Goal-Driven Phenotyping Through Spectral Imaging for Grape Aromatic Ripeness Assessment

In this paper, we describe a systematic approach to the design of an active spectral imaging system for in vivo phenotyping. Our approach takes into account two major factors: spectral sensitivity of the sensor and spectral composition of the illuminant. Similarly to previous works, we adopt a scheme consisting on dimensionality reduction and SVR regression of target chemical parameters from spectral datacubes. We find that high prediction accuracies may be achieved for different sets of parameters depending on the illuminant. Furthermore, in most cases the combination of a single monochromatic illuminant with a dichromatic image sensor (passband and stopband) suffices, which paves the way for the design of tailored low cost imagers. Besides, we demonstrate in vivo estimation of aromatically relevant compounds of white and red grape varieties, not addressed before to our knowledge.