Nedyalko Petrov

Radar Emitter Signals Recognition and Classification with Feedforward Networks

A possible application of neural networks for timely and reliable recognition of radar signal emitters is investigated. In particular, a large data set of intercepted generic radar signal samples is used for investigating and evaluating several neural network topologies, training parameters, input and output coding and machine learning facilitating data transformations. Three case studies are discussed, where in the first two the radar signals are classified in two broad classes – with civil or military application, based on patterns in their pulse train characteristics and in the third one trained to distinguish between several more specific radar functions. Very competitive results of about 82%, 84% and 67% are achieved on the testing data sets.

Self-organizing maps for texture classification

A further investigation of our intelligent machine vision system for pattern recognition and texture image classification is discussed in this paper. A data set of 335 texture images is to be classified into several classes, based on their texture similarities, while no a priori human vision expert knowledge about the classes is available. Hence, unsupervised learning and self-organizing maps (SOM) neural networks are used for solving the classification problem. Nevertheless, in some of the experiments, a supervised texture analysis method is also considered for comparison purposes. Four major experiments are conducted: in the first one, classifiers are trained using all the extracted features without any statistical preprocessing; in the second simulation, the available features are normalized before being fed to a classifier; in the third experiment, the trained classifiers use linear transformations of the original features, received after preprocessing with principal component analysis; and in the last one, transforms of the features obtained after applying linear discriminant analysis are used. During the simulation, each test is performed 50 times implementing the proposed algorithm. Results from the employed unsupervised learning, after training, testing, and validation of the SOMs, are analyzed and critically compared with results from other authors.

Linguistic composition based modelling by fuzzy networks with modular rule bases

This paper proposes a linguistic composition based modelling approach by networked fuzzy systems that are known as fuzzy networks. The nodes in these networks are modules of fuzzy rule bases and the connections between these modules are the outputs from some rule bases that are fed as inputs to other rule bases. The proposed approach represents a fuzzy network as an equivalent fuzzy system by linguistic composition of the network nodes. In comparison to the known multiple rule base approaches, this networked rule base approach reflects adequately the structure of the modelled process in terms of interacting sub-processes and leads to more flexible solutions. The approach improves significantly the transparency of the associated model while ensuring a high level of accuracy that is comparable to the one achieved by established approaches. Another advantage of this fuzzy network approach is that it fits well within the existing approaches with single rule base and multiple rule bases.

Advanced modelling of complex processes by rule based networks

This work presents an application of the novel theory of rule based networks for building models of processes characterised by uncertainty, non-linearity, modular structure and internal interactions. The application of the theory is demonstrated for a flotation process in the context of converting a multiple rule based system into an equivalent single rule based system by linguistic composition of the individual rule bases. During the conversion process, the transparency of the multiple rule based system is fully preserved while its accuracy is improved to a level comparable with the accuracy of the single rule based system.

Sets with incomplete and missing data — NN radar signal classification

We investigate further the problem of radar signal classification and source identification with neural networks. The available large dataset includes pulse train characteristics such as signal frequencies, type of modulation, pulse repetition intervals, scanning type, scan period, etc., represented as a mixture of continuous, discrete and categorical data. Typically, considerable part of the data samples is with missing values. In our previous work we used only part of the radar dataset, applying listwise deletion to get rid of the samples with missing values and processed relatively small subset of complete data. In this work we apply multiple imputation (MI) method, which is a model based approach of dealing with missing data, by producing confidence intervals for unbiased estimates without loss of statistical power (using both complete and incomplete cases). We employ MI to all data samples with up to 60% missingness, this way increasing more than twice the size of the initially used data subset. We apply feedforward backpropagation neural network (NN) supervised learning for solving the classification and identification problem and investigate and critically compare the same three case studies, researched in the previous paper and report improved, superior results, which is a consequence of the implemented MI and improved NN training.

