Estimation of Missing Data in Remote Sensing Images Using t-SVD based tensor completion

This paper focuses on estimating the lost information from hyperspectral remote sensing images affected by missing samples. During acquisition or transmission remote sensing images may suffer from missing information. When it is affected by missing information, it cannot be used for further applications such as object detection and classification. Therefore the need to predict the missing data in existing remote sensing images becomes a necessity. The performance of the existing tensor completion techniques are poor when data is missing in all bands of a hyperspectral remote sensing image because they do not adequately utilise the spectral information in all bands. This paper proposes a new technique to overcome these problems by using Tensor Singular Value Decomposition (t-SVD) based Tensor Completion (TC). The spectral as well as spatial information in all bands of a hyperspectral remote sensing image could be utilized to detect all missing data, wherefore existing tensor completion techniques are a failure. The proposed method have been experimentally proved to outperform the existing methods when data is missing in either all of the bands or some of the bands.