论文部分内容阅读
剖析了用神经网络实现特征主元提取（ＰＣＥ）、自组织特征影射（ＳＯＦＭ）、类扩展自组织语义影射（ＳＯＳＭ）和改进的特征细化自组织影射．通过对运载工具的特征压缩，进行可视性分析，结果表明ＰＣＥ和ＳＯＦＭ都能显示事物间的类似程度和关系结构，具有语义影射的功能．特征细化的ＳＯＦＭ同样能达到类扩展ＳＯＳＭ细化分类的功能，它克服了类扩展的ＳＯＳＭ增加输入特征的维数、增加不必要的计算量、输入特征与影射结果不相一致的缺点．
This paper analyzes the feature principal component extraction (PCE), self-organizing feature projection (SOFM), class-extended self-organizing semantic shadowing (SOSM) and improved feature refinement self-organizing projection using neural networks. Through the compression of vehicle characteristics, the visibility analysis shows that both PCE and SOFM can show the similarities and relationships between things and have the function of semantic mapping. The feature refinement of SOFM can also achieve the function of class extension SOSM refinement classification. It overcomes the disadvantages of class extension SOSM such as increasing the dimensionality of input features, increasing unnecessary computational complexity, and inconsistent input features and projection results.