基于径向基函数神经网络的混合像元分解

Mixed Image Cell Decomposition Based on Radial-basis Function Neural Networks

遥感图像中普通存在着混合像元 ,对这部分像元进行分类 (即混合像元分解 )是遥感图像处理中的难点。基于主分量分析的混合像元分解算法是一种较为成熟的算法 ,但它存在着计算量大 ,适应性差等缺点。在深入研究混合像元分解原理的基础上 ,提出了用径向基函数神经网络拟合分解结果超平面 ,以实现混合像元分解的算法 ,实验结果证明 :该算法的结果与基于主分量分析的混合像元分解算法结果相近 (相关系数达到 0 99) ,而计算量大大减少 ,具有较强的适应性。

Remote sensing images contain a lot of mixed image cells, and it is difficult to classify these cells. Mixed image cells decomposition algorithm based on principle component analysis is a widely used algorithm, but the large computation amount and less flexibility are its main drawbacks. By researching the curve fitting (approximation)theory of the radial basis function neural networks, and the principles of the mixed image cells decomposition algorithm based on principle component analysis, the paper proposes a new decomposition algorithm, which uses the radial basis function neural networks to fit (approximate) the hyperplane of the decomposition results of the principle component analysis algorithm. Experimental results prove that the results of the new algorithm are almost the same with the results of the principle component analysis algorithm (correlation coefficients are above 0.99). However, the new algorithm has much less computation complexity and more flexibility than the principle component analysis algorithm.