A Quick Band-to-Band Mis-Registration Detection Method for Sentinel-2 MSI Images

A band-to-band mis-registration (BBMR) error often occurs in remote sensing (RS) images acquired by multi-spectral push broom spectrometers such as the Sentinel-2 Multi-spectral Instrument (MSI), leading to adverse impacts on reliability of further RS applications. Although systematic registration conducted during image production process corrects most BBMR errors, there are still quite a few being observed with discernible BBMR. Thus, quick detection method is needed assess quality online products. We here propose hybrid framework for detecting between visible bands MSI images. This comprises three main steps: first, candidate chips captured based Google Earth Engine (GEE) spatial analysis functions shrink valid areas inside scenes potential target chips. The redundant data pertaining local operation thus narrowed down. Second, spectral abnormal precisely extracted from every single chip, excluding influences clouds and water surfaces. Finally, matched pixel bands, best-fit coordinates then determined compare tolerance. Here, proposed was applied 71,493 Level-1C covering China its surrounding GEE platform. From these images, 4356 442 were detected inter-band offsets among bands. Further manual visual inspection revealed that had an accuracy 98.07% at chip scale 88.46% scene scale.