关键词: LPC谱, 特征提取, 支持向量机, 神经网络, 目标识别

Abstract: Radiated noises of ships are very complicated. It is an imperative and difficult task to look for the new feature from ships noises in the classification of underwater targets. On the basis of the principle of the linear predictive coding(LPC), an weighting overlaps average LPC power spectrum density estimation algorithm is proposed. And an algorithm (One-Against-All: OAA) of multi-class Support Vector Machines (SVMs) is defined. Then the extracted feature vectors and SVMs、BP neural network were used to classify three different classes of targets. Results show that the statistical classification corrective rate exceeds 95.02%, the weighting overlaps average LPC spectrum has better discernment performance than the general LPC spectrum Feature, and the SVMs performance is better than that of BP neural network.

Key words: LPC Spectrum, Feature Extraction, Support Vector Machines, Neural Network, Target Recognition