Threat estimation systems

Abstract

A threat estimation system arranged to provide a first artificial neural network (ANN) that operates in accordance with a first set of weight values, and an engagement planner. The threat estimation system is arranged to receive a source of sensor data; estimate from the sensor data, using the first ANN, a threat result for a threat indicated by the sensor data; and determine from the threat result, using the engagement planner, a threat action to be performed. The threat estimation system is further arranged to receive an updated first set of weight values for the first ANN, and update the first ANN to operate in accordance with the updated first set of weight values.

Description

[0001]XA23409 Threat estimation systemsField of the InventionThe present invention concerns threat estimation systems. More particularly,but not exclusively, the invention concerns threat estimation systems that use artificialneural networks (ANNs) to perform threat estimation, methods of updating suchthreat estimation systems, and methods of training threat state prediction ANNs.Background of the InventionThreat state estimation systems are a key component of systems such asground-based air defence systems, air-to-air products and weapon aiming. Suchsystems use sensor data to infer characteristics of a threat in a defence context, forexample using radar data in an air-defence context to infer the class / type of a threat.The performance of such systems can be critical. Existing systems requireupdates to be deployed, to take account of new types of threat, and for generalperformance improvements. Existing systems rely on a model-based approach, forexample using models based on Kalman filters to identify threats. Each threat modelconsists of a bespoke definition of equations, parameters and heuristics. Thesedefinitions need to be created and tested prior to deployment, which is time-consuming and requires a high degree of specialist skill and knowledge from theengineer developing the definitions. As a consequence, the process of creating anddeploying updates can be onerous and slow to provide.It would be desirableth less input by sp t be able to create and deploy updates more quickly andflexibly, wi ecialist engineers required.The present invention seeks to mitigate some or all of the above-mentionedproblems. Alternatively or additionally, the present invention seeks to provideimproved threat estimation systems, improved methods of updating threat estimationsystems, and improved methods of training threat state prediction ANNs. XA23409 -2- Summary of the InventionIn accordance with a first embodiment of the invention there is provided athreat estimation system comprising:a processor, andmemory, wherein the threat estimation system is arranged, using the processor andmemory, to provide:a first artificial neural network, ANN, that operates in accordance with a firstset of weight values; and an engagement planner;wherein the threat estimation system is arranged to:receive a source of sensor data;estimate from the sensor data, using the first ANN, a threat result for a threatindicated by the sensor data; anddetermine from the threat result, using the engagement planner, a threat actionto be performed;and wherein the threat estimation system is further arranged to:receive an updated first set of weight values for the first ANN; andupdate the first ANN to operate in accordance with the updated first set ofweight values.By using a first ANN, the functionality of the threat estimation system can beprovided by ANN training on existing data, without requiring time-consumingdesigning of bespoke algorithms by highly-skilled engineers. However, by having thefirst ANN then provide the threat result to an engagement planner, which thendetermines the threat action to be performed, a modular system can be provided inwhich the disadvantages of having an ANN simply provide the complete operation ofthe threat estimation system can be avoided. In particular, due to their nature theunderlying design assumptions and basis of operation of ANN-based systems isobfuscated compared to bespoke algorithms, and their operation can be lesspredictable, paiticularly where they receive input dissimilar to that on which theyhave been trained. By having the output of the first ANN passed to an engagement XA23409 -3-planner, which then uses that output to determine the threat action to take, thosedisadvantages can be mitigated. In addition, output from other ANNs, or other non-ANN-based systems, can be used as well by the engagement planner to determine thethreat action to take.Further, additional training of the first ANN can be performed to take accountof new types of threat, or on better andlor larger datasets to provide improvedperformance of the first ANN. This can be done remotely, for example by theproviders of the threat estimation system, in order to determine an updated set ofweight values for the first ANN, without the first ANN of the threat estimation systemitself being required. As no designing of bespoke algorithms is required, this can beless time-consuming to perform and not require such a high level of skills. Theupdated set of weight values can then be provided to the threat estimation system, andused to update the set of weight values of the first ANN, so the first operation of thefirst ANN is updated. This allows updates to be more quickly and flexibly provided.The sensor data may be radar data. Alternatively or additionally, the sensordata