论文部分内容阅读
局部线性嵌入算法(LLE)能很好保存数据点的局部性质,因此有很好的数据可视化效果,但它不是一种很好的面向分类的特征提取方法。因为它存在样本外点学习能力差和忽略了样本类别信息的缺点。对此,本文提出一种分类型局部线性嵌入算法。所提方法通过计算重构误差来判定样本类别,并引进平移向量和缩放因子对距离修正,显著提高类别可分性。在对高光谱影像进行分类的试验中验证了该方法的有效性。
Local linear embedding algorithm (LLE) can preserve the local properties of data points very well, so it has good data visualization effect, but it is not a very good classification-oriented feature extraction method. Because of the shortcomings of poor sample learning ability and neglect of sample type information. In this regard, this paper presents a classification of local linear embedding algorithm. The proposed method determines the sample category by calculating the reconstruction error, and introduces the translation vector and the scaling factor to correct the distance, which improves the separability of the category significantly. The validity of this method is verified in the experiment of classifying hyperspectral images.