Description:TECHNICAL FIELD
[0001] The present disclosure relates to analysing of actions and activities of a target. In particular, the present disclosure pertains to a method for providing timely inference of intent of a target of interest and a system thereof.

BACKGROUND
[0002] Background description includes information that may be useful in understanding the present disclosure. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed disclosure, or that any publication specifically or implicitly referenced is prior art.
[0003] Analysing the actions and activities of a target aircraft and providing timely inference of its intent is of particular interest for air defence. The aerial attacks on friendly assets are carried out by enemy aircrafts using certain activity profiles. The activity profile refers to the three-dimensional trajectory and kinematics patterns followed by the enemy aircraft. The activity profile chosen for an attack by the enemy is determined by several factors such as performance of the attack aircraft, weapon and adapted tactics, perception of the air defence weapons deployed and geographical features. These factors dictate that a well-planned profile is carried out to inflict maximum damage on the target while minimising threat to the attacker. An ability to detect these profiles of an adversary’s aircraft is important for air defence surveillance operator to increase the level of situational awareness and also to take appropriate action in a timely manner.
[0004] Analysis of actions and activities of a fighter aircraft and inference of its intent falls under the classical domain of recognition and classification problems where in some decision or forecast has to be made based on available temporal sequences of observations. The problem falls under the category of activity and goal recognition and there are multiple solution approaches suggested by different researchers starting from logic based approaches, classical machine learning or probabilistic approaches to deep learning and brain inspired approaches. However, very few works pertains to military domain applications involving partially observable system.
[0005] One of the prior art relates fuzzy logic approach for intent inference based on the analysis of flight profiles for attack aircraft. In addition, in said prior art, the environmental context of the tracked aircraft is also taken into consideration during the execution of the inference process. Though the fuzzy logic techniques are particularly suitable for modelling problems with inherent imprecision properties, but such techniques from certain degree of rigidity and require introducing of all knowledge manually through domain-expert, which may be prone to errors.
[0006] A large number of methods have been applied for intention and plan recognition work as explained by Han T. A. They have classified prior work roughly in two main groups. Consistency approach and probabilistic approach. The problem with the consistency approaches is that they cannot handle well the case where the current observed actions enable more than one intention. Probabilistic approach can give multiple intentions with corresponding probabilities. But further probabilistic approach has limitation of prior probabilities of intentions that are assumed to be fixed. This assumption is not always reasonable because those prior probabilities should in general depend on the situation at hand. The approach used by the Han T. A. considers the multiple intentions recognition case; however the multiple intentions need to be perfectly mutually exclusive. Another limitation of this approach is that it doesn’t consider the constant change in state of the nodes, i.e. temporal evolution of domain variables.
[0007] There is, therefore, a need to provide a solution that may obviate the above mentioned limitations, and enable effective, accurate, and timely inference of intent of a target of interest.

OBJECTS OF THE PRESENT DISCLOSURE
[0008] Some of the objects of the present disclosure, which at least one embodiment herein satisfies are as listed herein below.
[0009] A general object of the present disclosure is to obviate the above-mentioned limitations, and enable effective, accurate, and timely inference of intent of a target of interest.
[0010] An object of the present disclosure is to provide a system and method that relates to a unique design that is composed of three deep learning based networks, i.e., Membership Networks, Rule Network, and Inference Network, mimicking fuzzy logic architecture.
[0011] Another object of the present disclosure is to provide a system and method with a feature to choose a particular network among Membership Network, Rule Network, and Inference Network, and to train the network with data on every training.
[0012] Another object of the present disclosure is to provide a system and method having feature to update or re-train any of these three networks independently.
[0013] Another object of the present disclosure is to provide a system and method that derives membership functions for each feature from data dynamically through deep learning based network without any user intervention.
[0014] Another object of the present disclosure is to provide a system and method that employ long short-term memory (LSTM) and Dense Networks to derive knowledge from data dynamically without intervention of domain experts.
[0015] Another object of the present disclosure is to provide a system and method that is completely data driven and has the capability to derive inference from the features and knowledge captured in the previous layers of the network.
[0016] Another object of the present disclosure is to provide a system and method having feature to customize each network architecture based on volume of data available to training.

SUMMARY
[0017] Aspects of the present disclosure relate to analysing of actions and activities of a target. In particular, the present disclosure pertains to a method for providing timely inference of intent of a target of interest and a system thereof.
