基于Quick Shift算法的高光谱影像分类

Hyperspectral Imagery Classification Based on Quick Shift

论述了面向对象分类方法处理高光谱高空间分辨率影像的优势与流程;分析了快速漂移(Quick Shift)算法的原理,该算法在进行模式搜索时具有可控制模态选择和平衡"过分割"与"欠分割"的特点。将该算法应用于高光谱影像分割,可得到面向对象分类所需的较理想的"同质"影像对象。为提高影像分割的效率,提出了一种基于灰度共生矩阵的自适应核带宽确定方法,能够兼顾影像空间特征和光谱特征。最后采用最小距离分类法、支持向量机分类法与提出的分类方法进行了对比试验,实验结果表明了该方法的有效性。

The advantages and process of classifying hyperspectral imagery with high spatial resolution via object-oriented classification were expounded.The theory of Quick Shift was analyzed,which can control the mode selection and balance under-and over-fragmentation in mode seeking.The algorithm was applied to segment hyperspectral imagery to obtain the ideal homogeneity objects used in object-oriented classification.In order to improve the segment efficiency,an adaptive kernel bandwidth selection method based on grey-level co-occurrence matrix was proposed,the method can calculate the spatial features and spectrum features simultaneously.Lastly,the comparative experiments were performed via minimum distance classification,SVM,and the proposed method,the results proved the validity of the proposed method.