一种改进的双因子自适应FastICA算法

为解决快速独立分量分析算法（FastICA）对初值权值敏感的问题，提出了一种双收敛因子FastICA算法（Double Convergence Factor FastICA，DCF—FastICA）。该算法利用两个不同步长因子的FastICA算法进行组合，并通过梯度算法自适应调节两分离权值矩阵的组合系数，从而得到最优权值分离矩阵。理论分析与实验结果表明DCF—FastICA算法较之以往改进算法具有明显的优势，该算法不仅改善了初值权值敏感问题，而且可在几乎不损失分离精度的情况下，使平均分离速度较原算法提高近70%，迭代次数较原算法减少近80%。

An Improved Double Factor Adaptive FastICA Algorithm

A novel algorithm called double convergence factors FastICA (DCF-FastICA) was proposed to solve the problem that the FastICA algorithm is sensitive to the initial weights.Two FastICA algorithms with different step size factors were combined in this method,and the combination coefficient was adjusted using the gradient algorithm until the optimal separation matrix was obtained.Theoretical analysis and experimental simulation showed that the proposed algorithm can produce better separation result compared with the previous improved algorithms,the problem of initial weights sensitivity could be resolved with almost no loss of separation precision,the average separation speed is improved nearly 70% and the number of iterations reduced nearly 80% compared with the original FastICA algorithm.