提出一种聚类优化的快速独立成分分析(FastICA)解混算法,解决了FastICA在光谱信息解混过程中,对解混矩阵初始值敏感导致解混信息不稳定的问题。利用模糊C均值聚类对单一颜料光谱信息进行光谱特征降维,选择最具代表性的聚类结果作为解混矩阵初始值,通过FastICA牛顿迭代公式计算聚类优化后的解混矩阵,避免随机选取初始值对混合颜料光谱信息解混的影响。实验结果表明,与其他算法相比,解混结果平均误差值降低了0.57,平均适应度系数达到了99.67%,光谱角度匹配距离降低了0.53,所提算法增加了FastICA算法解混结果的稳定性,提高了混合颜料光谱信息解混精度。

A clustering-optimized fast independent component analysis (FastICA) de-mixing algorithm is proposed. It solves the problem of unstable de-mixing information caused by the sensitivity to the initial value of the de-mixing matrix for FastICA algorithm during the spectral information de-mixing process. Fuzzy C-means clustering algorithm is used to reduce the spectral characteristics of single pigment spectral information, the most representative clustering result is selected as the initial value of the de-mixing matrix, and the clustering-optimized de-mixing matrix is calculated by FastICA Newton iteration formula to avoid the effect of randomly selecting initial values on de-mixing the spectral information of mixed pigments. The experimental results show that, compared with other algorithms, the average error value of the unmixed results of this algorithm is reduced by 0.57, the average fitness coefficient is 99.67%, and the spectral angle matching distance is reduced by 0.53. The proposed method can increase the stability of the FastICA de-mixing results, and improve the de-mixing precision of the mixed pigment spectral information.