Fusion Of Multi-Temporal And Multi-Sensor Hyperspectral Data For Land-Use Classification

A common application of hyperspectral imaging is land-use pixel classification. We have access to hyperspectral data from the same area acquired by different spatial and spectral resolution hyperspectral sensors at different timeslots. One may think that by only using data from the sensor with the highest both spatial and spectral resolution will be the best option. This paper shows that better results in the classification accuracy rate are achievable with redundant information of the same sensor taken in different timeslots and that data from lower resolution both spatial and spectral sensors could also improve the pixel classification accuracy. In addition, a band selection process over the entire set of bands have proven to provide better classification rates using a very small number of spectral bands.