论文部分内容阅读
由于对多类问题的高维数据无法直接观察其聚类和分布特性，本文采用神网络法实现自适应主元特征提取（ＡＰＥＸ），以压缩特征空间的维数，并保持足够的信息来鉴别事物之间的类别。它可有效地提取信号的主要特征和抑制噪声。我们将高维数据压缩影射到２或３维，从而实现特征数据的可视性分析，显示物体对象间的类似程度和关系结构。并采用高阶函数的神经网络对其进行非线性分类，同时与ＢＰ网络的非线性分类能力进行了实验比较。结果表明高阶函数神经网络较ＢＰ网络分类能力强，训练速度快。
Due to the inability to directly observe the clustering and distribution characteristics of high-dimensional data of many kinds of problems, this paper adopts the neural network method to realize adaptive principal component feature extraction (APEX) to compress the dimension of feature space and keep enough information to identify Categories between things. It can effectively extract the main features of the signal and suppress noise. We map high-dimensional data compression to 2 or 3 dimensions to enable visibility analysis of the feature data, showing similarities and relationships between object objects. The neural network of high-order function is used to classify it non-linearly. At the same time, the nonlinear classification ability of BP neural network is compared experimentally. The results show that higher-order function neural networks have better classification ability and faster training than BP networks.