Land Cover Classification Of Remote Sensing In The Central Area Of Yunnan Province Based On Super-resolution Reconstruction

Super-resolution reconstruction technology is to break through the limitations of remote sensing hardware imaging conditions.It can obtain higher resolution than the known resolution of remote sensing images,which is an important means to improve the resolution of remote sensing images,and also a hot topic in field of digital image processing.Land cover classification is an important research work in remote sensing application field,especially large-scale regional land cover dynamic monitoring is the basis of scientific research on global change and sustainable development,and is also an important indicator component of International Geosphere Biosphere Program(short for IGBP)plan.MODIS is undoubtedly the preferred data source for large area classification.MODIS has the advantages of larger imaging area,rich spectral information,short return period,low cost and easy access,but the resolution is low.Therefore,it is of great significance to explore how to improve the resolution of MODIS images and along with its advantages to carry out a large area of land cover remote sensing classification.Take the MODIS in Central Yunnan Province as the research object.MAP,IBP,POCS,NUI and other reconstruction methods are used to carry out the super-resolution reconstruction.The reconstructed MODIS images are used to classify the land cover in Central Yunnan Province.The main contents and results are as follows:(1)Experimental study on super resolution reconstruction algorithm of remote sensing image.By studying the basic concepts of image super resolution reconstruction,different classification,commonly used research methods and research status at home and abroad,a preliminary experimental study on super resolution reconstruction technology is carried out.Study the mechanism of image degradation,motion estimation,fuzzy function identification,noisy degradation model estimation of the image.The images are reconstructed by sub-pixel registration,inverse filtering and convolution.Then,four algorithms are used to initialize the image super resolution reconstruction.The MODIS 250 m,OLI 30 m and 15 m resolution remote sensing images were used to evaluate the reconstructed images.The subjective and objective evaluation of the reconstructed image is carried out,including the standard deviation,average gradient,spatial frequency and information entropy.The results show that the MAP method of subjective contrast texture is clear and delicate,spectral characteristics are obvious,the objective evaluation index is the best.The other three reconstruction effects are IBP,NUI and POCS.(2)Application of MAP method to reconstruct remote sensing images in Central Yunnan.Based on the image motion blur and the optical fuzzy condition,the image degradation model is simulated and the model parameters suitable for the data in this area are worked out.Finally,reconstruct a super-resolution image in central Yunnan,and combined the subjective and objective and MTF curves to show the advantages of reconstructed images.(3)Land cover classification based on supervised classification using reconstructed images.Using the Support Vector Machine classifier to classify the reconstructed and original MODIS images,combined with land cover data collected in the field and remote sensing image of 2 m resolution.And develop training samples and test samples for classification and evaluation.The overall classification accuracy of the reconstructed image of the confused matrix is 74.82%,the Kappa coefficient is 0.7103,the overall classification accuracy of the original image is 70.64%,and the Kappa coefficient is 0.6552.For the MODIS image classification,the evaluation results show that the classification quality of the original and reconstructed data is very good,and the classifier and the training sample are crossed.Compared with the original image,the reconstructed image is improved by 4.18 percentage points,and the Kappa coefficient is improved by 5.51 percentage points.It is proved that the reconstructed image quality is improved from another angle and the initial application of reconstructed image is realized.The super resolution reconstruction of MODIS is realized and the land cover classification realized by combining with the actual situation in central Yunnan province.In the practical application,the super-resolution reconstruction stage,the motion estimation and the fuzzy function identification can't be precise position,resulting in the reconstruction image quality is not perfect,classification accuracy is not much better.Therefore,a more accurate quality estimation model is the focus of the research work of super resolution reconstruction.Changes in surface features and cloud interference are also major factors affecting the quality of reconstructed images.This work is also the focus of future research.