Contactless and non-destructive chlorophyll content prediction by random forest regression: A case study on fresh-cut rocket leaves

In green leafy vegetables, the retention of green colour is one of the most generally used index to evaluate the overall quality and freshness and it is associated to total chlorophyll content. Destructive chemical techniques and non-destructive chlorophyll meters represent the state-of-the-art methods to accomplish such critical task. The former are effective and robust but also expensive and time consuming. The latter are cheaper and faster but exhibit lower reliability, require the probe to touch the leaves and heavily depend on the positions chosen for sampling the leaf's surface. In this paper, a new approach to non-destructively predict total chlorophyll content of fresh-cut rocket leaves without contact is proposed. Fresh-cut rocket leaves were analysed for total chlorophyll content by spectrophotometer and SPAD-502 (used as reference values) and acquired by a computer vision system using a machinelearning model (Random Forest Regression) to predict total chlorophyll content. Finally, the trained and validated model will be used for on-line prediction of total chlorophyll content of unseen freshcut rocket leaves. The proposed system can match the physical and timing constraints of a real industrial production line and its performance (R2 = 0.90), measured on the case study of fresh-cut rocket leaves, outperformed the results of the SPAD chlorophyll meter (R2 = 0.79).