may be sonar data, video data, any other suitable type of sensor data, or acombination thereofThe threat estimation system may be a naval threat estimation system, anaircraft threat estimation system, a land threat estimation system, or may be used toevaluate threats in any other setting or combination of settings.The threat result may be a single result or a set of results. The threat actionmay be a single action or a set of actions. The threat action may be to perform a threatresponse such as launching a missile or set of missiles, and / or may be the display ofthe threat result or results to an operator. The threat estimation system may performthe threat action. The first ANN may be a recurrent neural network, RNN. Where the threatestimation system comprises a plurality of ANNs, some or all of them may be RNNs.The first ANN may be a threat classification ANN arranged to perform threatclassification, such that the threat result estimated by the threat classification ANN isa classification of the type of the threat indicated by the sensor data. The type of thethreat may identify the type of missile, or vessel / aircraft / vehicle, or any otherappropriate classification of the threat.The threat estimation system may be further arranged to: / afl-rtv,-4- / provide a novelty detection ANN that operates in accordance with a noveltydetection set of weight values, determine from the threat result estimated by the threat classification ANN ifthe threat indicated by the sensot data is of a type known to the first ANN, andwherein the engagement planner is arranged to determine the threat action tobe performed using the determination by the novelty detection ANN. This can allowthe threat estimation system to take account of the threat classification ANNproviding a threat classification for a threat which it was not trained upon.The threat estimation system may be further arranged to: receive an updated novelty detection set of weight values for the noveltydetection ANN; andupdate the novelty detection ANN to operate in accordance with the updatednovelty detection set of weight values.The novelty detection set of weight values for the novelty detection ANN maybe determined using a subset of the first set of weight values for the threatclassification ANN. The updated novelty detection set of weight values may similarlybe determined using a subset of the updated first set of weight values for the threatclassification ANN. The threat classification ANN may comprise a layer with internal memorystates that operates in accordance with a subset of the first set of weight values, andwherein the novelty detection ANN comprises a layer with internal memory statesthat operates in accordance with the subset of the first set of weight values. The threatclassification ANN and the novelty detection ANN may share a layer with internalmemory states. The layers with internal memory states may be long-short-term-memory (LSTM) layers. LSTM layers are commonly found in RNNs, which providea learned set of features of the input, and in this way can provides a staticrepresentation of the threat. Alternatively, the layers with internal memory states maybe gated recurrent units (GRUs), for example. By having the threat classificationANN and the novelty detection ANN use the same or related LSTM layers, the riskthat the novelty detection ANN might distinguish “novelties” based on threatcharacteristics different from those used by the threat classification ANN is mitigated,helping to avoid both false positives and negatives. XA23409 -5- The threat estimation system may be further arranged to store at least onepredetermined set of default settings, and the engagement planner may be arranged, inthe case that the novelty detection ANN determines that the threat indicated by thesensor data is not of a type known to the threat classification ANN, to determine thethreat action to be performed using the default settings. This can mitigate problemsthat can arise with systems that utilise ANNs components, as the underlying designassumptions made during development of ANN-based systems are more obfuscatedcompared to those for existing systems based on bespoke algorithms, and generallythe behaviour of systems under conditions where design assumptions have beenviolated is less predictable and robust for ANN-based systems than for bespoke-designed algorithms.The threat estimation system may be further arranged to: provide a further ANN that operates in accordance with a further set of weightvalues; estimate from the sensor data, using the further ANN, a further threat result forthe threat indicated by the sensor data; andwherein the engagement planner is arranged to determine the threat action tobe performed using the further threat result.The threat estimation system may be further arranged to:receive an updated further set of weight values for the further ANN; andupdate the further ANN to operate in accordance with the updated further setof weight values.The further ANN may be arranged to use the threat result estimated by the firstANN to estimate the further threat result from the sensor data. In particular, where thefirst ANN is a threat classification ANN, the threat classification result may be usedby the further ANN.The further ANN may be a threat state prediction ANN arranged to performthreat state prediction, such that the further threat result estimated by the threat stateprediction ANN is a future position of the threat indicated by the sensor data.The threat state prediction