FLODAF: Fuzzy Logic Applied to a Multi-Sensor Data Fusion Model

This work presents the development of a demonstrator for testing the performance of Multisensor Data Fusion applied to Naval Target Classification; the model used for demonstration has been developed by the authors for comparing different algorithms and techniques with special attention to Fuzzy Logic influence. Although this project is not classified, some data have been modified because of the confidential nature of the work itself, made in co operation with Italian Navy. The model combines signals/features coming from different sensors such as ESM, Radar, IFF and allow to enable different alternatives in term of data fusion architecture. ESM Alignement D Features Continue Fuzzyfication Alignement Posizionale Fuzzyfication Alignement Radar Alignement D Features Continue Fuzzyfication Alignement Posizionale Fuzzyfication Alignement IFF Alignement D Features Continue Fuzzyfication Alignement Posizionale Fuzzyfication Alignement Other Sensors Alignement D Features Continue Fuzzyfication Alignement Posizionale Fuzzyfication Alignement Tracking & Classification Tracking & Classification Tracking & Classification Gating Data Alignment & Association Co rre lati on Tracking Classification Identity Declaration Identity Declaration Identity Declaration In this Figure it is possible to see how features coming from different sensors are treated and processed in the general software architecture. It is even possible to expand the model to include other sensors Not-classified experimental results are summarised in the paper in reference to Dempster-Shafer and Bayesian methods changing parameters and architecture on 9 different scenarios representing surface, and air targets (i.e. support ships, cruisers, missiles, aircrafts and helicopters) from friend, foe and neutral parties. The demonstrator includes a Graphic User Interface and it has been implemented using C++ for Windows NT platform as software tool for supporting FLODAF (Fuzzy Logic Data Fusion) project devoted to evaluate the benefits and requirements related to the application of fuzzy logic in the target classification. The demonstrator uses a special Database including the features of different platforms, emitters and devices; the authors defined the record format for gathered information and Databases for storing reliability and tolerance on measurement, fuzzy sets and probability functions. The experimental results have been analysed by using DOE (Design of Experiments) and multivariable sensitivity analysis for identify the most critical factors. The evaluation of fuzzy logic potential within the framework of data fusion applied to naval sensors was tested through the development of a technological demonstrator. This ad hoc analysis tool was implemented in C++ and, under certain hypotheses, can be used to carry out scenario evaluations. The tool was tested and then used to carry out an experimental analysis to identify the influence of the para meters considered. The technological demonstrator focused on studying the fusion process in relation to a classification of targets. With this in mind operations were carried out on developed scenario data and it was possible to test just the data correlation phases on the single sensors to obtain the declarations/evidence intervals, implementing the supplied association of signals. This work could be defined innovative since it uses Fuzzy Logic in a process traditionally developed by using classic mathematical algorithms. Recently Lockheed Martin Canada R&D Team has addressed its efforts towards MSDF algorithms optimization in order to allow real time identification and tracking, and this work gave the authors the occasion to do another step in a field that was never developed before at least for the Italian Navy. The innovation was given by the introduction of Fuzzy Logic in Multi Source Data Fusion, this could improve the performance of the identification and tracking process wherever it's not possible or not satisfying to use traditional methods. The inputs of the demonstrator includes IFF, Radar, ESM features for classifying the target . Inside Data Fusion Processing are made also the fuzzyfication and defuzzyfication operations. The sequence of fuzzy logic processing can be divided into 2 broad functions: inference and defuzzyfication. Inference processing begins with the developement of the production rules in the form of “if then” statements, also referred to as fuzzy associative memory. The antecedent or condition block of the rule begins with “if” and the consequent or conclusion block begins with “then”. The value assigned to the consequent or conclusion block is equal to the logical product of the activation values of antecedent membership functions (fuzzy sets). The activation value is equal to the value of membership function at which it is intersected by the input variable at the time instant being evaluated. If the antecedent block for a particular rule is compound statement connected by “and”, the logical product is the minimum value of the corresponding activation values of the membership functions. If the antecedent block for a particular rule is a compound statement connected by “or”, the logical product is the maximum value of the activation values. When the logical product for the antecedent is zero, the value associated with the consequent membership function is also zero. A defuzzyfication operation is performed to convert the fuzzy values, represented by the logical products and consequent membership function, into a fixed and discrete output; the approach used is a correlation product inference, that preserves more information than correlation minimum inference. The Fuzzy sets used in the development of the demonstrator are the following: Fuzzy sets of ESM measures