智能识别中的降维新方法及其应用研究.pdf

摘要 样本问题，通过引入张量理论，重新构造MCSVMs支持向量机的目标函数，从而提出 同时降低了算法本身具有的时间和空间复杂度。而且Ker-MCSVMsm硼戗方法首次实现了 矩阵模式的非线性化。第七章针对经典SVM方法不能充分地反映样本内在几何结构以 及所蕴含的判别信息的问题，通过同时引入线性判别准则中的类内散度和局部保持投影 LPP的基本原理，重新构造SVM的目标函数，提出基于全局和局部保持的半监督支持 本间内在的全局和局部结构，同时还满足作为半监督方法必须依据的一致性假设原则。 Abstract Abstract Inthefieldof reductionmethodsofare intelligentpatternrecognition，feature widely usedto rawdatainorderto theeffectivenessof preprocesshigh—dimensional improve extractionandfeature featurereduction selection，two intelligentrecognition．Feature primary methods．arein andhavebeen todealwithcertain successfully exploreddepth applied paUem someconcrete findthat recognitionproblems．Whiledealing谢th recognitionproblems，we currentclassicalfeaturereductionmethodsand of onesshow the partsimproved shortcomings oflowrobustnessandweak tocertain orderto generalizationability extent．Therefore，in their featurereductionmethodsare other ability,new proposedthroughintegrating improve into are asfollows． intelligentprocessingtechnologythem．Theypresented Partoneconsistsof 2and whichtwonovelfeatureselectionmethodsare chapter3，in discussed ordertosolvetheweak 2，in ability respectively．Inchapter generalizationproblem