Development of a linear mixed model to predict the picking time in strawberry harvesting processes

In manual fruit and vegetable harvesting, picking time statistics can be used to improve labour management and optimise the design and operation of harvest-aiding machines, such as conventional cross-row conveyors or recently proposed robotic transport carts. In this study, a dataset of 161 picking times from 18 workers was collected in commercial strawberry fields in Salinas, California, and a set of conditional linear mixed models (LMMs) was formulated to model the amount of time (“picking time”) required by a picker to fill an empty tray with harvested crop. The LMMs were based on different combinations of the following influencing factors: picker speed, time of day, plant spacing, and picking cart style. The significance of effects of these factors was investigated and the LMMs were compared with each other using cross-validation (CV) techniques. The LMMs were also evaluated using a new dataset collected during the next year's harvest season. The best predictive LMM was found to be a heterogeneous model with “picker speed”, “time of day”, and “picking cart” factors. The model had a prediction error of 134.9 s based on 10-fold CV, and 136.8 s based on leave-one-out CV (LOOCV). The selected model predicts a priori mean and standard deviation of picking times for any given combination of factor levels. For instance, if picker speed is 'fast', the time of day is 'morning', and the picking cart is 'standard', the marginal predicted picking time is 477.1 ± 42.4 s. The proposed methodology and model structures offer a practical tool for strawberry picking time modelling, which could also be applied to other manually harvested specialty crops such as raspberries, cherry tomatoes, and table grapes.