Investigating Intensity and Transversal Drift in Hyperspectral Imaging Data

Abstract

When measuring data with hyperspectral cameras drift in the data distribution occurs over time and when the sensing device is changed. Frequently, this drift is a combination of intensity and wavelength shifts. In this contribution, we demonstrate that transfer component analysis together with subsampling constitutes a particular efficient and simple technology for spectral offset elimination which is applied to avoid the negative impact of drift on the classification performance. We demonstrate that this approach performs on par or better in comparison to established methods, and we also provide a theoretical motivation why this technology can deal with both, intensity as well as wavelength shift provided bounds on the smoothness of the functional data are given.

From hyperspectral to multispectral sensing – from simulation to reality: A comprehensive approach for calibration model transfer

“…Furthermore, LEDs and other electronic sensor components are sensitive to temperature change. This causes intensity and transversal drift in our spectra, something treated in [6]. Robust methods against drifts are presented in [6] and briefly covered in Section 3.…”

“…This causes intensity and transversal drift in our spectra, something treated in [6]. Robust methods against drifts are presented in [6] and briefly covered in Section 3.…”

“…This simplification necessitates standardizing each data set componentwise. We ultimately subtract the mean spectrum for each sensor in every sample [6].…”

“…The model trained with coefficients of Chebyshev polynomials, representing a latent space, rather than with spectra. This method's suitability for dealing with shifts has been demonstrated in [6]. In our study, the data in the latent space were z-transformed to normalize the large differences of the Chebyshev coefficients.…”

“…This paper uses a specially designed benchmark application to analyze the precision of these three sensor levels, focusing on different approaches to transferring the calibration model from simulation (2) to reality (3) by using only a subset of labeled samples, as presented in Figure 1. Standard mathematical methods [4], and the authors' experience with model transfer in the context of interoperability [5,6] are used to do this.…”