Title :
Principal component analysis for classifying passive sonar signals
Author :
Soares-Filho, William ; De Seixas, Jose Manoel ; Pereira Caloba, L.
Author_Institution :
IPqM, Brazilian Navy Res. Inst., Rio de Janeiro, Brazil
Abstract :
Principal component analysis in the frequency domain is used for neural identification of the radiated noise from ships. For comparison, components are extracted from three different approaches: linear (PCA) and nonlinear (NLPCA) principal component analysis, and neural discriminating analysis (NDA). The classifier using NDA achieves a classification efficiency of about 93% using only 3 components, while the classifiers using PCA and NLPCA need up to 33 components to reach the same efficiency
Keywords :
neural nets; principal component analysis; signal classification; sonar signal processing; neural discriminating analysis; neural identification; passive sonar signals; principal component analysis; sonar signal classification; Acoustic noise; Acoustic sensors; Frequency domain analysis; Machinery; Marine vehicles; Narrowband; Principal component analysis; Sensor arrays; Signal processing; Sonar equipment;
Conference_Titel :
Circuits and Systems, 2001. ISCAS 2001. The 2001 IEEE International Symposium on
Conference_Location :
Sydney, NSW
Print_ISBN :
0-7803-6685-9
DOI :
10.1109/ISCAS.2001.921380