[0018] According to an aspect, the present disclosure pertains to a system for providing timely inference of intent of a target of interest (ToI). The system comprising: a processing unit comprising a processor operatively coupled to a memory storing instructions executable by the processor, wherein the processing unit is operatively coupled with a learning engine comprising a membership network segment, a rule network segment, and an inference network segment, the processing unit configured to: fuse sets of track data, associated with the ToI, received from more than one detectors; extract, through the membership network segment, one or more low-level features from the fused set of track data; capture, through the rule network segment, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and determine, through the inference network segment, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly provide timely inference of the intent of the ToI.
[0019] In an aspect, the system may be configured to normalize, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and correspondingly the one or more low-level features may be extracted from the fused set of track data.
[0020] In other aspect, the system may be configured to designate, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight may be assigned to each of the one or more rule values; and correspondingly the temporal dynamics and context of observations may be captured by the system.
[0021] In another aspect, the rule network segment may comprise long short-term memory (LSTM) and Dense Network architecture.
[0022] In another aspect, the system may be configured to train the learning engine based on training-and-testing data sets, and wherein the learning engine may assign a specific weight to at least one of the one or more rule values based on said training.
[0023] In an aspect, the system may be configured to estimate, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly provide timely inference of the intent of the ToI.
[0024] In an aspect, the more than one detectors may comprise different type of RADARs.
[0025] According to another aspect, the present disclosure pertains to a method for providing timely inference of intent of a target of interest (ToI). The method comprising steps of: fusing, at a processing unit operatively coupled with a learning engine, sets of track data, associated with the ToI, received from more than one detectors; extracting, through a membership network segment of the learning engine, one or more low-level features from the fused set of track data; capturing, through a rule network segment of the learning engine, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and determining, through an inference network segment of the learning engine, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly providing timely inference of the intent of the ToI.
[0026] In one aspect, the method may comprise a step of normalizing, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and wherein, the one or more low-level features may be correspondingly extracted from the fused set of track data.
[0027] In other aspect, the method may comprise a step of designating, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight may be assigned to each of the one or more rule values; and correspondingly, the temporal dynamics and context of observations may be captured.
[0028] In another aspect, the method may comprise a step of estimating, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly the method may comprise providing timely inference of the intent of the ToI.
[0029] Various objects, features, aspects and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which like numerals represent like components.

BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The accompanying drawings are included to provide a further understanding of the present disclosure, and are incorporated in and constitute a part of this specification. The drawings illustrate exemplary embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0031] FIG. 1 illustrates an exemplary block diagram of the proposed system, to illustrate its overall working, in accordance with an embodiment of the present disclosure.
[0032] FIG. 2 illustrates an exemplary context diagram of the proposed system, to illustrate its overall working, in accordance with an embodiment of the present disclosure.
[0033] FIG. 3 illustrates a diagram representing neural fuzzy network architecture associated with the proposed system, in accordance with an embodiment of the present disclosure.
[0034] FIG. 4 illustrates a diagram representing membership network associated with the proposed system, in accordance with an embodiment of the present disclosure.
[0035] FIG. 5 illustrates a diagram representing rule network associated with the proposed system, in accordance with an embodiment of the present disclosure.
[0036] FIG. 6 illustrates a diagram representing inference network associated with the proposed system, in accordance with an embodiment of the present disclosure.
[0037] FIG. 7 illustrates a flow chart of the proposed method, to illustrate overall working, in accordance with an embodiment of the present disclosure.

DETAILED DESCRIPTION
[0038] In the following description, numerous specific details are set forth in order to provide a thorough understanding of embodiments of the present invention. It will be apparent to one skilled in the art that embodiments of the present invention may be practiced without some of these specific details.
[0039] If the specification states a component or feature “may”, “can”, “could”, or “might” be included or have a characteristic, that particular component or feature is not required to be included or have the characteristic.
[0040] As used herein the description and throughout the claims that follow, the meaning of “a,” “an,” and “the” includes plural reference unless the context clearly dictates otherwise. Also, as used in the description herein, the meaning of “in” includes “in” and “on” unless the context clearly dictates otherwise.
[0041] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such details as to clearly communicate the disclosure. However the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosures as defined by the appended claims.
