Remote Sensing Image Scene Classification Based On Local Descriptor And Feature Learning

In recent years,remote sensing technology is developing rapidly and is widely applied in the many fields,and remote sensing images are used in large quantities.As the key technology of image processing,scene classification has attracted great attention in a wide range and is a hotspot in the field of remote sensing image processing.Through continuous efforts,the image scene classification has formed many methods of classic classification and algorithms.At present,the method of scene classification can be divided into two kinds: based on the underlying-features and based on the learning-features.The underlying-features are easy to obtain and the calculation is simple,but the classification effect has the upper limit.The middle semantic of learning-features can describe the characteristics of the scene more closely than the underlying features,but the process is complicated and the development is not yet mature.It also needs to improve the effect of the scene classification.Based on the characteristics of remote sensing images and the shortcomings of traditional methods of scene classification,this paper makes a deep study on the scene classification of remote sensing images,the main contents are as follows:(1)Scene classification of remote sensing images based on Completed Dual-Cross Patterns(CDCP).In view of the low resolution of remote sensing images and the high similarity of scene,we have made some improvements to LBP and proposed CDCP.On the one hand,the neighborhood changes from single-layer structure to double-layer structure,because the texture changes of double-layer structure can reflect the texture features around the central pixel better than the single-layer structure.On the other hand,we consider the coding of the absolute value of the gray scale difference,which is good for identifying some scenes with high similarity.(2)Scene classification of remote sensing images based on the learning-features.This method is mainly optimized for learning dictionaries and encoding features.We use online dictionary learning algorithm for dictionary learning,online dictionary learning algorithm to learn the dictionary by reconstructing the sample and sparse coding,the resulting dictionary is more expressive;we use locality-constrained Linear coding algorithm to encode features,Compared with sparse coding,locality-constrained Linear coding algorithm can not only achieve smaller reconstruction errors,but also has local smooth sparsity.In the experiment,we analyzed the experimental results and compared the classification accuracy's difference between the original method and the improved method.The experimental results show that the proposed method can achieve better classification results in this paper.