ANN may be arranged to take as input a flagindicating if a valid input is being provided, and may be arranged:when the flag indicatcs a valid input is provided, determine the position of thethreat at a next time step using sensor data; andXA23409-6- when the flag indicates a valid input is not provided, determine the position ofthe threat at a next time step using the position of the threat determined by the threatstate prediction ANN at the preceding time step.This can allow threat state prediction, which can be of particular importance,to be improved, particularly for “long” prediction teams (for example more than fiveprediction steps into the future). The flag can be used during training, allowing themultiple training samples from a single trajectory, with different ratios of“observations” (for which the flag is set to indicate a valid input is being provided tothe threat state prediction ANN), and “prediction” (for which the flag is set to indicatea valid input is not being provided to the threat state prediction ANN, so the threatstate prediction ANN is forced to use its own prediction). The flag can be also usedduring deployment, being set to indicate a valid input is being provided where anobservation is available, and to indicate a valid input is not being provided where aprediction is required. This can provide an improvement over conventionalapproaches, as during training the threat state prediction RNN is required to learn topredict the state for varying time-horizons, rather than just a single time-step ahead.The further ANN may be a reachable area prediction ANN arranged toperform reachable area prediction, such that the further threat result estimated by thereachable area prediction ANN is a reachable area of the threat indicated by the sensordata.The further ANN may be a launch point prediction ANN arranged to performlaunch point prediction, such that the further threat result estimated by the launchpoint prediction ANN is a launch point of the threat indicated by the sensor data.The threat estimation system may comprise any combination of one or moreANNs, to provide any combination of one or more of the above flinctionalities.In accordance with a second embodiment of the invention there is provided acomputer program product comprising machine-readable instructions arranged, whenexecuted on a computing device comprising a processor and memory, to provide anyof the threat estimation systems described above.XA23409-7- In accordance with a third embodiment of the invention there is provided amethod of updating any of the threat estimation systems described above, comprisingthe steps of: training an ANN for a threat estimation system using sensor data and desiredthreat results for threats indicated by the sensor data;determining the set of weight values in accordance with which the ANNoperates; generating an updated set of weight values from the determined set of weightvalues; providing the updated set of weight values to the threat estimation system; andupdating an ANN of the threat estimation system to operate in accordancewith the updated set of weight values. The updated set of weight values may just be set of weight values inaccordance with which the ANN operates, or may be a subset or a n1odificationthereof The updated set of weight values may be provided as a data file transmitted tothe location of the threat estimation system, or on a storage medium, or in any otherdesired manner.In accordance with a fourth embodiment of the invention there is provided amethod of training a threat state prediction ANN for a threat estimation system asdescribed above, comprising the steps of: receiving at least one sample of sensor data, wherein the sample comprises aset of positions of a threat at different times;generating a plurality of samples from the at least one sample, wherein eachgenerated sample comprises a subset of the set of positions of the threat at differenttimes; and training the threat state prediction ANN using the generated plurality ofsamples. In this way, a threat state prediction ANN with improved performance can beprovided, as during training the ANN is required to learn to predict the state forvarying time-horizons, rather than just a single time-step ahead as with conventionaltraining methods.wnl;, -8- It will of course be appreciated that features described in relation to one aspectof the present invention may be incorporated into other aspects of the presentinvention For example, the method of the invention may incorporate any of thefeatures described with reference to the apparatus of the invention and vice versa5 Description of the DrawingsEmbodiments of the present invention will now be described by way of10 example only with reference to the accompanying schematic drawings of which:Figure is a schematic diagram of a threat estimation system according to a firstembodiment of the invention;Figure 2 shows in more detail the threat classification RNN and novelty detection15 RNI’4 of the threat estimation system of Figure 1,Figure 3is a schematic diagram of a training system for the threat estimationsystem of Figure 1,Figure 4 is a schematic diagram of the workflow of the training by the trainingsystem of Figure 3,20 Figure 5 are graphs representing samples used in the training of the threat stateprediction RNN of Figure 1;Figure 6 is a flowchart showing the deployment of an update to the threatestimation system of Figure 1; andFigure 7 is a schematic diagram of a computing device in accordance with an25 embodiment of the invention. 