光谱谐波分析的新型HAC非监督分类器

A New HAC Unsupervised Classifier Based on Spectral Harmonic Analysis

高光谱影像分类是识别影像信息的重要途径之一,研究其算法对地物识别、动态变化监测和专题信息提取等方面具有重要意义。非监督分类由于其具有无须先验知识的特点,被广泛应用于高光谱影像分类。结合谐波分析理论提出一种新的高光谱影像非监督分类算法,即谐波分析分类器(harmonic analysis classifier,HAC)。首先,该算法统计第一谐波分量并绘制其直方图,根据波峰数目及位置确定初始地物类别和聚类中心像元。然后将待分类像元光谱的波形信息映射到谐波分解次数、振幅和相位的特征空间中,利用同类地物在特征空间中表现聚集性这一特征,根据最小距离原则对待分类像元进行归类。最后,计算聚类中心像元间的欧式距离,通过设置距离阈值完成类间合并,从而达到高光谱影像分类的目的。提取两种地物类别的光谱曲线,经谐波分析后得到谐波分解次数、振幅和相位量,并分析其在特征空间中的分布情况验证了HAC算法的正确性。同时将HAC算法应用到EO-1卫星的Hyperion高光谱影像得到其分类结果,通过对比K-MEANS,ISODATA和HAC算法的高光谱影像分类结果,证实HAC算法作为一种非监督分类方法在高光谱影像分类方面具有较好的应用性。

Hyperspectral images classification is one of the important methods to identify image information,which has great significance for feature identification,dynamic monitoring and thematic information extraction,etc.Unsupervised classification without prior knowledge is widely used in hyperspectral image classification.This article proposes a new hyperspectral images unsupervised classification algorithm based on harmonic analysis（HA）,which is called the harmonic analysis classifer（HAC）.First,the HAC algorithm counts the first harmonic component and draws the histogram,so it can determine the initial feature categories and the pixel of cluster centers according to the number and location of the peak.Then,the algorithm is to map the waveform information of pixels to be classified spectrum into the feature space made up of harmonic decomposition times,amplitude and phase,and the similar features can be gotten together in the feature space,these pixels will be classified according to the principle of minimum distance.Finally,the algorithm computes the Euclidean distance of these pixels between cluster center,and merges the initial classification by setting the distance threshold.so the HAC can achieve the purpose of hyperspectral images classification.The paper collects spectral curves of two feature categories,and obtains harmonic decomposition times,amplitude and phase after harmonic analysis,the distribution of HA components in the feature space verified the correctness of the HAC.While the HAC algorithm is applied to EO-1satellite Hyperion hyperspectral image and obtains the results of classification.Comparing with the hyperspectral image classifying results of K-MEANS,ISODATA and HAC classifiers,the HAC,as a unsupervised classification method,is confirmed to have better application on hyperspectral image classification.