Research On Feature Extraction Algorithm Based On Manifold Learning

With the advancement and development of information technology,people will inevitably face a variety of high-dimensional data.How to deal with these data is a major challenge for researchers in pattern recognition and data mining.The traditional linear dimensionality reduction algorithm has very low applicability or even failure when dealing with high-dimensional data,which has attracted the attention of researchers at home and abroad.Manifold learning began to enter people's field of vision.Since it was proposed,it has aroused the enthusiastic response from the academic community.Its original idea of dimensionality reduction has become a research hotspot.Research results based on manifold learning algorithms are continuously applied and developed in various fields.In this paper,based on the improvement of the performance of feature extraction algorithms,based on the analysis of the basic principles and characteristics of manifold learning algorithms,some in-depth studies on manifold learning algorithms are carried out.The main contents and work done in this paper are as follows:1.In order to better satisfy the assumption of local linearity,this paper analyzes the characteristics of the Local Linear Embedding algorithm and proposes a local linear embedding algorithm based on the local tangent space mapping theory.This algorithm is to determine the K nearest neighbor after the points on the domain are mapped to the tangent space of the point,and the point is linearly reconstructed in the tangent space.The performance of the local linear embedding algorithm was verified by running on various data sets.2.The classic Isometric Mapping algorithm needs to determine the best K value by trial and error,but the equidistant mapping algorithm has high complexity,and multiple experiments will make the algorithm's efficiency in practical applications lower.This paper presents an equal distance mapping algorithm that can automatically select the neighborhood.Combining the principal component analysis method to perform the best field selection for each point in the sample and then performing the equidistant mapping,which greatly facilitates the use of the algorithm.