一种改进的加权L_1正则化稀疏光谱解混算法

An improved sparse spectral unmixing algorithm based on weighted L_1 regularization

针对普通的加权L1正则化方法在进行稀疏光谱解混时,对空间信息利用不足的问题,提出了一种基于修正权值的L1范数正则化稀疏光谱解混方法。在加权L1优化求解过程中,根据当前的解以及空间信息得到下一步迭代的权值,能更好地利用混合像元丰度系数的稀疏性。实验结果表明,在较低的信噪比时,基于权值修正的加权L1正则化稀疏光谱解混比传统的迭代加权L1正则化的方法精度高。

Aiming at the shortcoming of spatial information exploitation in hyperspectral sparse unmixing based on normal reweighted L1 regularization, a method of sparse unmixing based on corrected reweighted L1 regularization is proposed. In the optimization process of weighted L1 , the method uses the value of current solution and spatial in- formation to get the weights for next iteration, which makes the sparsity of fractional abundances of mixed pixel be represented better. Experimental results indicate that the accuracy of sparse unmixing based on corrected reweight- ed L1 is higher than that of traditional reweighted L1 regularization for low signal-to-noise ratio hyperspectral images.