A statistical technique for noise identification and restoration of hyperspectral image cube

Abstract

Hyperion, onboard EO-1 satellite is a hyperspectral sensor that scans the earth in 242 contiguous bands with 10 nm spectral resolution. The data has numerous applications in geology, agriculture, water resources, and many other fields. However, due to the miscalibration of detectors, the Hyperion dataset suffers from a few inherent errors like bad bands, bad columns, stripes, etc. which degrade the quality of the image. Thus, detection of these errors and their correction is an essential step before any scientific interpretation. The available literature suggests manual and semi-automatic techniques for error removal, which are time-consuming, complex, and not efficient. In the present study, a statistical technique was developed to detect the errors and correct the stripping artifacts, which is simple, less time-consuming, and fully automatic, thereby reducing processing time and human errors which might be induced if manually processed.