On the influence of spectral calibration in hyperspectral image classification of leaves

Automatic detection of physiological changes in leaves using close range hyperspectral data is becoming a new tool for biologists. Given the geometry of leaves, the reliability of spectral data strongly depends on a careful spectral and geometric calibrations. In this paper, we evaluate the effect of several calibration approaches on automatic classification of leave regions. For our experiments we employ an in-vivo leaf scanning system, then an unsupervised classifier is applied on each calibrated and non-calibrated image and the biological relevance of the output is evaluated using vegetative indexes. Finally, we make recommendations about how to improve the hyperspectral image processing pipeline for this kind of data sets.