Optimizing Classification of Spread Spectrum Signals Based on Features Extraction

摘要

Spread Spectrum techniques (SS) attract theattention they are widely used in wireless communicationsand radar. Direct sequence spread spectrum and frequencyhopped spread spectrum are the two most usedtechniques. In this research, the clustering of these signalsis performed by feature-based model. Features areextracted by Gray Level Co-occurrence Matrix (GLCM),gray level run-length matrix, cumulants, moments, PCA,KPCA and Fast-ICA features. Clustering by GLCMfeatures gets the best result which is one of the commontextures features extraction techniques. The selection ofrelevant features is the big challenge. Therefore, the maincontribution is to optimize the SS identification based onclustering techniques by decreasing the number offeatures without accuracy degradation which is based onfilters, sequential forward selection and binarymetaheuristic search strategies such as Binary ParticleSwarm Optimization (BPSO), Genetic Algorithm (GA),bat feature selection (BBA) and hybrid WhaleOptimization Algorithm with Simulated Annealing andTournament mechanism (WOASAT). BPSO as a wrappermethod is proposed to the optimization as it outperformsthe other techniques in terms of accuracy and selectedfeatures with k-means, k-medoids or HAC.