Super-resolution land cover mapping by deep learning

Feng Ling,Giles M Foody

doi:10.1080/2150704x.2019.1587196

Abstract

ABSTRACTSuper-resolution mapping (SRM) is a technique to estimate a fine spatial resolution land cover map from coarse spatial resolution fractional proportion images. SRM is often based explicitly on the use of a spatial pattern model that represents the land cover mosaic at the fine spatial resolution. Recently developed deep learning methods have considerable potential as an alternative approach for SRM, based on learning the spatial pattern of land cover from existing fine resolution data such as land cover maps. This letter proposes a deep learning-based SRM algorithm (DeepSRM). A deep convolutional neural network was first trained to estimate a fine resolution indicator image for each class from the coarse resolution fractional image, and all indicator maps were then combined to create the final fine resolution land cover map based on the maximal value strategy. The results of an experiment undertaken with simulated images show that DeepSRM was superior to conventional hard classification and a suite of popular SRM algorithms, yielding the most accurate land cover representation. Consequently, methods such as DeepSRM may help exploit the potential of remote sensing as a source of accurate land cover information.

Highlights

The mixed pixel problem has long been recognised as a major constraint to land cover mapping from remotely sensed imagery, especially if acquired at a coarse spatial resolution and/or for the mapping of highly fragmented landscapes
Super-resolution land cover mapping (SRM) provides a further major enhancement by locating the class fractional components predicted by a soft classification geographically in the area represented by mixed pixels (Atkinson 2009; Foody et al 2005; Ge et al 2014; Ling et al 2010)
These spatial dependence models have been widely used in SRM, they can be oversimplified and may be inadequate for the representation of complex land cover mosaics such as those found in highly fragmented landscapes (Ling et al 2016) and the quality of the final map is highly influenced by the suitability of the specific model used (Muad and Foody, 2012)

Summary

Introduction

The mixed pixel problem has long been recognised as a major constraint to land cover mapping from remotely sensed imagery, especially if acquired at a coarse spatial resolution and/or for the mapping of highly fragmented landscapes. The spatial dependence can be calculated at the sub-pixel scale (Atkinson 2005), the sub-pixel/pixel scale (Ling et al 2013; Mertens et al 2006) and at multiple scales (Ling et al 2014; Chen et al 2018) These spatial dependence models have been widely used in SRM, they can be oversimplified and may be inadequate for the representation of complex land cover mosaics such as those found in highly fragmented landscapes (Ling et al 2016) and the quality of the final map is highly influenced by the suitability of the specific model used (Muad and Foody, 2012)

Methods

Results

Conclusion