil, such that the flyback current is reduced and the switch off time is shortened.

Until the present invention, as far as the inventor is aware, no one seems to have 5 appreciated that sharp threat objects such as knives have a shorter decay time than most other body-worn objects such as phones, keys, and coins. Zips and buttons can also be detected, but as these are very small, typically only when relatively close to the detector and can therefore be included in non-threat items. Even closed flick knives have been detected using the apparatus and methods of the invention. -The apparatus and methods 10 of the invention are able to detect smaller amounts of metal further away than prior art metal detectors, due to the fast turn off time and resultant sensitivity to quickly decaying signals (that are normally missed). It will detect zips and buttons, and folded flick knives although the signals from the knives might be small these still decay equally quickly, if compared, say, to a mobile phone.

The present invention, by shortening the duration of the flyback pulse, enables previously hidden eddy currents to be revealed. The fast turnaround of pulses facilitates the addition of digital signal processing to improve the signal to noise ratio and facilitating the measurement and removal of a background to firstly reveal, and secondly reveal in sufficient detail and clarity, the signal from the eddy currents of particular kinds of objects e.g. very sharp objects such as knives and it is also envisaged from very narrow cracks and defects in metals and other conductive materials.

It is also apparent that by providing apparatus or components of the apparatus for use in the methods of the invention about the person of an individual such as a police officer or prison guard, they can be alerted upon approach of another person carrying a concealed weapon. Various garments such as trousers, vests, jackets and hats and even gloves can be envisaged. For example a transmitter coil may be present in a piece of street furniture and a receive coil may be present in the garment. When in a glove this allows, say, a security guard to extend an arm towards a potential assailant reducing the detection distance to any objects on the assailant without the guard having to come closer than up to 2 m from the potential assailant (1m from the person to the glove, and 1m from the glove to the guard's body). The receiver (Rx) coil is then hidden in the glove. Indeed, the receiver coil could also be in a detector wand of some length (which may then be somewhat harder to hide).

It will be apparent to those skilled in the art from this disclosure that responses from thin bladed objects such as knives are generally different than other body carried clutter such as phones, keys etc. Further these can be differentiated by the generally shorter-lived response and/or different amplitudes seen in the response (e.g. when detected in the same zone(s) (range of distances) or at fixed distances). Further, it has been shown that a neural network can classify the results. It has also been shown that the careful removal of background using data averaging and improvements in signal to noise means that the system is very sensitive to tiny eddy current response signals.

It has also been shown that the system works better when the response is fastest (e.g. with lower inductances, and/or lower voltages across the coils, and/or an improved (e.g. optimised) damping circuit, and/or perhaps a separate receiver coil spatially separated from the transmit coil, perhaps in a field free region). One or more or all of these improve(s) (e.g. optimise(s)) the detectability of the response of short-lived eddy current decays such as those from knives.

It has also been shown that a spatially separated receiver can have more turns (which improves sensitivity) rather than trying to get the transmit coil to double up as a receiver coil. Good results have been seen using a large 80 cm square system, with a small (e.g. much smaller) receiver coil with many turns located in the middle of the transmit coil e.g. as seen in Figure 2, which has been demonstrated to routinely detect thin bladed objects at 1 metre or more distance from, in this instance, the common plane of the Tx and Rx coils.

Whilst the apparatus is described particularly in connection with the detection of concealed weapons such as knives, potentially at the same time as non-harmful objects such as mobile phones or keys, the apparatus and methods of the invention may be applicable to other forms of conductive items, particularly metal item detection, for example in emergency rooms or during surgery, in clinics and hospitals, The apparatus and methods of the invention are also applicable to the detection of very short-lived eddy currents e.g., from defects or discontinuities or breaks in conductive objects such as pipes, wires, plates and so on. It is particularly useful if these are present in a fluid medium such as air or water so that, if such items are not in themselves moveable, the apparatus can be moved to one side and can generate a representative background from an adjacent segment of the fluid medium that does not contain the object of interest.

These and other applications of the methods and apparatus of the invention will be apparent to those skilled in the art from the present disclosure.