30 A threat estimation system in accordance with a first embodiment of theinvention is now described with reference to Figure to 6. XA23409 -9- Figure is a schematic diagram of the threat estimation system 100. The threatestimation system 100 comprises a plurality of RNNs, whose training is described indetail later below. The threat estimation system takes as input real-world radar data source 101,i.e. “live” radar data from a radar system. In particular, radar track data from the real-world radar data source 101 is used, giving the latest observed 3D threat position andvelocity, range and range rate.The radar track data from the real-world radar data source 101 is passed to athreat classification RNN Ill. The threat classification RNN ill generates anestimated threat type for the threat, i.e. a classification of the type of the threat, basedon the behaviour of the threat as shown by the radar track data. The threat type mayfor example be the type of missile that the threat is.The threat classification RNN 111 is shown in more detail in Figure 2. As willbe understood by the skilled person, as the threat classification RNN 111 is an RNN,it comprises a long-short term memory (LSTM) layer 130, which generates a latentspace embedding, i.e. the learned set of features of the threat, that provides a staticrepresentation of the input radar track data (which is a variable length time-series) thatis used by the threat classification RNN 111 to classify the threat type. The latentspace embedding output by the LSTM layer 130 is passed to a fully-connected layer131, whose output is passed to a softmax layer 132, and then to a classifier layer 133,which provide the output of the threat classification RNN Ill, i.e. the estimated threattype for the threat.In addition, the latent space embedding output by the LSTM layer 130 of thethreat classification RNN 111 is used by a novelty detection RNN 112, which is alsoshown in Figure 1. The latent space embedding is passed to a one-class support vectormachine 140, which uses the Latent space embedding to generate a Boolean “isnovel?” flag. This flag indicates whether the threat is novel, i.e. dissimilar to types ofthreat considered during the training of the RNNs of the threat estimation system 100.The use of the shared latent space embedding by both the threat classification RNN 111 and the novelty detection RNN 112 reduces the risk that could otherwisearise that a novelty detection algorithm might distinguish “novelties” based on a set ofthreat characteristics that do not include the characteristics to which the systemXA23409- 10-performance is sensitive (and so don’t adequately indicate that the system is operatingoutside of the conditions it was designed for), andlor include characteristics that areirrelevant to system performance (and so provide false positives of “novelties”). The threat classification generated by the threat classification RNN 111 is alsopassed to a threat state prediction RNN 113, a reachable area prediction RNN 114,and a launch point prediction RNN 115, each of which is also passed the radar trackdata from the real-world radar data source 101. The threat state prediction RNN 113generates an estimated future position for the threat, i.e. a 3D trajectory for the threatbetween the current time and predicted impact time. The reachable area predictionRNN 114 generates an estimated reachable area for the threat, i.e. a 2D polygonrepresenting the area on the surface the threat can reach. The launch point predictionRNN 115 generates an estimated launch point for the threat, i.e. the 3D position fromwhich the threat was launched.The threat type for the threat generated by the threat classification RNN 111,the “is novel?” flag generated by the novelty detection RNN 112, and the futureposition for the threat generated by the threat state prediction RNI’J 113, are eachpassed to an engagement planner 120. The engagement planner 120 uses these inputsto generate appropriate engagement instructions, i.e. parameters used to control theengagement with the threat, which are passed to a missile launcher 121. While theengagement planner 120 uses outputs from the various RNNs, it is not itself an RNN,but instead uses a bespoke set of logical rules, optimization routines and lookup tablesto generate the engagement instructions.The engagement instructions may include an allocation matrix of specificdefender missiles to specific threats, firing times / windows for missile, estimated “flyout” times for defender missiles, desired turnover angles for missile soft-launch, orany other appropriate parameters. The engagement planner 120 can also generatesensor priorities that are passed back to the real-world radar data source 101 to controlthe radar, in particular a prioritized list of threats for the radar to maintain tracks on,and!or improve the resolution of to support the engagement.In particular, thc “is novcl?’ flag gcncratcd by the novelty detection RNN 112is used by the engagement planner 120 to determine whether it is appropriate to use XA23409 -11-the outputs from the RNNs when generating the engagement instructions, and inparticular, in the event that the threat has been flagged as novel by the noveltydetection algorithm, the engagement planner 120 uses default values for theengagement instructions, rather than perform threat-specific optimization based on theoutput of the RNNs. The reason for this is that the flag indicates that the threat is dissimilar to the types of threat considered during the training of the RNNs, and sothey are operating outside of the conditions for which they were trained. This isparticularly important for systems utilizing ANN components, as the underlyingdesign assumptions made during development of ANN-based systems are moreobfhscated compared to those for existing systems based on bespoke algorithms, andgenerally the behaviour of systems under conditions where design assumptions havebeen violated is less predictable and robust for ANN-based systems than for bespoke-designed algorithms. The reachable area for the threat generated by the reachable area predictionRNN 114, and the launch point for the threat generated by the launch point predictionRNN 115, are passed to an operator display 122 to be displayed to an operator of thethreat estimation system 100. It will be appreciated that the outputs generated by the various RNNs of thethreat estimation system 100 could be used in various alternative or additional ways,for example with alternative or additional outputs being used to generate appropriateengagement instructions, and alternative or additional outputs being displayed to anoperator. Figure 3 is a schematic diagram of a training system for the threat estimationsystem 100. The training system 200 takes as input real-world data 201 of radar data,and modelling and simulations data 202, and are used to generate a training dataset203 of threat radar tracks. The training dataset 203 can then be used to train a set ofRNNs 204 corresponding to the RNNs of the threat estimation system 100, inparticular, a novelty detection RNN 205, a threat classification RNN 206, a threat state prediction RNN 207, a reachable area prediction RISIN 208 and a launch pointprediction RNN 209.Figure 4 is a schematic diagram of the workflow of the training by the training system of Figure 3. The workflow 300 begins with customer inputs 301 beingXA23409- 12-provided, consisting of real-world data gathering 310, and the providing of threatmodels 311, sensor models 312, and scenarios 313. The threat models 311, sensor models 312 and scenarios 313 are used for synthetic data generation 320; this synthetic data may be provided by the customer (i.e. user of the threat estimationsystem 100), may be generated as part.of the training workflow 330, or a combination.The real-world data 310 and the generated synthetic data 320 are used togenerate threat datasets 331. Design work 332 is undertaken to develop an algorithmarchitecture for the threat estimation system 100, and the threat datasets 331 are thenused for ML algorithm training 333 of the RNNs of the training system 200, based onthe algorithm architecture design 332, to generate learned parameters 334, using MLtraining methods which will be well known to the skilled person. Algorithmevaluation 335 is performed, i.e. evaluation of the output of the algorithms using thelearned parameters, and the results of this can be fed back into the ML algorithmtraining 333, and also synthetic data generation 320, to optimise the results of the training. Once a desired amount of training has been performed, the learned parameters334 can be used to perform system-level integration and testing 336, and if such testing is successfully passed, they can be deployed as an algorithm suite 337, i eused in live customer systems. The process deployment of the is described in moredetails below.The operation of the threat state prediction RNN 113, and the training of thecorresponding threat state prediction RNN 207, is now described in more detail withreference Figure 5. In general, threat state prediction is of particular importance to theplanning and execution of an engagement, meaning that its performance can heavilyinfluence the overall system performance. Threat state prediction is used to supportfiring policy decisions, effector trajectory planning, sensor scheduling, and manyother system functions. The threat state prediction RNN 113, by virtue of its particularimplementation, provides improved accuracy of state estimates, particularly for“long” prediction times (more than five prediction steps into the future).Threat state prediction generally requires predicting the value of a singlevariable S over a series of time steps. Conventionally, when training the true value ofthe variable at time k, Sk, is used as the RNN input, and the target value, Tk is chosen XA23409 -13-to be the value of S at the next time-step, i.e. Sk+]. During deployment, in order topredict multiple steps into the future, the RNN will be “primed” with the observedstates so far (So, Si,..., Si). Its own predictions, Zk, are then fed back into the RNNwith a single step delay, repeatedly, until the desired prediction horizon is reached.In contrast, with the threat state prediction RNN 113 a concept of “observationphase” and “prediction phase” is used, and an “inputs valid?” flag is included in theinput to the threat state prediction RNN 113. This flag is used during training,evaluation and deployment of the threat state prediction RNN 113. In addition, thetraining data for the threat state prediction RNN 113 (or rather the correspondingthreat state prediction RNN 207) is modified to derive multiple samples from a singletrajectory, where each derived sample has a different ratio of “observations” to“predictions”, as shown by the samples 401, 402 and 403 of Figure 5. In other words,from a single trajectory of k time steps, rather than using this to train for a desiredresult for time k based using the set of