Dictionary Construction Method for Hyperspectral Remote Sensing Correlation Imaging

Abstract

The correlation imaging technique is a novel imaging strategy which acquires the object image by the correlation reconstruction algorithm from the separated signal light field and reference light field, with the advantages such as super-resolution, anti-interference and high security. The hyperspectral remote sensing correlation imaging technique combined the correlation imaging and hyperspectral remote sensing has the ability to detect the spectral properties of ground objects more than the above advantages. The spatial and spectral images can be reconstructed from very few measurements acquired by hyperspectral correlation imaging systems via sparsity constraint, making it an effective approach to solve the process and transition problem in high spatial and spectral resolution. While in the hyperspectral remote sensing correlation imaging, due to the complexity and variance of the spatial and spectral properties of the target scene and lack of prior knowledge, it is difficult to construct an effective dictionary for the hyperspectral reconstruction. The fixed dictionaries such as DCT (Discrete Cosine Transform) dictionary and wavelet dictionary are mainly used in the reconstruction up to present, and these dictionaries contain fixed and limited characteristics, which is hard to present different hyperspectral scenes efficiently. This paper aims at the problem that in the hyperspectral remote sensing correlation imaging system the sparse representation of complex ground objects is difficult in image reconstruction, resulting in low quality of reconstructed images. By combining the sparse coding and dictionary learning theory of signal processing, a related research on the construction method of hyperspectral remote sensing sparse dictionaries is carried out. Through the construction of hyperspectral remote sensing sparse dictionaries, optimization in reconstruction, and application research in actual imaging systems, a set of sparse dictionary construction and usage methods in hyperspectral correlation imaging has been formed. Comparing with current methods such as the total variation constraints and the hyperspectral image kernel norm constraints, the hyperspectral remote sensing scene sparsity and hyperspectral image reconstruction quality have been effectively improved by the proposed method, which has guiding significance for the development of the correlation imaging field.