The algorithm research of low-rank matrix spectral reconstruction for ground targets

Recently, micro-spectrometer based on filter array has received extensive attention in terms of cost and size. Yet, the spectrometer will produce large noise in the work, which has a great impact on the spectral reconstruction. In this paper, a low-dimensional filter array is selected based on the K-means-PSO(Particle Swarm Optimization) method to achieve the purpose of data dimensionality reduction, which further reduces the cost and processing difficulty of the micro-spectrometer. To address the redundancy and poor accuracy of spectral reconstruction data obtained by micro-spectrometers, a convex optimization algorithm constrained by three-segment regularization of a low-rank-matrix (IReg-Cvx algorithm) was proposed for spectral reconstruction in this study. In order to test algorithm universality and stability better, we selected 120 kinds of ground spectral curves, and the low-dimensional filter array is fused with the IReg-Cvx algorithm. Apply the corresponding constraints according to the different slopes of the curve, and the high-quality spectral reconstruction of the ground object target spectrum can be stably realized under the noise environment of 30, 25, and 20 dB.