observations for all time steps before k, desiredresults at multiple different time steps n can be trained for (when n may be equal to orless than k), based on different sets of observations prior to n, and not only the set ofobservations for all time steps before n.During the observation phase, the input to the threat state prediction RNN 113is the true state Sk, as in conventional systems, and the “inputs valid?” flag is set to betrue. During the prediction phase, the RNN input is set to null, the target value for theRNN is chosen to be Sk (rather than Sk+J), and the “inputs valid?” flag is set to befalse. In order to predict multiple steps into the future, the threat state prediction RNN113 is again “primed” with the observed states so far (So, Si,..., &), and then iterated(with null inputs) until the desired prediction horizon is reached. However, thisprovides a performance improvement over the conventional approach, because duringtraining, the RNN is required to learn to predict the state for varying time-horizons,rather than just a single time-step ahead.The deployment of an update to the threat estimation system 100 is nowdescribed with reference to the flowchart of Figure 6. First, the RNNs of the systemare trained on new real-world data (step 501), for example as described above withreference to Figures 3 to 5. This results in a set of learned parameters 334, which areweight values extracted from the various trained RNNs (step 502), i.e. weight values XA23409 -14-for their nodes. The extracted weight values are the used to generated an updated setof weight values for the RNNs (step 503). The updated set of weight values maysimply be all of the extracted weight values, or a subset of them and / or a modificationthereof, as desired.The updated set of x;eight values is then proided to the threat estimationsystem WO (step 504), e.g. as a data file transmitted to the location of the threatestimation system 100, or provided on a storage medium. The updated set of weightvalues are then used to update the weight values of one or more of the various RNNsof the threat estimation system 100 (step 505), i.e. the threat classification RNN 111, *novelty detection RNN 112, threat state prediction RNN 113, reachable areaprediction RNN 114, and / or a launch point prediction RNN 115 as appropriate.In this way, the various RNNs of the threat estimation system 100 are updatedto operate based on the training of the corresponding RNNs of the training system200. Embodiments of the invention include the methods described above performedon a computing device, such as the computing device 800 shown in Figure 7. Thecomputing device 800 comprises a data interface 801, thought which data can be sentand received, for example over a network or using a data storage medium such as aCD or USB storage device. The computing device 800 further comprises a processor802 in communication with the data interface 801, and memory 803 incommunication with the processor 802. In this way, the computing device 800 canreceive data via the data interface 801, and the processor 802 can store the receiveddata in the memory 803, and process it so as to perform the methods described herein.Each device, module, component, machine or function as described in relationto any of the embodiments described herein may comprise a processor and / orprocessing system or may be comprised in apparatus comprising a processor and / orprocessing system. One or more aspects of the embodiments described herein comprise processes performed by apparatus. In some embodiments, the apparatus comprises oneor more processing systems or processors configured to carry out these processes. Inthis regard, embodiments may be implemented at least in pail by computer softwarestored in (non-transitory) memory and executable by the processor, or by hardware, orby a combination of tangibly stored software and hardware (and tangibly stored XA23409 -15-firmware). Embodiments also extend to computer programs, particularly computerprograms on or in a carrier, adapted for putting the above described embodiments intopractice. The program may be in the form of non-transitory source code, object code,or in any other non-transitory form suitable for use in the implementation of theprocesses. The carrier may be any entity or device capable of carrying the program,such as a RAM, a ROM, or an optical memory device, etc. While the present invention has been described and illustrated with referenceto particular embodiments, it will be appreciated by those of ordinary skill in the artthat the invention lends itself to many different variations not specifically illustratedherein. By way of example only, certain possible variations will now be described.Where in the foregoing description, integers or elements are mentioned whichhave known, obvious or foreseeable equivalents, then such equivalents are hereinincorporated as if individually set forth. Reference should be made to the claims fordetermining the true scope of the present invention, which should be construed so asto encompass any such equivalents. It will also be appreciated by the reader thatintegers or features of the invention that are described as preferable, advantageous,convenient or the like are optional and do not limit the scope of the independentclaims. Moreover, it is to be understood that such optional integers or features, whilstof possible benefit in some embodiments of the invention, may not be desirable, andmay therefore be absent, in other embodiments.

