Remote Sensing Target Classification And Recognition Research Based On Sparse Representation

Target recognition is a hot research field of computer vision and artificial intelligence.It is widely used in face recognition,human-computer interaction,military,medicine and so on.In recent years,the sparse representation classification method obtained widespread attention by many researchers because of its good classification effect and better robustness.It has been successfully applied in practical problems of image processing and achieved good results.With the continuous development of remote sensing technology,remote sensing images are widely used in many fields,such as civil,military,and many other important aspects.The classification of remote sensing image is one of the key parts of remote sensing applications.In particular,the target detection and recognition in the field of military applications all need to classify the target images.Some scholars applied the sparse representation classification and recognition method to remote sensing images and achieved good recognition results.But remote sensing images tend to have problems such as poor visual contrast,low resolution and target with different angles of rotation,differences of different levels with the space structure and so on.So it is necessary to improve these problems related to sparse representation method.This paper studies and summarizes the research results of sparse representation classification methods by scholars at home and abroad.Two kinds of remote sensing target classification and recognition algorithms are put forward in this paper based on the characteristics of remote sensing images:1.A remote sensing target recognition method based on the extended dictionary and sparse representation classification is proposed in this paper.First,the training and test samples are enhanced with dyadic wavelet transform.Next,a feature dictionary is constituted by extracting SIFT features from the enhanced images.Then,an extended dictionary which contains an original training dictionary and a feature dictionary simultaneously for sparse representation is composed,so that the extended dictionary can be more discriminative,and the recognition rate can be higher.And the influence of SIFT features using random projection is also analyzed.Experiments show that the method is robust to the recognition of the remote sensing target.2.A remote sensing target recognition approach based on Gabor multi-scale adaptive weighting and sparse representation classification is proposed in this paper.First,Gabor wavelet transform is performed on the training samples and testing samples.The Gabor features in each direction are synthesized to be approximate isotropic.Then,fusion features are achieved by adaptive weighting summation and PCA dimensional reduction according to the characteristics contained in each scale.Thus a fusion feature dictionary is obtained replacing the original training dictionary so that the dictionary can be more discriminative,and the recognition rate can be higher.Experiments show that the method is robust to the recognition of the remote sensing target.This paper proposes two improved methods for the classification of sparse representation classification and recognition-The dictionaries’ discrimination ability is improved by expending or improving the dictionary thereby enhancing the sparse representation classifier’s capability and improving the recognition rate.Experiments in 10 classes of remote sensing images contained different rotation angles show that the two methods proposed in this paper have better robustness.