CAJ | 학술논문

主成分分析是声场景分类中常用的特征选择方法。针对主成分分析的局限性，提出一种基于互信息的主成分分析方法。这一方法引入类别信息，用不同声场景条件下特征之间的互信息矩阵之和替代传统主成分分析中的协方差矩阵，计算其特征向量与特征值，特征向量表示由原始特征空间向新的主成分空间的转换系数，特征值则用于计算主成分的累计贡献率并判断主成分维数。声场景分类实验结果表明，该方法较之传统主成分分析方法降维效果更好，辅以神经网络分类器，计算得到的分类正确率更高。
주성분분석시성장경분류중상용적특정선택방법。침대주성분분석적국한성，제출일충기우호신식적주성분분석방법。저일방법인입유별신식，용불동성장경조건하특정지간적호신식구진지화체대전통주성분분석중적협방차구진，계산기특정향량여특정치，특정향량표시유원시특정공간향신적주성분공간적전환계수，특정치칙용우계산주성분적루계공헌솔병판단주성분유수。성장경분류실험결과표명，해방법교지전통주성분분석방법강유효과경호，보이신경망락분류기，계산득도적분류정학솔경고。
Principal Component Analysis (PCA) is a common method for feature selection. To improve selection, an algorithm of Principal Component Analysis based on Mutual Information (PCAMI) is presented. PCAMI introduces the category information, and uses the sum of mutual information matrices between features under different acoustic environments to replace the covariance matrix. The eigenvectors of the sum matrix represent the transformation coeffi-cients and the eigenvalues of the sum matrix are used to calculate the accumulative contribution rate to determine the dimensionality of the principal components. The experiments of acoustic environment classification show that relative to the traditional PCA, PCAMI has better performance in dimensionality reduction and in classification accuracy under the assistance of neuron network.