基于WSN的战场声目标多特征联合智能分类识别

本文利用WSN对战场声目标进行分类识别。由于战场上目标的声音信号构成非常复杂，采用单一特征很难全面反映其特点，提取多种特征来构成联合特征向量显得非常重要。然而简单的多特征联合识别，给识别带来维数灾难。本文提出，首先对滤波后的信号进行功率谱特征提取，再利用KPCA进行降维处理，将处理后的特征向量与最优小波包能量提取的特征向量组合，最后利用支持向量机完成对战场五类声目标的识别。仿真实验表明,采用这种方法能有效地提高分类识别的准确度。

The Acoustic Target in Battlefield Intelligent Classification and Identification with Multi-Features in WSN

The method of acoustic target classification in battlefield based on WSN was proposed. A single acoustic features can hardly reflect full characteristics of the target. Because target acoustic signals in battlefield, which consist of many different components, are very complex. Multiple features should be extracted to form jointed eigenvector. However identification with simple jointed features is a dimension disaster. In this paper, firstly, power spectrum density estimation was used to extract the feature of acoustic signals which was smoothed with wavelet packet. Then kernel principle component analysis was used to reduce the dimension of the feature. Secondly, the jointed eigenvector, which consisted with the reduced feature and the other feature extracted by best basis of wavelet packet, was obtained. Lastly, the multiple classifiers, which ware made up by support vector machine and error-correcting output codes, was used to identify the five types of acoustic target in battlefield. Simulation test shows that the accuracy of the target classification and identification is highly enhanced.