Remote Sensing Image Target Recognition Based On Sparse Representation

Since artificial intelligence has become a national strategy,both the emerging Internet and traditional industries have invested a lot of manpower and financial resources to conduct in-depth research.Among them,target recognition is the most widely used technology branch in artificial intelligence,and it is involved in enterprise applications,military science and technology,and medical technology.In recent years,the target identification method based on sparse theory shows a better recognition effect and robustness,which makes it become a hot topic for many scholars.Many scholars have applied it to practical problems such as image processing and achieved very good results.With the rapid development of remote sensing technology in the 20th century,as a field observation technology capable of providing real-time multi-temporal,multi-spectral,and wide-range information,it has been widely used in various industries,such as civil and military applications.There is no lack of its presence.Accurate robust classification and recognition of remote sensing images has become a hot topic and challenge in the field of target recognition research.Some scholars have used sparse representation classification recognition methods to classify and recognize remote sensing images and found that this method has a good recognition effect.However,in remote sensing images,the diversity of objects,the rich information contained in the image leads to poor visual contrast,and the image objects have different rotation angles.Therefore,when using traditional sparse representation theory to classify remote sensing image targets,it is necessary to consider Remote sensing image features and algorithm improvement.This paper studies and summarizes the research results of sparse representation classification and recognition methods by scholars at home and abroad.According to the characteristics of remote sensing images,the following two improved algorithms are proposed:1.An algorithm for classification and recognition of remote sensing images based on Sift features and joint sparse model is proposed.Firstly,the image enhancement of binary training wavelet transform is performed on the training image and the test image,which reduces the influence of the visual contrast difference of the remote sensing image to some extent.Then the Sift feature of the enhanced image is extracted as a joint sparse input sample,so that the public features and private features are more adapted to the rotation characteristics of the remote sensing image,and the recognition rate is improved.In addition,this paper also analyzes the recognition rate of the algorithm under different dimensions after random projection.Experimental results show that the proposed algorithm has a better effect on the recognition rate of remote sensing images.2.A method of remote sensing target recognition combined with adaptive weighted feature dictionary and joint sparse is proposed.First,the Gabor wavelet transform is performed on both the training set and the test set,and the Gabor features in each direction are adaptively weighted and summed.Then using the joint sparse theory to extract the public features and private features as the joint feature dictionary proxy original dictionary,so that the atoms in the dictionary have a stronger ability to discriminate between samples,and improve the recognition rate of the classification.Finally,the experimental results show that this algorithm has significantly improved the recognition rate.Based on the sparse representation classification and recognition methods,this paper proposes two improved methods for the characteristics of remote sensing images.Through the improvement of feature vectors and dictionaries,the sparse reconstruction vectors can be improved in their discriminative ability and sparse representation classifiers can be classified.Stronger,improve recognition rate.Experiments are carried out on the experimental data set composed of target images of 10 types of remote sensing aircraft.The results show that both of the proposed methods have better recognition rate and robustness.