Decision Fusion and Micro-Doppler Effects in Moving Sonar Target Recognition

Abstract

.is paper proposes a method for underwater target recognition based on micro-Doppler eects (called STR_MD) using a majority voting ensemble classi:er weighted with particle swarm optimization (PSO) (called MV-PSO). .e micro-Doppler eect refers to amplitude/phase modulation of the received signal by rotating parts of a target such as propellers. Since dierent targets’ geometric and physical properties dier, their micro-Doppler signature is dierent. .is inconsistency can be considered an eective issue (especially in the frequency domain) for sonar target recognition. To demonstrate the eectiveness of the proposed method, both simulated and practical micro-Doppler data are produced and applied to the designed STR_MD. Also, MV-PSO with six well-known basic classi:ers, k-nearest neighbors (k-NN), Naive Bayes (NB), decision tree (DT), MLP_NN, support vector machine (SVM), and random forest (RF), has been used to evaluate the performance of the proposed method. .is ensemble classi:er assigns an instance to a class that most base classi:ers agree on. However, basic classi:ers in a set seldom work just as well. .erefore, in this case, one strategy is to weigh each classi:cation depending on its performance using PSO. .e performance parameters measured are the recognition score, reliability, and processing time. .e simulation results showed that the correct recognition rate, reliability, and processing time for the simulated data at SNR = 5 dB and 10° viewing angle were 98.50, 98.89, and 9.81 s, respectively, and for the practical dataset with RPM= 1200, 100, 100, and 4.43, respectively. .us, MV-PSO has a more encouraging performance in STR_MD for simulated and practical micro-Doppler sonar datasets.