Research On Super-resolution Mapping Based On Land Cover Distribution Patterns For Remote Sensing Imagery

Providing up-to-date accurate land cover information is the core of many resource management,planning and monitoring programs.Therefore,the land cover maps containing abundant land cover distribution features are widely used in agriculture,forestry,geology,oceanography and other fields of scientific research and practical application.Remote sensing images have become viable sources of effective land cover classification due to their macroscopic view and map format properties.However,restricted by the real object distribution patterns and the data acquisition equipment,mixed pixels are widely present in medium and low spatial resolution remote sensing images,posing a huge challenge to land cover mapping.Traditional soft classification technology analyze the mixed pixels of remote sensing images quantitatively by estimating the proportion of different land cover in each mixed pixel,failing to determine the spatial distribution of various kinds of land cover at the sub-pixel level.Super-resolution mapping technology achieve obtaining land cover distribution information of remote sensing images at the sub-pixel level by segmenting the mixed pixels into smaller units,effectively making up for the lack of soft classification techniques and having important research value in the field of remote sensing image processing.In this paper,the super-resolution mapping methods of remote sensing images have been analyzed and studied according to the different land cover distribution patterns.The specific work is arranged as follows:Firstly,the research status of SRM method is analyzed.Several commonly used SRM algorithms are studied and analyzed,whose effects are verified and compared through two experiments.Aiming at the problems existing in these super-resolution mapping methods,the SRM research based on two different patterns of land cover distribution is proposed.For the simple land cover distribution patterns,the object size is usually larger than pixel resolution,and the object boundary is regular and clear relatively.The traditional spatialdependence-based SRM methods can complete the land cover mapping of such patterns efficiently.However,the existing algorithms ignore the directionality of the spatial dependence.To solve this problem,the SPSAM with anisotropic spatial dependence model(SPSAMA),which takes full account of the directionality of spatial attraction among the boundary pixels,is proposed in this paper.The proposed algorithm not only maintains the high efficiency of SPSAM algorithm,but also optimizes the land cover mapping results of the object border and further improves the super-resolution mapping accuracy.As to the complex land cover distribution patterns,the size of the object is smaller than that of mixed pixel generally,whose object spatial structure is more complicated and the boundary shape is more fragmented.In this case,relying on the spatial dependence solely can hardly locate the sub-pixel The learning-based SRM methods have been developed to be valid solutions to these patterns,nevertheless there are some limitations.For example,high-resolution remote sensing images serving as auxiliary or training data,should cover the same area as the input image or have similar land cover distribution patterns with the input image,which are not readily available in practice.Dealing with these problems,this paper proposes a new SRM algorithm based on joint dictionary sparse representation(JDSR).According to the transfer learning mechanism,the two dictionaries for the low-and high-resolution image patches jointly training from natural images with sparse representation are applied to the super-resolution mapping of remote sensing imagery.JDSR optimize the sub-pixel sharpening process and achieve the goal of improving the super-resolution mapping accuracy of the complex land cover distribution patterns in the absence of auxiliary remote sensing data.Experimental results show that the JDSR algorithm is effective in complex land cover distribution.