应用盲源分离方法分离岩矿混合像元

Application of Blind Source Separation Method in Decomposing Mixed Pixel

采用盲源分离方法解混岩矿混合像元,获取岩矿组分信息.分析了常用的Fast ICA、Robust ICA方法,从算法稳定性、分离信号质量和迭代计算效率3个方面,比较不同目标函数及寻优过程的优势和不足:Robust ICA在算法稳定性和迭代计算效率上具有较大优势,分离信号质量并不是最佳;Fast ICA对初始值和步长比较敏感,计算可能不收敛,也可能陷于局部最优;峭度为目标函数的Fast ICA有较好的分离信号质量,但算法稳定性不如负熵为目标函数的Fast ICA;用于负熵近似的非二次函数对算法稳定性和迭代计算效率有较大影响,原因是非二次函数影响迭代计算步长,较小的步长算法稳定性较好,但是迭代计算效率降低.实际运用中,应根据岩矿混合像元光谱特点,选择恰当的混合像元分离方法,在不同性能之间达到平衡.

Blind Source Separation( BSS) methods can be used in rock mineral decomposition to obtain the component information. Fast ICA and Robust ICA are two widely used BSS methods,but their performances are affected by objective function and searching process. The study compares different variants of the two methods from the perspective of stability,efficiency and quality. Robust ICA outperforms the rest in terms of stability and efficiency of algorithm,but the quality of decomposed signals is not the best. Fast ICA is sensitive to initialize solution and step size,while the searching process might be non-convergent or trapped in partial optimum. When Fast ICA takes kurtosis as objective function,it improves the quality of decomposed signals,but its stability of algorithm is inferior to Fast ICA with negentropy as objective function.Furthermore,as the non-quadratic functions used for negentropy approximation affect iterative step size,they have significant influence on the algorithm stability and efficiency. Smaller step size leads to better stability,while reduces efficiency. In practice,the BSS methods should be chosen based on spectral characteristics of mixed rock-mineral pixel to achieve a balance of stability,efficiency and quality.