CAJ | 학술논문

利用KL散度衡量增量非负矩阵分解效果，提高非负矩阵分解性能；施加行列式、稀疏性和相关性等约束条件，保证盲源信号分离的唯一性和性能；采用自然梯度下降法并选择合适的学习速率，得到源分离算法，该算法利用前一次分离结果和现在的输入信号矢量，迭代更新分离矩阵。仿真表明，KL-INMF盲源分离算法性能优于基于欧式距离INMF的盲源分离算法。
이용KL산도형량증량비부구진분해효과，제고비부구진분해성능；시가행렬식、희소성화상관성등약속조건，보증맹원신호분리적유일성화성능；채용자연제도하강법병선택합괄적학습속솔，득도원분리산법，해산법이용전일차분리결과화현재적수입신호시량，질대경신분리구진。방진표명，KL-INMF맹원분리산법성능우우기우구식거리INMF적맹원분리산법。
The Kullback-Leibler divergence was used to measure effects of incremental non-negative matrix factorization ( INMF ) in order to increase performance of NMF.The constraints of determinant, sparsity and correlations were imposed to ensure the unique and the performance of blind sources separation .A blind source separation algorithm(KL-INMF) was obtained by using a natural gradient descent method and selecting the appropriate learning rate.The algorithm iteratively updates the separation matrix through using the results of the last time separation and the current signals.Simulation results shown that the performance of KL-INMF blind source separation algorithm is better than that of the blind source separation algorithm based Euclidean distance INMF (ER-INMF).