[0042] Embodiments explained herein relate to analysing of actions and activities of a target. In particular, the present disclosure pertains to a method for providing timely inference of intent of a target of interest and a system thereof.
[0043] According to an embodiment, the present disclosure pertains to a system for providing timely inference of intent of a target of interest (ToI). The system can include: a processing unit comprising a processor operatively coupled to a memory storing instructions executable by the processor, wherein the processing unit is operatively coupled with a learning engine comprising a membership network segment, a rule network segment, and an inference network segment, the processing unit configured to: fuse sets of track data, associated with the ToI, received from more than one detectors; extract, through the membership network segment, one or more low-level features from the fused set of track data; capture, through the rule network segment, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and determine, through the inference network segment, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly provide timely inference of the intent of the ToI.
[0044] In an embodiment, the system can be configured to normalize, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and correspondingly the one or more low-level features can be extracted from the fused set of track data.
[0045] In other embodiment, the system can be configured to designate, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight can be assigned to each of the one or more rule values; and correspondingly the temporal dynamics and context of observations can be captured by the system.
[0046] In an embodiment, the rule network segment can include long short-term memory (LSTM) and Dense Network architecture.
[0047] In another embodiment, the system can be configured to train the learning engine based on training-and-testing data sets, and wherein the learning engine can assign a specific weight to at least one of the one or more rule values based on said training.
[0048] In another embodiment, the system can be configured to estimate, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly provide timely inference of the intent of the ToI.
[0049] In an embodiment, the more than one detectors can include different types of RADARs.
[0050] According to another embodiment, the present disclosure pertains to a method for providing timely inference of intent of a target of interest (ToI). The method can include steps of: fusing, at a processing unit operatively coupled with a learning engine, sets of track data, associated with the ToI, received from more than one detectors; extracting, through a membership network segment of the learning engine, one or more low-level features from the fused set of track data; capturing, through a rule network segment of the learning engine, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and determining, through an inference network segment of the learning engine, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly providing timely inference of the intent of the ToI.
[0051] In an embodiment, the method can include a step of normalizing, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and wherein, the one or more low-level features may be correspondingly extracted from the fused set of track data.
[0052] In another embodiment, the method can include a step of designating, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight can be assigned to each of the one or more rule values; and correspondingly, the temporal dynamics and context of observations can be captured.
[0053] In another embodiment, the method can include a step of estimating, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly the method can include providing timely inference of the intent of the ToI.
[0054] According to an embodiment, the proposed system relates a novel deep learning based approach designed with neural-fuzzy architecture for analysing the actions and activities of a target of interest (ToI), and correspondingly providing timely inference of the ToI, thereby predicting intent of the ToI. Therefore, the proposed system can serve a great value and play an important role in an air defence scenario.
[0055] FIG. 1 illustrates an exemplary block diagram of the proposed system, to illustrate its overall working, in accordance with an embodiment of the present disclosure.
[0056] Referring to FIG. 1, the proposed system 100 (also, referred to as system 100, herein) can be efficiently utilized for providing timely inference of intent of a target of interest (ToI), wherein the system 100 can include a processing unit 108, which in turn can include a processor operatively coupled to a memory storing instructions executable by the processor. Further, the processing unit 108 can be operatively coupled with a learning engine 110 that can be configured to make the system 100 intelligent and smart by equipping the system 100 with artificial intelligence.
[0057] In one embodiment, the learning engine 110 can include a multi-layered neural network. In other embodiment, the learning engine 110 can include a membership network segment 112, a rule network segment 114, and an inference network segment 116. In an exemplary embodiment, all the network segments, i.e., the membership network segment 112 (also, referred to as membership network 112), the rule network segment 114 (also, referred to as rule network 114), and the inference network segment 116 (also, referred to as inference network 116) can be designed by segmenting distinct layers of the multi-layered neural network.
[0058] In an embodiment, the system 100 can include more than one detectors 102 (also, referred to as detectors 102, herein) operatively coupled to the processing unit 110, where the detectors 102 can be configured to generate sets of data packets based on the detection of the ToI. In an exemplary embodiment, the detectors 102 can include RADARs.
