Remote sensing image data fusion analysis method based on NMF algorithm

Abstract

With the continuous development of human production and life, single type remote sensors are constrained by the external environment or their own factors, and the remote sensing image data obtained can no longer meet current needs. The research aims to propose a fusion model to address this issue. A fusion model of the NMF algorithm based on the projection gradient optimization rule is proposed by combining the sensor observation model, non-negative matrix factorization (NMF, a method of decomposing a non-negative matrix into the product of two non-negative matrices) algorithm, and linear spectral hybrid model. Among them, endmembers (referring to the pure spectral components that make up the mixed pixels in hyperspectral unmixing) participate in the construction of linear spectral mixing models. Simulation data experiments show that the NMF fusion algorithm results in high spatial resolution and minimal distortion of spectral information. Compared with the least squares algorithm, its peak signal-to-noise ratio average is improved by 1.2–4.5 dB, the general image quality index average is improved by 0.0–0.01, the spectral value error mean difference range is 0.02–0.68, and the root mean square error mean difference range is 7.64–46.64, overall better. In the real data fusion experiment, the information entropy value range of the fusion result of this algorithm is 5.0–6.4, and the image clarity improvement value range is 3.6–6.4. The algorithm proposed in this study combines high spatial resolution and high spectral resolution in remote sensing image data, which is of great significance for the widespread application of remote sensing technology in various industries.