Research On Manifold Learning Method For Feature Extraction Of Visual Target

With the development of science and technology,computer vision has become one of the research hotspots.Visual target feature extraction methods are an important field in computer vision,and visual target feature extraction has been widely used in real life,such as target extraction,target detection and target tracking.This paper proposed a Laplacian Eigenmaps(LE)regularized semi-supervised target feature extraction algorithm(SSRLE)based on Robust Principal Component Analysis(RPCA),which is based on the ideas that the prior supervision information can improve the accuracy of target extraction and the target after separating the background can improve the classification accuracy,to use the relationship between target extraction and semi-supervised classification problem.On the basis of ensuring the global structure of the data,this paper adds the LE regularization of the weight matrix of the adaptive neighborhood graph to ensure the local structure of the data,and it can overcome the problem that the traditional LE algorithm is affected by the neighbor value k.Under the action of the prior information,the target and background can be separated well.The method use target data and supervision information to train linear classifiers and manifold smoothing hypothesis to achieve prediction of unlabeled data and achieve better classification results.Considering that the shortcoming of ordinary graphs that can only represent the relationship between two data points,which will seriously damage the complex structure information between data points,it is necessary to extend simple graphs to hypergraph learning.This paper proposed a semi-supervised target feature extraction method based on hypergraph and SSRLE.It expands the Laplacian matrix to the hypergraph Laplacian matrix,which can guarantee the complex structure information between data points through the hypergraph,and constructs the feature-feature interaction matrix to connect the complex relationship between the data features.Meanwhile the paper improves the accuracy of the algorithm in target extraction and classification problems.This paper verifies the effectiveness of the proposed algorithm from target extraction and classification.It also compares the effectiveness of different ratios of supervision information on target extraction and the effect of different semi-supervised algorithms on classification.Because of the semi-supervised target feature extraction method based on the hypergraph has a strong dependence on the value,this paper finally analyzes the parameters k.