الفهرس Only 14 pages are availabe for public view
 |  | from | 92 |  |  |
| | | from | 92 | | |
Abstract
Identifying naval targets (surface or subsurface) is one of the most complicated and time consuming naval operations. Although the sonar operator can perform the task of classification by both listening to the sound on headphones and looking for features in a series of rolling spectrograms, but he could not identify the target using
the same technique.
The main problem that originated this thesis was how to enable the sonar operator to identify the naval targets that are near his platform, using the underwater sound they produce (that is received by the passive sonar), automatically.
In this thesis, we present a comparative study between two well-known identification engines. The first one is based Continuous Hidden Markov Model (CHMM), while the 2nd one is based on Artificial Neural Network (ANN) to identify the naval target. Mel frequency cepstral coefficients (MFCCs), Perceptual Linear Prediction (PLP), and Relative Spectral (RAST A) PLP were selected as the features used to represent the passive sonar signal. The general Gaussian density distribution HMM was developed for the CHMM-based engine. Elman network was developed for the ANN¬based engine. The effect of speed, distance and direction of the target on the identification process was studied.
The results had shown that there was no difference in using either MFCC or PLP with the CHMM-based engine as they gave the same results for Identification Rate (IR) at 91.67% while changing target range, 100% while changing target direction, 58.3% while changing the target speed for simulated targets and 100% when applying to real targets. The results came down to 91.67%, 83.33%, 25%, and 84.2% respectively when using the RAST A features for the same engine.
On the other hand, PLP gave the best performance when using with the ANN-based identification engine at results 91.67%, 83.33%, 50%, and 84.2% respectively. Results came down to 83.3%, 75%, 41.67%, and 73.68% respectively when MFCC features were used and 33.3%, 25%, 16.6%, and 21.05% respectively when RASTA¬PLP features were used.
As an overall conclusion, we had the best performance coming out from the CHMM¬based identification engine with either MFCC or PLP as signal features. Also we found that target speed is the factor that has the maximum effect on the IR, as it decreases the IR by about 40%, due to the decrease of the SNR.