Abstract: In recent years, the research work on mining and reconstruction of high-dimensional, heterogeneous, and massive remote sensing image data using matrix decomposition methods has been reported continuously. In this paper, we review the cutting edge of matrix decomposition application in both domestic and international studies since 2000, focusing primarily on the progress of matrix decomposition, its key improvements and application optimizations; the research trends in data reconstruction and feature mining methods based on matrix decomposition; and the challenges faced by matrix decomposition methods in the field of remote sensing image reconstruction. The results indicate that over the past 20 years, the related research has shown a gradual increase in volume, with a shift in focus from traditional decomposition methods to hyperspectral image analysis, image reconstruction, low-rank efficiency decomposition methods, and sparse feature mining. The proportion of research on sparse matrix decomposition, probabilistic matrix decomposition, and non-negative matrix decomposition increased year by year, with the most active application research in the fields of geography, remote sensing, and ecology. The non-negative matrix decomposition method exhibits strong capabilities in image feature mining and is anticipated to achieve complete reconstruction of surface information from satellite remote sensing images at low-rank levels. Utilizing matrix decomposition methods to extract features and reconstruct image data from remote sensing images demonstrates both feasibility and significant potential for further expansion.