基于RBF神经网络的图像融合复原方法研究

RBF Neural Network Based Image Fusion Restoration Approach

提出了一种基于径向基函数(RBF)神经网络的多通道图像数据融合复原方法,研究了该方法在多光谱图像复原上的应用.将软竞争学习策略和自适应调整隐节点相结合对网络进行优化训练.利用多光谱卫星图像数据,对所提出的方法进行仿真实验.实验结果表明该融合复原方法提高了复原图像的质量;改进后的学习算法能够保证学习准确度和较短的训练时间;实验还表明RBF神经网络的多通道复原和单通道复原、传统的维纳滤波及最大后验概率方法相比,在改善图像像质上具有明显的优越性.

A radial basis function （RBF）neural network based image fusion restoration approach for multiple spectrum images is proposed and investigated. The spatial resolution improvement of images with poor resolution can be achieved by fusing the images of various resolutions. The simulated experimental results show the effectiveness of the method. The image quality is improved by fusing the correlative and redundant information between images. The modified algorithms ensure quick training and mapping precision of the networks. The research also illustrates that using multi-channel information is more effective than using a single one and the scenario conducted has advantages over conventional methods.