Ming Nie1, Xiaobing Dai2,*
1 Electric Power Research Institute of Guangdong Power Grid Co., Ltd., Guangzhou 510080, China
2 Wuhan University, Wuhan 430068, China
Corresponding Author: Xiaobing Dai
Abstract: Spectra play an increasingly important role in object detection, recognition, and identification. A key challenge in object identification using infrared spectroscopy is that spectral profiles change with variations in temperature and pressure, making robust feature extraction difficult. Thus, a stable descriptor for infrared spectra is essential for invariant feature extraction under varying environmental conditions. In this paper, we propose a novel spectral descriptor grounded in the concept of curvature. The fundamental insight is that the relative curvature across different scales remains unchanged despite environmental perturbations. The Curvature Scale Space (CSS) method is adopted as the foundation, upon which we develop the Normalized Curvature Scale Space (NCSS) descriptor for invariant spectral representation.
Keywords: Gas spectral recognition; Invariant feature extraction; Curvature Scale Space; Spectral descriptor
References
Innovation Series is an academic publisher publishing journals and books covering a wide range of academic disciplines.
Francesc Boix i Campo, 7 08038 Barcelona, Spain