[0059] In other embodiment, the system 100 can include one or more input units 104 (also, referred to as input units 104, herein) operatively coupled to the processing unit 110, where an entity can feed any data related to the ToI, using the input units 104, into the system 100. In another embodiment, the input units 104 can be operatively coupled between the detectors 102 and the processing unit 110, wherein the input units 104 can act as an intermediate channel for transmitting the generated sets of data packets from the input units 104 to the processing unit 110. In an exemplary embodiment, the input units 104 can include, but not limited to, keyboard, joystick, GUI-interface, translator, and compiler.
[0060] In an embodiment, the proposed system 100 can involve processing of observations and information associated with human cognitive process, in which uncertainties and imprecision is usually inherent. In one embodiment, the proposed system 100 can include a unique implementation of deep neural network in three design layers, which can mimic the fuzzy logic architecture to deal with the random and imprecise behaviour of the partially observable.
[0061] In one embodiment, the proposed system 100 can train the learning engine 110 by feeding multiple training-and-testing data sets into distinct layers of the deep neural network. In an exemplary embodiment, distinct layers of the deep neural network can be trained independently. In another exemplary embodiment, the learning engine 110 can be configured to learn automatically during the training, and further it can determine and estimate various aspects related to an input data on real-time basis.
[0062] In an embodiment, input data to the network may include fused track-data obtained from multiple sensors and radars. Further, a low-level feature extraction from the input data can happen at the first layer of networks (Membership Networks), which may be followed by long short-term memory (LSTM), and Dense Networks (Rule Network) to capture temporal dynamics and context of the observations pertaining to various activities and goals of any partially observed agent. Further, the proposed system 100 can be configured to learn automatically said observations and related contexts from the data.
[0063] In an embodiment, a layer of inference network can finally provide prediction in a timely manner in terms of possibility values for each of the activities and goals learned by the system 100. In another embodiment, aerial attacks on friendly assets can be carried out by enemy aircrafts using certain activity profiles, where the activity profiles can refer to the three-dimensional trajectory and kinematics patterns followed by the ToI, i.e., enemy aircraft. In another embodiment, the activity profile chosen for an attack by the enemy aircraft can be determined by several (low-level) features, such as trajectory, performance of the ToI, weapon and adapted tactics, perception of the air defence weapons deployed and other geographical features.
[0064] In an embodiment, implementation of the proposed system 100 can provide a possibility value for enemy air tracks following the activity profiles. In an exemplary embodiment, track kinematics of the ToI can be taken as input via a track source 202 that acts as an input unit 102, then the track kinematics can be provided to a neural fuzzy network 204, and further value of possibility for various activity profiles can be obtained as output.
[0065] According to an embodiment, a conventional fuzzy control system, as illustrated in FIG. 2, can involve an algorithm that may include three steps namely input step, process step, and inference step for determining activities and intent of a ToI.
[0066] In an embodiment, in the input step, the conventional fuzzy control system can map each input variable to membership functions and can find out membership value and truth value from each function defined for that variable. In an exemplary embodiment, let’s say there are 5 membership functions for each variable then there can be 5 membership value for the same. Further, course, speed, altitude, location sensitivity index (LSI), and time of reporting of the track can be used as feature set pertaining to the mapped variables.
[0067] In an embodiment, in the process step, the conventional fuzzy control system can invoke appropriate rules based on the truth value obtained from the input step, and further can generate result for each of the invoked rules. In an exemplary embodiment, a set of rules can be pre-stored in the conventional system, where the set of rules can be encrypted and stored based on knowledge of experts.
[0068] In an embodiment, in the inference step, the conventional fuzzy control system can combine the result of each rule and infer a final output value.
[0069] In an implementation, in order to predict attack profile using the conventional fuzzy control system following data/ information is requird to be chosen and defined –
• Membership functions for each variable,
• Set of rules for each attack profile, and
• inference logic to combine result from each rule.
[0070] However, the proposed system 100, as illustrated in FIG. 3, can be adapted to mimic a fuzzy control network with the learning capability for each of the step as mentioned in the FIG. 2. In an embodiment, the proposed system 100 can include ‘k’ membership networks, i.e., membership network 1, membership network 2… membership network k, for ‘k’ features. Further, output of the membership networks can be fed to the rule network 114, and its output then can be fed to the inference network 116.
[0071] In an embodiment, in the membership network 112, as illustrated in FIG. 4, for instance, if five membership functions are required for each variable, then unlike the conventional fuzzy control system, instead of defining five membership functions, the proposed system 100 can define a neural network as a membership network 112, which can execute instructions to produce five membership values as outputs. In an exemplary embodiment, the membership network 112 can include a two-layer dense network.
