Research And Application On Land-Use Classification Of Remote Sensing Image By Sparse Representation Based On Multi-Feature Modeling

Accurately extracting land cover information is one of the important bases for the rational use and development of land resources,and the application of remote sensing technology to land cover classification has many advantages.When many mature remote sensing classifiers and implementation strategies are classified in the ecologically fragile areas of desert grasslands,their advantages are limited.Because the ecologically fragile areas of desert grasslands have a variety of surface cover types,significant edge effects,weak anti-jamming capability and other characteristics.In order to improve the accuracy of land use classification under the premise of a small amount of labeled samples in the ecologically fragile areas of desert grasslands,this paper selects Landsat remote sensing data and takes Luoshan area of Ningxia as an example to conduct a preliminary discussion on the classification and dynamic analysis of land use/coverage.The related content and conclusions are as follows:(1)In order to improve the accuracy of object classification in the premise of a paucity of labeled samples,this paper selected Landsat8 OLI(Operationa Land Imager)images from 2013 to 2015 as the data source,and selected three classification methods including maximum likelihood classifier,support vector machine and sparse classifier based on dictionary learning.We only use spectral features to compare the accuracy of different classifiers.The results show that the sparse classification based on dictionary learning has the highest classification accuracy for the original spectral features compared with the MLC and SVM,and has excellent learning effect and performance,especially in the case of a small number of labeled samples.(2)In view of the richness of land use types in the ecologically fragile areas of desert grasslands,we used a total of 15-dimensional features including spectrum,vegetation,terrain,architecture and water information instead of using one feature alone or directly using the original spectral features.In order to improve the classification accuracy,we combined the spectral features with the vegetation,terrain,architecture and water information features respectively.Using the sparse classifier based on dictionary learning,we compared the effect of different feature combinations under the small sample training set.The research results show that the higher the feature dimension is,the higher accuracy is not always.The final best feature combination is b1~b7,NDVI,DEM,NDBI+VAR(B5)and MNDWI.There is a 12-dimensional feature combination.(3)Using the best feature combination and the best classifier selected above,we tookLandsat5 TM and Landsat8 OLI images from 2001 to 2015 as data sources,and extracted the land cover information of the study area from 2001,2006,2010 and 2015 in the premise of a paucity of labeled samples.Then we analyzed the changes of land cover during the 15 years in the study area from the quantitative statistics of land cover types,the transfer matrix and the spatial variation.It is shown that: 1)From 2001 to 2006,due to the construction of Hongsibao irrigation project and ecological resettlement project,the area of irrigated land increased significantly,the area of dry land decreased,and artificial land decreased,which mainly converted into dry land and wild grassland.2)From 2006 to 2010,due to the key period of the ecological immigration project in the arid zone of central Ningxia,the area of dry land increased,the area of artificial land continued to decrease,the wild grassland increased,the bare land decreased.So,the land utilization began to be rationalized.3)From 2010 to 2015,with the mature development of ecological migration projects and ecological construction projects,the area of grassland,wetland and other areas has increased,the area of cultivated land has continued to grow,and the area of man-made land was decreased significantly.Land use is more and more intensive,the ecological environment has been improved,and the harmony between man and land has become higher and higher.