Improving spatial resolution of LANDSAT spectral bands from a single RGB image using artificial neural network

Spectral information provided by multispectral and hyperspectral sensors has a great impact on remote sensing studies. These sensors are embedded in aircrafts and satellites like the Landsat, which has more data freely available but lack the spatial resolution that suborbital sensors have. To increase the spatial resolution, a series of techniques have been developed like pansharpenning data fusion and more advanced convolutional neural networks for super-resolution, however, the later requires large datasets. To overcome this requirement, this work aims to increase the spatial resolution of Landsat spectral bands using artificial neural networks that uses pixel kernels of a single high-resolution image from Google Earth. Using this method, the high-resolution spectral bands were generated with pixel size of 1m in contrast to the 15m of pansharpenned Landsat bands. The evaluate the predicted spectral bands the validation measures Universal Quality Index (UQI) and Spectral Angle Mapper (SAM) were used, showing values of 0.98 and 0.16 respectively, presenting good results.