Super-resolution mapping of wetland inundation from remote sensing imagery based on integration of back-propagation neural network and genetic algorithm

Abstract

Mapping the spatio-temporal characteristics of wetland inundation has an important significance to the study of wetland environment and associated flora and fauna. High temporal remote sensing imagery is widely used for this purpose with the limitations of relatively low spatial resolutions. In this study, a novel method based on integration of back-propagation neural network (BP) and genetic algorithm (GA), so-called IBPGA, is proposed for super-resolution mapping of wetland inundation (SMWI) from multispectral remote sensing imagery. The IBPGA-SMWI algorithm is developed, including the fitness function and integration search strategy. IBPGA-SMWI was evaluated using Landsat TM/ETM + imagery from the Poyanghu wetland in China and the Macquarie Marshes in Australia. Compared with traditional SMWI methods, IBPGA-SMWI consistently achieved more accurate super-resolution mapping results in terms of visual and quantitative evaluations. In comparison with GA-SMWI, IBPGA-SMWI not only improved the accuracy of SMWI, but also accelerated the convergence speed of the algorithm. The sensitivity analysis of IBPGA-SMWI in relation to standard crossover rate, BP crossover rate and mutation rate was also carried out to discuss the algorithm performance. It is hoped that the results of this study will enhance the application of median-low resolution remote sensing imagery in wetland inundation mapping and monitoring, and ultimately support the studies of wetland environment.