Hyperspectral Compression Based on Tucker Decomposition

Tucker decomposition stands out as an effective method for hyperspectral data compression, with the core tensor playing a pivotal role in determining its quality . Consequently, the selection of an appropriate core tensor is a focal point in the realm of data compression. This study conducted three experiments, examining spatial size, waveband number, and comprehensive analyses to discern the optimal core tensor. Our findings reveal a significant relationship between the size of the core tensor decomposed by Tucker and the data compression rate. Notably, the spatial size of the core tensor was found to be inversely proportional to the compression rate. Specifically, a core tensor size constituting 2/5 of the data’s spatial dimensions and a waveband number comprising 3/10 of the data were identified as optimal. This configuration demonstrated the ability to effectively preserve both spatial and spectral information while maintaining a commendable compression rate. In conclusion, selecting a core tensor size of 2/5 of the data’s spatial size and 3/10 of the data’s waveband number emerged as a favorable choice, striking a balance between information retention and compression efficiency in the context of hyperspectral data compression using Tucker decomposition.