Título: Classificador Neural OnLine para Sonar Passivo utilizando um Processador Digital de Sinais de Alto Desempenho
Autores: Souza Filho, J. B. O.; Seixas, J. M.; Soares Filho, W.
Resumo: The noise radiated from ships in the ocean contains information about their machinery, being normally used by passive sonar systems for detection and identification purposes. This article describes the implementation of an on-line neural classifier using a high-speed 32-bit floating-point digital signal processor (ADSP21062). The radiated noise from ships of four classes was received by an hidrophone, which was placed far from the ship, and digitized by an analog-to-digital converter (AD1847). Digitized signals were preprocessed for feeding the input nodes of a feedforward multilayer neural classifier. The preprocessing and the neural network were coded in Assembly language, optimized for maximum speed, and accesssible through an user-friendly interface, running in an IBM-PC that hosts the DSP board. Several tests for evaluating the accuracy and speed performance of the DSP implementation were performed, comparing the prototype with an off-line implementation developed using both MATLAB scripts and C and Fortran languages.
Código DOI: 10.21528/CBRN2001-084
Artigo em pdf: 5cbrn_084.pdf
Arquivo BibTex: 5cbrn_084.bib