[0072] According to one embodiment, the processing unit 108 can fuse sets of track data associated with the ToI, and received from at least one of the detectors 102. Further, the processing unit 108 can extract, through the membership network 112, one or more low-level features from the fused set of track data. In an embodiment, the system 100 can be configured to normalize, through the membership network 112, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and correspondingly the one or more low-level features can be extracted from the fused set of track data.
[0073] In an exemplary embodiment, as illustrated in the FIG. 4, activation used by the membership network 112 in first layer can be tan(h) function, as represented by block 404, where the tan(h) function may normalize input values/ features, which can be obtained at block 402, in range -1 to 1. Further, second layer of the membership network 112 can be a dense layer which may consider every output value from the first layer, and can correspondingly give output value in between 0 to 1 using sigmoid function, as represented by block 406. In an implementation, every neuron in the second layer of the membership network 112 can be seen as a membership function of a variable used in the conventional fuzzy control system.
[0074] Referring to FIG. 5, the rule network 114 of the learning engine 110 can be composed of LSTM (Long Short Term Memory) cell 502 and dense layer. In an embodiment, the rule network 114 can consider output of each input variable from the membership network 112 as its input. In an exemplary embodiment, number of neurons in the dense layer can be said to be number of rules.
[0075] In an embodiment, in case of the conventional fuzzy control system, one has to write set of rules for each attack profile, however, in case of the proposed system 100, written set of rules for each attack profile are not required, instead the rule network 114 can easily learn universal set of rules pertaining to set of rules for each attack profile. Moreover, learning capability of the LSTM cell 502 (also, referred to as LSTM 502, herein) can boost the learning of track pattern. Further, output value of each neuron can give a rule value.
[0076] Referring to FIG. 6, the inference network 116 of the learning engine 110 can learn membership functions for each variable, set of rules for each profile, and attack profile possibility, all at same time with data using back propagation algorithm. In an exemplary embodiment, the proposed system 100 can be tested for three activity profiles, i.e., Low Altitude Toss, Offset Popup Delivery, and High Altitude Dive.
[0077] In an embodiment, the proposed system 100 can include a computer-implemented model that may require a GPU based deep learning server along with machine learning framework (tensorflow/ pytorch) for training and testing of the model. The trained model, thereafter, can be deployed and run on generic GPU based workstation for field implementation.
[0078] Referring to FIG. 7, the proposed method 700 (also, referred to as method 700, herein) for providing timely inference of intent of a target of interest (ToI). In an embodiment, the method 700 can include fusing, at step 702, at a processing unit operatively coupled with a learning engine, sets of track data, associated with the ToI, received from more than one detectors. In other embodiment, the method 700 can include extracting, at step 704, through a membership network segment of the learning engine, one or more low-level features from the fused set of track data.
[0079] In another embodiment, the method 700 can include capturing, at step 706, through a rule network segment of the learning engine, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features. In yet another embodiment, the method 700 can include determining, at step 708, through an inference network segment of the learning engine, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly providing timely inference of the intent of the ToI.
[0080] In an embodiment, the proposed method 700 can also include a step of normalizing, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, wherein, the one or more low-level features can be correspondingly extracted from the fused set of track data.
[0081] In an embodiment, the method 700 can also include a step of designating, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight can be assigned to each of the one or more rule values; and correspondingly, the temporal dynamics and context of observations can be captured.
[0082] In an embodiment, the method 700 can further include a step of estimating, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly the method can then include providing timely inference of the intent of the ToI.
[0083] In an embodiment, the proposed method 700 is novel and first of its kind with deep learning based design with mimicking fuzzy-logic architecture for analysing the actions and activities of fighter aircrafts and providing their future course of actions in advance.
[0084] In another embodiment, the proposed method 700 can enable learning of wide variety of patterns automatically from track data without the requirement of large database while preserving the flexibility and expressiveness.
[0085] As used herein, and unless the context dictates otherwise, the term “coupled”; is intended to include both direct coupling (in which two elements that are coupled to each other contact each other) and indirect coupling (in which at least one additional element is located between the two elements). Therefore, the terms “coupled to”; and “coupled with”; are used synonymously. Within the context of this document terms “coupled to”; and “coupled with”; are also used euphemistically to mean “communicatively coupled with” over a network, where two or more devices are able to exchange data with each other over the network, possibly via one or more intermediary.
