一种用于线性光谱解混算法验证的模拟数据生成方法

An Approach to Generate Synthetic Hyperspectral Data Used for Evaluating Linear Spectral Unmixing Algorithms

线性光谱解混已成为一种通用的光谱解混方法并已发展出大量算法,对这些算法进行客观评价,是该算法得到推广应用的重要基础。由于实测数据获取困难且花费很大,大多数研究中广泛采用模拟数据进行算法验证。针对目前常用的基于Dirichlet分布的模拟方法在实际应用中遇到的问题,结合实验分析了Dirichlet分布的参数设置对模拟结果的影响,进一步结合高光谱数据在其特征空间中的单形体几何特征,探讨了实际应用中参数取值受到的限制,并通过分析Dirichlet分布概率密度函数值的空间分布特征,提出参数的合适取值范围为(0,1.5)。

Linear spectral unmixing has been intensively studied for more than ten years, now it is a com- monly accepted approach to spectral unmixing, and a number of algorithms have been developed, but the objective evaluation of these algorithms is the foundation to the wide use of them. As the acquisition of reli- able ground-truth data is difficult and expensive, the use of real scenarios is limited, the simulated data have been used widely in the evaluation of these algorithms. The approach based on Dirichlet distribution is one of the common used methods for the simulation of hyperspectral data,it is more flexible and the Dirichlet density is suited to model abundance fractions. However,the parameters of the Dirichlet distribution affect the simulated results ,and improper parameters will cause the simulated data to lose the property that the hyperspectral data should preserve. This paper analyses the influence of the Dirichlet parameters on simula- ted results, and discusses the restrictions of the parameters value, and provides the recommended parame- ters range （0,1.5）.