Feature Extraction Model Of Multi-constraint Deep Subspace Sparse Optimization

Image processing and pattern recognition are playing important roles in researching field of leading technology.The achieved result has been widely applied in artificial intelligence,military science and technology,human lives and so on.As an significant part of image recognition,feature extraction has crucial influence on recognition rate and operation time.This year,the algorithm of deep learning,as a new feature extraction method,is superior to traditional feature extraction method in semantic description of image features.Based on the deep structure of the deep learning model,this paper studies the deep feature extraction model of the deep subspace model by combining the domestic and abroad relevant research results.First of all,it is necessary to improve the robustness of the model.To achieve it,the relevant theory of sparse representation is incorporated into the deep subspace model,and the feature extraction model is constrained by the sparse optimization method.In order to solve the noise interference in the image,the error matrix is constructed.Then,the optimal solution is made to the sparse characteristic matrix and the error matrix in each subspace spatial feature mapping.The noise interference is separated from the sparse features to keep the feature extraction sparse.Secondly,in order to solve the problem of distance measurement in the deep subspace,the self-learning theory of distance metric is proposed.In addition,the optimal distance metric mapping matrix is determined by solving the maximum internal class distance in deep subspace.The deep subspace has the characteristics of higher feature dimension and more complicated data distribution than the classical single subspace.According to the measure learning operation,the distinction information between the samples can be maximum kept during the mapping process of each subspace spatial feature.That is the way of how to obtain the corresponding optimal distance metric in the specific subspace.Under this metric,the internal class distance of the sample feature should be as far as possible and the distance within the class should be as small as possible to contribute the follow-up identification classification.Last but not least,the inherent symmetry of the face image and some object images proposes to apply the symmetry constraint to the deep subspace feature extraction model,which can improve the feature abstraction ability of the model for the symmetric object image.The face image has a typical symmetry which brings convenience for the recognition.In the internal layer feature mapping,applying the symmetry metric matrix in the process of internal layer feature mapping and adding the symmetry constraint can effectively explore the special position relation between organs on the face.It not only can make the feature extraction result reasonable and improved but also can facilitate the follow-up identification.