高分类率水下目标特征量提取方法研究

根据人眼分类双谱图时的特点，在双谱对称性所确定的三角形区域内提出了两种双谱幅值矩阵元素抽取方案，即沿平行于副对角线方向的抽取方案和沿列向量方向的抽取方案。对所抽取的元素，采用简单的求和或最大值操作进行特征提取，形成4种特征向量。利用支持向量机的一对一多分类方法进行目标分类，实验表明：在由双谱对称性确定的三角形区域内，采用沿平行于副对角线方向的元素幅值抽取方案，对所抽取元素幅值使用求和方法得到的特征向量具有非常高的正确分类率。由此方法获得的特征向量对于A、B、C三类水下目标辐射噪声的分类率达到了100%，得到的其他特征向量的平均分类正确率均稳定在95%以上。

Research on feature extraction algorithm of high classification accuracy for underwater targets

According to the characteristics of human eyes＇classifying bispectra,two extraction schemes for bispectra magnitudes matrix elements are proposed in a triangle area defined by bispectrum symmetry.They are extraction scheme along the direction parallel to the vice diagonal and that along the direction of column vector,respectively.And 4 feature vectors are obtained by simple summation operation or maximum operation to the extracted elements.By using One-against-One（OAO） method of multi-classification of Support Vector Machine（SVM）,it is validated that very high mean classification accuracy can be obtained by summation operation to the first extraction scheme.The classification accuracy of the feature vector obtained by the above method are 100% for the radiated noise of underwater targets in three types A,B and C.And the mean classification accuracy for the other feature vectors proposed by this paper are steadily above 95%.