An accurate and fast neural method for PCA extraction

Principal component analysis (PCA) is a characteristic extraction method, whose main objective function is the reconstruction of the original data space. PCA is a linear optimal method, in the sense of mean squared error, and is applied in a wide variety of knowledge areas. In this paper, a new neural method for PCA extraction is proposed and compared, in terms of accuracy and computational costs, to other well accepted neural extraction methods, such as GHA and APEX. The performance comparison was evaluated using preprocessed spectra from passive sonar signals. It was verified that the proposed method performed better than all other methods, exhibiting easier implementation, lower computational costs and higher accuracy.