A Hybrid Approach to Super-Resolution Mapping of Remotely Sensed Multi-spectral Satellite Images Using Genetic Algorithm and Hopfield Neural Network

Though super-resolution mapping of multi-spectral remote sensing satellite images is used to locate the multiple classes within a mixed pixel, it has limitations, for example, low rate of convergence leading to low accuracy and higher time consumption. This paper demonstrates a hybrid approach of genetic algorithm and Hopfield neural network which can not only speed up the process but also identify the exact global optimal solution using super-resolution mapping. The experiments are carried out for Landsat ETM+image of different dimensions (10x10, 25x25, 64x64, 128x128 and 256x256) and with map sizes of 2, 3, 4, 5 and 6. Map size indicates the number of all sub-pixels which are resolved from a pixel. The overall accuracy is 89.44%, 90.25%, 91.09%, 92.56%, and 93.62%, respectively, for map sizes 2, 3, 4, 5, and 6, thus showing an increase of nearly 2% accuracy for hybrid genetic algorithm. The time taken was also reduced by half of that for the genetic algorithm. Thus, the efficiency of this novel approach to map land cover classes in a coarse pixel with greater accuracy and appreciably lesser time has been demonstrated.