[0086] While the foregoing describes various embodiments of the invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof. The scope of the invention is determined by the claims that follow. The invention is not limited to the described embodiments, versions or examples, which are included to enable a person having ordinary skill in the art to make and use the invention when combined with information and knowledge available to the person having ordinary skill in the art.

ADVANTAGES OF THE PRESENT DISCLOSURE
[0087] The present disclosure provides a system and method that enables effective, accurate, and timely inference of intent of a target of interest.
[0088] The present disclosure provides a system and method that relates to a unique design that is composed of three deep learning based networks, i.e., Membership Networks, Rule Network, and Inference Network, mimicking.
[0089] The present disclosure provides a system and method with a feature to choose a particular network among Membership Network, Rule Network, and Inference Network, and to train the network with data on every training.
[0090] The present disclosure provides a system and method having feature to update or re-train any of these three networks independently.
[0091] The present disclosure provides a system and method that derives membership functions for each feature from data dynamically through deep learning based network without any user intervention.
[0092] The present disclosure provides a system and method that employ long short-term memory (LSTM) and Dense Networks to derive knowledge from data dynamically without intervention of domain experts.
[0093] The present disclosure provides a system and method that is completely data driven and has the capability to derive inference from the features and knowledge captured in the previous layers of the network.
[0094] The present disclosure provides a system and method having feature to customize each network architecture based on volume of data available to training.

 
, Claims:1. A system for providing timely inference of intent of a target of interest (ToI), the system comprising:
a processing unit comprising a processor operatively coupled to a memory storing instructions executable by the processor, wherein the processing unit is operatively coupled with a learning engine comprising a membership network segment, a rule network segment, and an inference network segment, the processing unit configured to:
fuse sets of track data, associated with the ToI, received from more than one detectors;
extract, through the membership network segment, one or more low-level features from the fused set of track data;
capture, through the rule network segment, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and
determine, through the inference network segment, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly provide timely inference of the intent of the ToI.
2. The system as claimed in claim 1, wherein the system is configured to normalize, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and correspondingly the one or more low-level features are extracted from the fused set of track data.
3. The system as claimed in claim 2, wherein the system is configured to designate, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight is assigned to each of the one or more rule values; and correspondingly the temporal dynamics and context of observations are captured by the system.
4. The system as claimed in claim 3, wherein the system is configured to train the learning engine based on training-and-testing data sets, and wherein
the learning engine assigns a specific weight to at least one of the one or more rule values based on said training.
5. The system as claimed in claim 3, wherein the rule network segment comprises long short-term memory (LSTM) and Dense Network architecture.
6. The system as claimed in claim 3, wherein the system is configured to estimate, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly provide timely inference of the intent of the ToI.
7. A method for providing timely inference of intent of a target of interest (ToI), the method comprising steps of:
fusing, at a processing unit operatively coupled with a learning engine, sets of track data, associated with the ToI, received from more than one detectors;
extracting, through a membership network segment of the learning engine, one or more low-level features from the fused set of track data;
capturing, through a rule network segment of the learning engine, temporal dynamics and context of observations pertaining to actions and activities of the ToI based on the extracted one or more features; and
determining, through an inference network segment of the learning engine, intent of the ToI taking into consideration the captured temporal dynamics and context of observations, and correspondingly providing timely inference of the intent of the ToI.
8. The method as claimed in claim 7, wherein the method comprises a step of normalizing, through the membership network segment, the fused set of track data by allocating cascaded set of token values to membership functions associated with every variable present in the fused set of track data, and
wherein, the one or more low-level features are correspondingly extracted from the fused set of track data.
9. The method as claimed in claim 7, wherein the method comprises a step of designating, through the rule network segment, one or more rule values to the normalized set of track data, wherein a weight is assigned to each of the one or more rule values; and correspondingly, the temporal dynamics and context of observations are captured.
10. The method as claimed in claim 7, wherein the method comprises a step of estimating, through the inference network segment, possibility of one or more attack profiles taking into consideration the designated one or more rule values, and weights of the respective rule values, and correspondingly the method comprises providing timely inference of the intent of the ToI.