空间权重自适应的MRF高光谱图像模糊聚类方法

An Adaptive Spatial Weight Markov Random Field Fuzzy Clustering for Hyperspectral Images

传统模糊聚类方法以像元光谱信息为基础,通过相似性准则在特征空间内进行自动聚集。高光谱图像聚类过程往往受到混合像元和“同物异谱”现象的影响,造成结果噪声和破碎严重,导致算法难以适应于高光谱图像地物识别。针对传统聚类算法的不足,考虑邻域像元间相关性和连续性即上下文特征,文章提出了一种新的基于空间权重自适应马尔科夫随机场模型(markov random field,MRF)的高光谱图像模糊聚类算法,在模糊C-均值聚类目标函数中引入空间项,并采用自适应权重系数控制其在聚类中的影响程度,将空间信息自适应地引入聚类过程中。通过模拟及真实高光谱数据实验证明,较仅使用光谱及分类后处理滤波算法,该算法有效提高了高光谱图像聚类的精度和抗噪能力。

Conventional pixelwise clustering methods mainly take advantage of spectral features while ignoring spatial relationship with neighboring pixels.The accuracy of the clustering is affected by the“salt and pepper”effect caused by the mixed pixels and spectral variety.Addressing the problem,an adaptive spatial weight Markov random field fuzzy clustering algorithm is proposed when considering the contextual information between the pixels and its neighbors.A spatial weight coefficient is used to control the degree of the spatial contribution in the objection function of the fuzzy C-means.Experimental results of a synthetic hyperspectral data set and a real hyperspectral image demonstrate that the clustering accuracy and the anti-noise ability for hyperspectral images are improved when comparing with the spectral clustering and post clustering methods.