P applied voltage pulse (usually a series of pulses, repeated regularly) F flyback pulse R recovery time D data acquisition period B(t) background response (may be averaged or summed) D(t) detection response (when a target is in the detection zone) (may be averaged or summed) A(t) = D(t)-B(t) absolute detection response (may be averaged or summed or derived from averaged or summed B(t) and/or D(t)) apparatus (coil(s), driving circuit, measuring circuit) 12 damping circuit 14 damping resistor 16 (active) transient suppression device(s) (e.g. one or more Zener diodes) 18 trim potentiometer resistor coil (single Transmitter (Tx) / Receiver (Rx) coil) 20A Tx coil, 20B Rx coil, 20C compensation coil (Cx) 22 pulse control (from microcontroller) 24 optical isolator 26 power mosfet 28 pick-off circuit (may include one or more amplifiers) pulse generator 32 Vcc (voltage common collector) -power input line 34 ground response measurement circuit (may include one or more amplifiers) 42 response (knife present) recovery time R = 20ps 44 response (knife present) recovery time R = 30ps 46 background response (after background subtraction -hovers around zero) 48 response (mobile phone present) (after background subtraction) microcontroller (e.g. within PC, laptop etc.) 52 normalised background subtracted response (knife present) 54 normalised background subtracted response (drinks can present) 56 normalised background subtracted response (knife present) -200 averaged sweeps 58 normalised background subtracted response (mobile phone present) -200 averaged sweeps 62 output data (dots) and fitted curve (line) (knife present) 64 output data (dots) and fitted curve (line) (mobile phone present) 66 output data (dots) and fitted curve (line) (both mobile phone and knife present together) 70 training data response -background no threat present, no object present, or mobile phone is present 72 training data response -various knives present 74 training data response -background, no object present 76 training data response -large kitchen knife, further away 78 training data response -mobile phone, further away training data response -large folding knife 82 training data response -background, no threat present (no object present or keys present) 84 training data response -knife and mobile phone both present 100 a method of detecting conductive objects a method of classifying detected responses 300 a method of training an artificial neural network (ANN) 400 a method of using an ANN

