Automatic segment-wise restoration for wide irregular stripe noise in SDGSAT-1 multispectral data using side-slither data

Abstract

The raw orbital images captured by the new launched SDGSAT-1 Multispectral Image for Inshore (MII) are plagued by wide-irregular stripe noise, due to inconsistent unit response. This paper proposed a new method for destriping wide-irregular stripe noise in MII using the characteristics of side-slither data. Firstly, the raw side-slither data was standardized using line detection to guarantee that each row observed the same ground object. Then, the whole orbital side-slither image was segmented into blocks of equal length, and it was found that the response of wide-irregular stripe noise is consistent within a certain length. The Inverse Distance Weight was used to interpolate the DN values of striped pixels as referenced values, and the segmented length was determined by calculating Pearson correlation coefficient between the original and referenced DN values. Thirdly, the Random Sample Consensus (RANSAC) algorithm was used to find the inliers and calculate the correction parameters, after it was discovered that the original and referenced DN values had a linear correlation. The proposed method, SIR (consists of image segmentation, pixel interpolation and RANSAC fitting), can directly destripe the raw orbital image. One orbital side-slither data and six ordinary orbital data were selected for verification. Twelve state-of-the-art methods were chosen for comparison with SIR. The accuracy scores of SIR on three assessment indexes were higher than those of twelve other methods. The destriping outcomes for the images of city, cloud, forest, and river demonstrated the effectiveness of SIR in correcting wide-irregular stripe noise in MII images.