Land Cover Classification In Cloudy And Hilly Regions Based On Optical And SAR Data

Land cover classification has fundamental and critical significance for Land Resources Survey and Assessment and Global Changes.One of most important and basic works of these researches is to acquire land use information accurately and efficiently,that is to say,how to get land cover information quickly and accurately with multi-source remote sensing data.But in Southwest China(Chongqing,Sichuan,Guizhou),as the representative of cloudy and hilly regions,where the subtropical humid monsoon climate is dominant,the air humidity is high and having more foggy day,cloud cover above 20% of Landsat OLI images in a year accounts for more than 90%.These brings great difficulties and challenges to the classification of land cover in mountain areas.Therefore,in this paper,the area we researched is situated in parts of Chongqing.First,based on two different clouds of Landsat OLI data,combined with NDVI,texture,elevation and slope information,we used object-oriented method,determined the membership function and established the classification rules to classify land cover.Then,because the imaging way of microwave data is different from optical data,it's not affected by clouds and fogs,we attempted to use fully polarimetric ALOS-2 image,after a series of processing,including polarization filtering,terrain correction and polarimetric decomposition,we used the RGB composition image of pauli,supplemented by texture,elevation and slope information,established the corresponding classification rules,also classified the land cover by means of object-oriented approach.Finally,we tried to fuse the Landsat OLI and ALOS-2 data,and based on the fusion data,classified land cover by objectoriented and nearest pixel method.The main conclusions of the study are as follows:(1)Different clouds of optical data causing different classification results,the overall accuracy is 58.09% and the kappa coefficient is 0.4913 for Landsat OLI data of 26.28% clouds cover;the overall accuracy is 82.74% and the kappa coefficient is 0.791 for Landsat OLI data of 8.43% clouds cover.The data classification accuracy of less cloud data is about 30% higher than more cloud.It shows that the cloud amount has a great influence on the accuracy of the optical image classification.So in the cloudy mountains,other data should be searched for the classification replaced for optical data.(2)Synthetic aperture radar can transmit energy at microwave frequencies,it's unaffected by weather conditions,we based on the Range-Doppler location model,ortho-rectified ALOS-2 images and normalized the backscatter coefficients,maximally eliminated the effect of terrain.After many times of debugging,we determined the segmentation scale and established the corresponding algorithm of ground object classification for the ALOS-2 data,and the classification results show that,the overall accuracy is 83.80% and the kappa coefficient is 0.8039.(3)In order to extracted land cover information better,we fused Landsat OLI and ALOS-2 data by multiplicative method after continuous attempt.For Landsat OLI and ALOS-2 fusion data,the object-oriented nearest neighbor classification method was used to classify the image.The results show that the overall accuracy of the classification is 86.97% and the kappa coefficient is 0.8447,which achieves a great classification accuracy.(4)For the overall classification accuracy,Landsat OLI and ALOS-2 fusion data is the highest,Landsat OLI data(cloud amount 8.43%)and ALOS-2 data are equivalent,Landsat OLI data(cloud cover 26.28%)is the lowest.For the classification accuracy of various land cover types,the classification precision of Landsat OLI and ALOS-2 fusion data is higher for woodland,farmland,shrub and grassland,garden,water body and artificial building.Landsat OLI data only has a comparative advantage on the extraction of shrub and grassland,ALOS-2 data has the relative advantages of woodland,water and artificial buildings extraction.(5)The better options for the classification of land cover in the cloudy and hilly areas are as follows: during the low cloud and fog period,Landsat OLI data with simple acquisition methods and easy data processing should be selected for classification.Of course,Landsat OLI and ALOS-2 fusion data can be used for land information extraction.when high accuracy is required or need to be re-classified for a first-grade land category.For periods of cloudy,there is no available optical remote sensing data,and fully polarized ALOS-2 data is well substitutable.