Abstract in English:

In recent years, hyperspectral imaging (HSI) has emerged to become a crucial method for both remote sensing and close-range surface analysis. In this paper, we present substantial upgrades of our previously published system for multispectral and hyperspectral surface analysis (Hegemann et al., 2017). Besides minor changes in illumination, we carefully evaluated different approaches for reflectance correction using up to eight calibration standards. Wavelength correction, which ensures an exact wavelength fit, is also done using a calibration standard. Therefore, our calibration pipeline provides high-quality hyperspectral data that is mostly independent of the hardware acquiring it, as we remove the impact of illumination and sensor sensitivity and consequently solely dependent on the sample. Additionally, as the main contribution, we present a method to acquire hyperspectral images from a non-planar surface using spectrometers without a time-consuming auto- focus at every pixel posi tion. We do this by generating a registered depth map from gray value images of the sample. Since annotated hyperspectral data is in high demand, we also contribute two initial pixel-wise labeled close-range hyperspectral datasets generated with our upgraded system for further research and benchmarks.