Detection of Saturation in High-Resolution Pushbroom Satellite Imagery

Over the last decade, DigitalGlobe has launched a series of commercial Earth imaging satellites. These high-resolution satellite imageries provide an increasingly abundant data source for remotemapping of the Earth surface and its temporal variability. Among the factors affecting image quality is saturation of the charge-coupled device due to improper setting of the time delay integration level for the imaged surface, which results in along-track striping over areas of high radiance. We present and demonstrate an algorithm for the local detection of saturation striping by a wavelet transform, used to detect periodic variations of brightness (i.e., striping) with varying frequencies at different locations, combined with the use of unidirectional brightness gradients. The algorithm is applicable to raw, orthorectified, and resampled imagery. We test the algorithm using panchromatic images acquired by the GeoEye-1 and WorldView 1-3 sensors over polar regions. Saturation area classification masks generated by the algorithm agree well with manually identified areas of saturation. Manual validation of the algorithm applied to over 6000 images in Iceland reveals a high (>80%) success rate when the saturation levels are 2% or higher. Our general methodology may be widely applicable to periodic noise detection in imagery.