Research On Scene Classification Of Remote Sensing Image Based On Lie Group

With the science and technology continues to progress and development,kinds of remote sensing satellites such as optical,thermal infrared,and radar that can obtain remote sensing images can acquiring high-resolution remote sensing images.And the resolution of remote sensing images was increasing,the specifications continue to expand,and the surface information that can be recorded is also richer and more detailed.But the traditional remote sensing image classification method can not afford the processing needs of remote sensing images in increasingly complex scenes.For a remote sensing image data set,remote sensing image scene classification can automatically recognize different scenes according to human visual perception methods,such as airports,forests,ports,farmland,etc.,through deep semantic understanding of various image features.Remote sensing image scene classification,as an important branch of remote sensing image application field,has become a practical research hotspot in remote sensing image processing.Since the past two dacades,many researcher and scholars have carried out a lot of discussion and achievements on the scenes classification of remote sensing image,and proposed many classic remote sensing image scene classification methods,and achieved good results.However,due to the complexity of the remote sensing image acquisition process,the diversity between feature categories,and the variability between spatial layout structures,etc.,scene classification still has a lot of room for improvement.Based on the related research of Lie Group in mathematical theory,this paper proposes a research on scene classification of remote sensing images based on Lie Group kernel learning.The main research includes:(1)By studying the characteristics of various features and common feature extraction methods,comprehensively analyzes the advantages and disadvantages between them,summarize the previous research and application results of Lie Group,and propose a model to construct image scene semantic features based on Lie Group.Extract scene image features,reduce redundant and disturbing information in the remote sensing image,and construct a Lie Group covariance feature with high-level,representative.(2)Combined with the strong application ability of Lie Group in classification,the theory of Lie Group is introduced for scene classification of remote sensing image.Using the conversion relationship between Lie Group and Lie Algebras,the support vector machine algorithm is optimized based on the measurement form of the inner product space of Lie Groups and the principle of kernel functions,and kernel functions suitable for scene classification of remote sensing images are obtained.We design a scene classification model for remote sensing images based on Li Group kernel learning to improve the scene classification performance of remote sensing images.(3)The public remote sensing image data set of UC Merced Land Use and NWPU-RESISC45 were selected for experiment.By comparing the classification results of the proposed model and other methods in the classification of remote sensing image scenes,the effectiveness of Lie group kernel learning in the classification of remote sensing image scenes is explored.The experimental results show that the classification model proposed in this paper can achieve a classification accuracy of 94.21% on the data set UC Merced Land Use and 98.91% on the data set NWPURESISC45.Compared with the commonly used machine learning algorithms in recent years,the angle of the problem solved and the method used in this paper have certain reference significance for future generations to study Lie Group and remote sensing image scene classification.