Claims

XA23409 - 16- Claims 1. A threat estimation system comprising:a processor, andmemory, wherein the threat estimation system is arranged, using the processor and memory, toprovide: afirst artificial neural network, ANN, that operates in accordance with a firstset of weight values; andan engagement planner;wherein the threat estimation system is arranged to:receive a source of sensor data;estimate from the sensor data, using the first ANN, a threat result for a threatindicated by the sensor data; anddetermine from the threat result, using the engagement planner, a threat actionto be performed;and wherein the threat estimation system is flwther arranged to:receive an updated first set of weight values for the first ANN; andupdate the first ANN to operate in accordance with the updated first set ofweight values.

2. A threat estimation system as claimed in claim 1, wherein the first ANN is arecurrent neural network, RNN.

3. A threat estimation system as claimed in claim I or 2, wherein the sensor data isradar data.

4. A threat estimation system as claimed in any preceding claim, wherein the firstANN is a threat classification ANN arranged to perform threat classification, such thatthe threat result estimated by the threat classification ANN is a classification of thetype of the threat indicated by the sensor data.

5. A threat estimation system as claimed in claim 4, further arranged to:XA23409- 17- provide a novelty detection ANN that operates in accordance with a noveltydetection set of weight values;determine from the threat result estimated by the threat classification ANN ifthe threat indicated by the sensor data is of a type known to the first ANN; andwherein the engagement planner is arranged to determine the threat action to beperformed using the determination by the novelty detection ANN.

6. A threat estimation system as claimed in claim 5, wherein the threat estimationsystem is further arranged to:receive an updated novelty detection set of weight values for the noveltydetection ANN; andupdate the novelty detection ANN to operate in accordance with the updatednovelty detection set of weight values.