Claims

Claims 1 An apparatus for detecting conductive objects (e.g. in a fluid medium) comprising: - a first coil as a transmitter coil (Tx) for transmitting a magnetic field, the transmitter coil (Tx) arranged with respect to a predetermined detection zone; the first coil, or a second coil, as a receiver coil (Rx) for detecting a magnetic field induced in one or more conductive object(s) within the predetermined detection zone; a power supply for applying intermittent voltage pulses to the transmitter coil (Tx); - a means for impeding the effect of a flyback pulse on the receiver (Rx) coil at the end of a voltage pulse such that data acquisition can commence.
2. An apparatus according to claim 1 in which a recovery time R after the voltage pulse is the duration of the flyback pulse, and/or in which a recovery time R after the voltage pulse is a characteristic decay time of the flyback pulse.
3. Apparatus according to claim 1 or 2 in which the means for impeding the effect of the flyback pulse is configured such that a or the recovery time R after a voltage pulse ends is less than an eddy current decay time in at least one target object of interest.
4. An apparatus according to any preceding claim in which the means for impeding the effect of a flyback pulse comprises at least one transient suppression component associated with the transmitter coil (Tx).
5. An apparatus according to any of claim 4 in which the at least one transient suppression component comprises at least one Zener diode in reverse bias configuration associated with, optionally parallel to, the transmitter coil (Tx).
6. An apparatus according to any of claims 4 to 5 in which the at least one transient suppression component comprises a chain of two or more Zener diodes in series.
7. An apparatus according to any of claims 4 to 6 in which a damping circuit comprising a damping resistor, is provided in parallel with the transmitter coil (Tx).
8. An apparatus according to any preceding claim in which the means for impeding the flyback pulse is configured to damp the flyback pulse in a time period selected from: 50ps, 45ps, 40ps, 35ps, 30ps, 25ps, 20ps, 8ps 5ps.
9. An apparatus according to any preceding claim in which the applied pulse has a predetermined limited voltage.
10. An apparatus according to any preceding claim comprising a pick-off circuit for measuring a signal from the receiver coil.
11. An apparatus according to any preceding claim in which the transmitter coil (Tx) has a predetermined limited inductance.
12. An apparatus according to any preceding claim comprising a control unit (e.g. a microcontroller) for controlling one or more of: timing of delivery of voltage pulses, pulse spacing, pulse length, pulse mark-space ratio, pulse voltage, and timing of measurement of the response.
13. An apparatus according to any preceding claim comprising a control unit for controlling transmission of a pulse from the transmission coil (Tx) into a fluid medium during a first time period Ti; and also for controlling measurement of the induced signal in the receiver coil (Rx) during a second time period T2 after the end of the first time period and, optionally, for removing or ignoring a third time period T3 of the induced signal in the receiver coil (Rx) forming a first part of a second time period (12), such that data is collected during a data acquisition time period after the end of the third time period T3.
14. An apparatus according to any preceding claim in which the pulses are one or more of: square pulses, equi-spaced, of the same length, of varied lengths, of constant length during a particular data acquisition sequence.
15. An apparatus according to any preceding claim in which the receiver coil (Rx) is separate from the transmitter coil (Tx).
16. An apparatus according to any preceding claim in which the receiver coil (Rx) is smaller than the transmitter coil (Tx).
17. An apparatus according to any preceding claim in which the receiver coil (Rx) is of a different size and/or shape and/or number of coils from the transmission coil (Tx).
18. An apparatus according to any preceding claim in which the means for impeding comprises or further comprises a compensation coil (Cx) for creating a magnetic-free region between the transmitter coil (Tx) and the receiver coil (Rx).
19. An apparatus according to claim 17 or 18 in which the transmitter coil (Tx), receiver coil (Rx) and the compensation coil (Cx) are generally or substantially co-planar and/or the compensation coil (Cx) lies within the transmitter coil (Tx) and the receiver coil (Rx) lies within the compensation coil (Cx).
20. A method for detecting conductive objects (e.g. in a fluid medium), in a pulse induction detection apparatus such as that in claims 1 to 19 or, in an apparatus comprising at least a first coil, the first coil acting as a transmitter coil (Tx), and the first, or a second coil, acting as a receiver coil (Rx), the method comprising: -applying a voltage pulse to the transmitter coil (Tx) for transmitting a magnetic pulse from the transmitter coil (Tx) into a predetermined detection zone; - impeding the effect of a flyback pulse on the receiver (Rx) coil at the end of a voltage pulse whereby data acquisition can commence. 20
21. A method according to claim 20 in which the step of impeding comprises one or more of: impeding the effect of the flyback pulse such that a or the recovery time R after a voltage pulse ends is less than an eddy current decay time in at least one target object of 25 interest; - using at least one transient suppression component; - using a damping circuit having at least one transient suppression component; damping a flyback pulse such that the duration of the flyback pulse is less than the expected time for the eddy currents to decay in one or more target conductive objects of 30 interest.
22. A method according to claim 20 or 21 comprising, in a predetermined detection zone, transmitting a magnetic field from the transmitter coil (Tx) and determining a background response (B(t)) on a receiver coil (Rx) when no target is present in a detection zone; introducing a target into the detection zone; transmitting a magnetic field from the transmitter coil (Tx) and detecting a response on the receiver coil (Rx) when a target is present in the detection zone; and, subtracting the background response (B(t)) from the detected response (D(t)).
23. A method according to any of claims 20 to 22 comprising determining an absolute amplitude of a response by subtracting a background response (B(t)) from a detected response (D(t)).
