Artificial neural networks in the improvement of spatial resolution of thermal infrared data for improved landuse classification

The spatial resolution of remotely sensed (RS) data in the thermal infrared (TIR) range is very coarse compared to the very fine resolutions in the visible (VIS) and near infrared (NIR) ranges. Despite, the information on emissive properties of TIR data that is complementary to the reflective properties of the VIS and NIR data, the application of TIR data has been rather restricted, mainly due to its coarse spatial resolution. Artificial neural networks (ANN) have proved to be far superior [Govindaraju, R. S. and Rao, A. R., 2000][Heermann, P. D. and Khazenei, K., 1992] to the statistical methods in many applications. Studies have been carried out on the applicability of ANN in the improvement of effective spatial resolution of Landsat-5, TM band 6 (TIR) daytime and nighttime data. The present paper reports the methodology developed and the results of the studies. The results are compared with those of a statistical approach.