Infrared blind spectral deconvolution with low-rank sparse regularization for small object tracking

Abstract

Infrared spectral signal often exists the random noise and peak overlap problems, which is limited its widely applications. To address those issues, we proposed a novel blind spectral deconvolution model with low-rank sparse regularization (LSR) for small object tracking. Specifically, a new infrared spectral deconvolution (ISD) model integrating the rank of ground-truth spectral matrix is constructed to regularize the rank of degraded infrared spectral lines by L1 norm. In addition, a method based on alternating solutions between the latent spectral line and instrument function (INF) is used to optimize the ISD model. Finally, the experimental results are compared with other decomposition methods. The proposed LSR method not only recovers spectral structural details but also suppresses noise effectively. The restored infrared spectrum is more accurate in the application of IR spectral feature extraction and recognition.