24. A method according to any of claims 22 to 23 comprising repeating the step of determining a background response (B(t)) and determining a summation of background responses (sum B(t)) or average background response (Av B(t)).
25. A method according to any of claims 22 to 24 comprising repeating the step of determining a detected response (D(t)) when a target is present and determining a summation of detected responses (sum D(t)) or an average detected response (Av D(t)) when a target is present.
26. A method according to any of claims 22 to 25 comprising subtracting a background response (B(t)), or an average or summed background response (Av B(t), sum B(t)), from a detection response (D(t)) when a target is present, or from an average or summed detection response when a target is present (sum D(t), Av (D(t)), to determine a background subtracted detected response (A(t)).
27. A method according to any of claims 22 to 26 comprising normalising the amplitude of a background subtracted detected response (norm A(t)).
28. A method according to any of claims 20 to 27 comprising fitting one or more curves to one or more of: a detected response (D(t)), a background subtracted detection response (A(t)), a normalised background subtraction detection response (norm A(t)).
29. A method according to claim 28 comprising extracting one or more decay constants from the fit and using such to classify a target object, and/or part of a target object and/or a characteristic of a target object.
30. A method according to any of claims 20 to 29 comprising one or more of the following steps: -controlling the voltage of the applied pulse controlling number of turns on the transmission coil (Tx); controlling number of turns on the receiver coil (Rx), - controlling one or more dimensions and/or shapes of at least one coil (e.g. Tx, Rx, Cx) -controlling diameter of the wire of the transmission coil; - controlling diameter of the wire of the receiver coil (Rx); -controlling the number of repetitions of the detection response; - controlling number of repetitions of the background response; controlling separation of the receiver coil (Rx) and transmission coil (Tx); providing a compensation coil (Cx); - controlling number of turns of the compensation coil (Cx); -controlling the diameter of wire of the compensation coil (Cx); controlling the distance to the target (average, medium, maximum, minimum) from at least one coil (Tx, Rx, Cx); -controlling the distance to the target from, for example, a centre of the transmission coil (Tx) or other fixed point on the transmission coil (Tx) coil; controlling the distance from compensation coil (Cx) to receiver coil (Rx) and/or transmission coil (Tx).
31. A method of preparing training data for a neural network model comprising using the apparatus of claims 1 to 19 and/or the methods of claims 20 to 30 to gather training data, the training data comprising: a) one or more input components, the input components comprising one or more of: a detection response (D(t)); a background subtracted (e.g. absolute) detection response (A(t)); a normalised, background subtracted (e.g. absolute) detection response (norm A(t)); and b) one or more output components, the output components comprising one or more characteristic(s) of an objects being detected e.g. one or more of: nature of the object; a mobile phone; several mobile phones; one key; several keys; a knife; a plurality of knives; characteristic information about the conductive object e.g. its sharpness and/or its depth and/or its width and/or its size; c) optionally, using one or more output components selected from: threat; no threat; sharp; not sharp; thin; not thin; blade; not blade; edge; not edge; defect; no defect; discontinuity; no discontinuity.
32. A method of preparing training data according to claim 31 in which one or more input components comprise one or more characteristics of the measuring apparatus and/or measuring arrangements.
33. A method according to claim 32 in which the characteristics of the apparatus and/or measuring arrangements comprises one or more of: the voltage of the applied pulse(s) - number of turns on the transmission coil (Tx); number of turns on the receiver coil (Rx); one or more dimensions and/or shapes of at least one coil (Tx, Rx, Cx) - diameter of the transmission coil (Tx) wire; -diameter of the receiver coil (Rx) wire; number of repetitions of the detection response; - number of repetitions of the background response; -separation of the receiver coil (Rx) and transmission coil (Tx); presence of a compensation coil (Cx); number of turns of the compensation coil (Cx); - diameter of wire of the compensation coil (Cx); distance to the target (average, medium, maximum, minimum) from at least one coil (Tx, Rx, Cx); -distance to the target from, for example, a centre of the transmission coil (Tx) or other fixed point on the transmission coil (Tx) coil; -distance from compensation coil (Cx) to receiver coil (Rx) and/or transmission coil (Tx).
34. A computer-implemented method of training a neural network comprising using the apparatus of claims 1 to 19 and/or the method of claims 20 to 30 and/or training data prepared using claims 31 to 33, the method comprising: providing data to a neural network comprising one or more of: -a detection response (D(t)); - an absolute detection response (A(t)); - a normalised, absolute detection response (norm A(t)); and/or characteristics of the object being detected e.g. sharp / not sharp, threat / no threat.
35. An apparatus comprising a neural network trained in classifying target objects using the apparatus of claims 1 to 19, and/or using the method of claims 20 to 30, and/or using training data from the method of claims 31 to 33, and/or the computer-implemented method of claim 34, optionally wherein the neural network was trained using an absolute detection response (A(t)) or a normalised absolute detection response (norm A(t)) for classifying objects.
36. A method of classifying conductive objects comprising using the apparatus of any of claims 1 to 19 or 35, 37 or 38 and/or using the method of any of claims 20 to 30 and/or training data prepared using claims 31 to 33, and/or the computer implemented method of claim 34 to classify objects.
37. An apparatus according to any of claims 1 to 19 or 35 concealed in a poster, floor tile, doorway, doorway frame, and/or an item of street furniture.
38. An item of clothing comprising an apparatus according to any of claims 1 to 19 or 35.