7. A threat estimation system as claimed in claim 5 or 6, wherein the novelty detectionset of weight values for the novelty detection ANN are determined using a subset ofthe first set of weight values for the threat classification ANN.

8. A threat estimation system as claimed in claim 7, wherein the threat classificationANN comprises a layer with internal memory states that operates in accordance with asubset of the first set of weight values, and wherein the novelty detection ANNcomprises a layer with internal memory states that operates in accordance with thesubset of the first set of weight values.

9. A threat estimation system as claimed in any of claims 5 to 8, wherein the threatclassification ANN and the novelty detection ANN share a layer with internalmemory states.

10. A threat estimation system as claimed in any of claims 5 to 9, further arranged tostore at least one predetermined set of default settings, and wherein the engagementplanner is arranged, in the case that the novelty detection ANN determines that thethreat indicated by the sensor data is not of a type known to the threat classificationANN, to determine the threat action to be performed using the default settings.XA23409 -18 -11. A threat estimation system as claimed in any preceding claim, further arranged to:provide a further ANN that operates in accordance with a further set of weightvalues; estimate from the sensor data, using the further ANN, a further threat result forthe threat indicated by the sensor data; andwherein the engagement planner is arranged to determine the threat action to beperformed using the further threat result.

12. A threat estimation system as claimed in claim 11, further arranged to:receive an updated further set of weight values for the further ANN: andupdate the further ANN to operate in accordance with the updated further setof weight values.

13. A threat estimation system as claimed in claim 11 or 12, wherein the further ANNis arranged to use the threat result estimated by the first ANN to estimate the furtherthreat result from the sensor data.

14. A threat estimation system as claimed in any of claims 11 to 13, wherein thefurther ANN is a threat state prediction ANN arranged to perform threat stateprediction, such that the further threat result estimated by the threat state predictionANN is a future position of the threat indicated by the sensor data.

15. A threat estimation system as claimed in claim 14, wherein the threat stateprediction ANN is arranged to take as input a flag indicating if a valid input is beingprovided, and is arranged:when the flag indicates a valid input is provided, determine the position of thethreat at a next time step using sensor data; andwhen the flag indicates a valid input is not provided, determine the position ofthe threat at a next time step using the position of the threat determined by the threatstate prediction ANN at the preceding time step.XA23409- 19-16. A threat estimation system as claimed in any of claims 11 to 13, wherein theifirther ANN is a reachable area prediction ANN arranged to perform reachable areaprediction, such that the further threat result estimated by the reachable areaprediction ANN is a reachable area of the threat indicated by the sensor data.

17. A threat estimation system as claimed in any of claims 11 to 13, wherein thethrther ANN is a launch point prediction ANN arranged to perform launch pointprediction, such that the further threat result estimated by the launch point predictionANN is a launch point of the threat indicated by the sensor data.

18. A computer program product comprising maehiñereadable instructions arranged,when executed on a computing device comprising a processor and memory, toprovide the threat estimation system as claimed in any preceding claim.

19. A method of updating a threat estimation system as claimed in any of claims ito17, comprising the stepsof training an ANN for a threat estimation system using sensor data and desiredthreat results for threats indicated by the sensor data;determining the set of weight values in accordance with which the ANNoperates;generating an updated set of weight values from the determined set of weightvalues; providing the updated set of weight values to the threat estimation system; andupdating an ANN of the threat estimation system to operate in accordancewith the updated set of weight values.

20. A method of training a threat state prediction ANN for a threat estimation systemas claimed in claim 14, comprising the steps ofreceiving at least one sample of sensor data, wherein the sample comprises aset of positions of a threat at different times;generating a plurality of samples from the at least one sample, wherein eachgenerated sample comprises a subset of the set of positions of the threat at differenttimes; andXA23409 -20 -training the threat state prediction ANN using the generated plurality ofsamples. 4

Images