Developing universal models for the prediction of physical quality in citrus fruits analysed on-tree using portable NIRS sensors

Abstract

The citrus sector seeks rapid, economical, environmentally-friendly and non-destructive technologies for monitoring external and internal changes in physical quality taking place in fruit during on-tree growth, thus allowing fruit quality to be evaluated at any stage of fruit development. The use of portable near-infrared spectroscopy (NIRS) sensors based on micro-electro-mechanical system (MEMS) technology, in conjunction with chemometric data treatment models, has already been studied for quality-control purposes in two citrus species: oranges and mandarins. The critical challenge is to develop robust and accurate universal models based on hundreds of highly heterogeneous citrus samples in order to design quality prediction models applicable to all fruits belonging to the genus Citrus, rather than models that can only be applied successfully to a single citrus species. This study evaluated and compared the performance of Modified Partial Least Squares (MPLS) and LOCAL regression algorithms for the prediction of major physical-quality parameters in all citrus fruits. Results showed that, while models developed using both linear (MPLS) and non-linear regression techniques (LOCAL) yielded promising results for the on-tree quality evaluation of citrus fruits, the LOCAL algorithm additionally increased the predictive capacity of models constructed for all the main parameters tested. These findings confirm that NIRS technology, used in conjunction with large databases and local regression strategies, increases the robustness of models for the on-tree prediction of citrus fruit quality; this will undoubtedly be of benefit to the citrus industry.