基于高维云模型和RBF神经网络的遥感影像不确定性分类方法

云模型是用自然语言值表示的某个定性概念与其定量表示之间的不确定性转换模型,RBF神经网络已经广泛应用于遥感影像分类。考虑到传统的RBF神经网络分类技术不能有效表达影像分类过程中存在的不确定性、难以自适应地确定隐含层神经元,本文提出了一个基于高维云模型和改进RBF神经网络的不确定性分类技术。利用高维正态云创建隐含层神经元,使RBF神经网络能充分表达影像分类过程中存在的不确定性。通过峰值法云变换和高维云算法自适应地确定最优隐含层神经元。通过基于概率的权值确定和频率阈值调整,进一步优化RBF神经网络的结构。实验表明,本文提出的方法有较高的分类精度,分类结果基本上与人眼目视解译一致。

Uncertainty classification method of remote sensing image based on high-dimensional cloud model and RBF neural network

Cloud model is an uncertainty conversion model between qualitative concept described by natural language and its quantitative expression.The RBF neural network has been applied widely to remote sensing image classification.Considering the traditional RBF neural network classification technique couldn’t effectively express uncertainty existing in image classification,and couldn’t determine adaptively hidden layer neurons,this paper proposed an uncertainty classification technique based on high-dimension cloud model and improved RBF neural network.Firstly,by using high-dimensional normal cloud models to construct hidden layer neurons,RBF neural network could fully express the uncertainty existing in image classification.Then,by using peak-based cloud transform and high-dimensional cloud algorithm,the optimal neurons of hidden layer were adaptively determined.Finally,by using probability-based weight determination and frequency threshold adjustment,the RBF neural network was further optimized.The experiments showed that the proposed method had higher classification accuracy and could produce good classification results which were consistent with visual interpretation of the human eye.