Rule base identification in fuzzy networks by Boolean matrix equations

This paper proposes a novel approach for modelling complex interconnected systems by means of fuzzy networks. The nodes in these networks are interconnected rule bases whereby the outputs from some rule bases are fed as inputs to other rule bases. The approach allows any fuzzy network of this type to be presented as an equivalent fuzzy system by linguistic composition of its nodes. The composition process makes use of formal models for fuzzy networks and basic operations in such networks. These models and operations are used for defining several node identification cases in fuzzy networks. In this case, the unknown nodes are derived by solving Boolean matrix equations in a way that guarantees a pre-specified overall performance of the network. The main advantage of the proposed approach over other approaches is that it has better transparency and facilitates not only the analysis but also the design of complex interconnected systems.

Supervised radar signal classification

This work investigates radar signal classification and source identification using three classification models: Neural Networks (NN), Support Vector Machines (SVM) and Random Forests (RF). The available large dataset consists of pulse train characteristics such as signal frequencies, type of modulation, pulse repetition intervals, scanning type, scan period, etc., represented as a mixture of continuous, discrete and categorical data. Typically, considerable part of the data samples contains missing values. In our previous work we used only part of the radar dataset, applying listwise deletion to clean the samples with missing values and processed relatively small subset of complete data. In this work we apply three different imputation techniques to deal with the missing data: multiple imputation (MI), K-Nearest Neighbour Imputation (KNNI) and Bagged Tree Imputation (BTI). We employ the imputation methods to all data samples with up to 60% missingness, this way increasing more than twice the size of the initially used data subset. Subsequently the three classifiers (NN, SVM, and RF) are employed and the results are analysed and critically compared based on their accuracy to assess the model with the best performance.

Identification of radar signals using neural network classifier with low-discrepancy optimisation

A hybrid low-discrepancy sequences optimisation approach is investigated and used for training neural network classifiers for recognition of radar signal emitters. Two sample case studies are developed in order to demonstrate and evaluate the presented approach. In the first one, generic intercepted radar signals are classified in two broad classes - with civil or military application, based on patterns in their pulse trains, whereas in the second one the classifier is trained to distinguish between several more specific radar functions. Very competitive results of about 84% accuracy are achieved on the testing data sets.

Classifiers accuracy improvement based on missing data imputation

Abstract In this paper we investigate further and extend our previous work on radar signal identification and classification based on a data set which comprises continuous, discrete and categorical data that represent radar pulse train characteristics such as signal frequencies, pulse repetition, type of modulation, intervals, scan period, scanning type, etc. As the most of the real world datasets, it also contains high percentage of missing values and to deal with this problem we investigate three imputation techniques: Multiple Imputation (MI); K-Nearest Neighbour Imputation (KNNI); and Bagged Tree Imputation (BTI). We apply these methods to data samples with up to 60% missingness, this way doubling the number of instances with complete values in the resulting dataset. The imputation models performance is assessed with Wilcoxon’s test for statistical significance and Cohen’s effect size metrics. To solve the classification task, we employ three intelligent approaches: Neural Networks (NN); Support Vector Machines (SVM); and Random Forests (RF). Subsequently, we critically analyse which imputation method influences most the classifiers’ performance, using a multiclass classification accuracy metric, based on the area under the ROC curves. We consider two superclasses (‘military’ and ‘civil’), each containing several ‘subclasses’, and introduce and propose two new metrics: inner class accuracy (IA); and outer class accuracy (OA), in addition to the overall classification accuracy (OCA) metric. We conclude that they can be used as complementary to the OCA when choosing the best classifier for the problem at hand.

Modular rule base fuzzy networks for linguistic composition based modelling

This paper proposes a linguistic composition based modelling approach by networked fuzzy systems that are known as fuzzy networks. The nodes in these networks are modules of fuzzy rule bases and the connections between these modules are the outputs from some rule bases that are fed as inputs to other rule bases. The proposed approach represents a fuzzy network as an equivalent fuzzy system by linguistic composition of the network nodes. In comparison to the known multiple rule base approaches, this networked rule base approach reflects adequately the structure of the modelled process in terms of interacting sub-processes and leads to more accurate solutions. The approach improves significantly the transparency of the associated model while ensuring a high level of accuracy. Another advantage of this fuzzy network approach is that it fits well within the existing approaches with single rule base and multiple rule bases.