Teager-Huang Analysis Applied to Sonar Target Recognition

In this paper, a new approach for target recognition based on the Empirical mode decomposition (EMD) algorithm of Huang et al. [11] and the energy tracking operator of Teager [13]-[14] is introduced. The conjunction of these two methods is called TeagerHuang analysis. This approach is well suited for nonstationary signals analysis. The impulse response (IR) of target is first band pass filtered into subsignals (components) called Intrinsic mode functions (IMFs) with well defined Instantaneous frequency (IF) and Instantaneous amplitude (IA). Each IMF is a zero-mean AM-FM component. In second step, the energy of each IMF is tracked using the Teager energy operator (TEO). IF and IA, useful to describe the time-varying characteristics of the signal, are estimated using the Energy separation algorithm (ESA) algorithm of Maragos et al .[16][17]. In third step, a set of features such as skewness and kurtosis are extracted from the IF, IA and IMF energy functions. The TeagerHuang analysis is tested on set of synthetic IRs of Sonar targets with different physical characteristics (density, velocity, shape,... ). PCA is first applied to features to discriminate between manufactured and natural targets. The manufactured patterns are classified into spheres and cylinders. One hundred percent of correct recognition is achieved with twenty three echoes where sixteen IRs, used for training, are free noise and seven IRs, used for testing phase, are corrupted with white Gaussian noise. Keywords—Target recognition, Empirical mode decomposition, Teager-Kaiser energy operator, Features extraction.