基于RFS和ART的高光谱图像主成分提取方法

Hyperspectral Image Principle Component Extraction Method Based on RFS and ART

针对传统高光谱图像主成分提取方法受数据分布状态和噪声影响大的缺点,提出基于区域特征光谱的ART(Adaptive Resonance Theory)神经网络主成分提取算法.首先通过多方向阈值空间邻域聚类提取区域特征光谱作为ART的输入模式,利用ART网络的自适应特性获取地物光谱矢量特征,并通过对光谱矢量聚类完成图像的主成分提取.对高光谱图像仿真结果表明:通过提取区域特征光谱,神经网络的数据处理量减少了约97%;算法能够较准确地提取图像主成分且提取效果明显好于K-均值算法.

Algorithms used to extract principle components of hyperspectral image are sensitive to noise and data distribution.A principle components extracting algorithm based on the region feature spectrum（RFS） and ART is presented.The algorithm firstly extracts region feature spectrum through spatial neighborhood clustering as input pattern vectors of the network,and then acquires the classificatory character adaptively.Finally,extraction is successfully achieved by using clustering spectral vectors.The experiments on hyperspectral images indicate that the size of data processed by network is reduced about 97%,and the extraction effect is obviously better than that by K-means algorithm.