基于Kohonen网络的小波阈值在遥感图像去噪中的应用研究

传统Donoho非线性小波阈值去噪方法中,阈值的选取一般是根据噪声方差设置的,但在实际的操作中很难对噪声方差的精确值,这种理论上的局限性使得各种基于方差估计的去噪方法并不能获得实际需要的去噪结果;针对阈值选取的问题,提出一种非线性小波变换阈值的kohonen神经网络的自调整学习训练方法对遥感图像进行去噪,该方法在小波变换的基础上结合了神经网络的非线性阈值自组织特征映射算法,阈值的选择根据训练图像进行学习;可以实现自调整寻找最优值,以满足实际,达到最优的去噪效果。

Study and Application of Wavelet Threshold Based on Kohonen Neural Network in Remote Sensing Image De-noising

In the de-noising of traditional non-linear wavelet threshold,we often select threshold by variance of noise,but that is difficultly obtained in practice,the disadvantage of theory cause that all kinds of the method based-on estimate of variance can't satisfy practical demand.Aim at threshold's choice,the paper propose new de-noising method of non-linear wavelet threshold based on Kohonen neural network in remote sensing image,which combines character mapping of threshold self-organize and neural network,chooses the best threshold according to the images,and reaches the best effect of de-noising.