Close-range hyperspectral imaging (HSI) of plants is now a potential tool for non-destructive extraction of plant functional traits. A major motivation is the plant phenotyping related applications where different plant genotypes are explored for different environmental conditions. HSI of Arabidopsis thaliana is of particular importance as it is a model organism in plant biology. In the present work, a portable HSI setup has been used for the monitoring of a set of 6 Arabidopsis thaliana plants. The plants were monitored under controlled watering conditions where 3 plants were watered as normal and the other 3 plants were given 50% of the normal volume of water. The images were pre-processed utilising the standard normal variate (SNV) and changes over time were evaluated using unsupervised clustering over the time series. The results showed an early detection of stress from day 4 onwards compared to the commonly used normalised difference vegetation index (NDVI), which provided detection from day 9.

Use of hyperspectral imaging (HSI) for automated characterisation of plants in a high-throughput plant phenotyping setup (HTPPS) is a challenging task. A challenge arises when the same plant is being monitored automatically during the experiment as it might not be in the same orientation as it was imaged last time. Such changes in orientation result in variations in illumination, which affects the signals recorded by the HSI setup. In addition, there are challenges with the use of threshold-based segmentation approaches such as normalised difference vegetation index (NDVI) for distinguishing between old and dead leaves, which might be observed in the later stages of experiments, from the soil background. Therefore, the potential of spectral normalisation for homogenising HS images and the use of supervised spectral set for plant segmentation is presented. Further, the effects of testing chemicals on plants were visualised using PCA of the HS images.