Contactless estimation of apple fruit respiration rate using machine learning models based on laser speckle imaging
Piotr M. Pieczywek, Artur Nosalewicz, Artur Zdunek
Abstract
This article presents the application of the laser speckle imaging method, a nondestructive and contactless method monitoring the gas exchange rate of intact apples. Red and infrared lasers in a typical setup for laser speckle imaging were used. A new parameter, speckle pattern relaxation time, has been proposed to evaluate speckle dynamics. Machine learning methods were used to develop a set of predictive models calibrated and validated against the respiration rate of two apple fruit cultivars measured with the flush system. The model with the highest performance used three variables: the speckle pattern relaxation time (τRED or τIR), fruit mass, and categorical variables describing apple varieties. This model provided satisfactorily low values of mean absolute prediction errors of 6.04%. Data from laser light scattering measurements combined with modern machine learning algorithms provided a nondestructive and fast method for estimating the apple fruit respiration rate. The developed solution has the potential for a wide range of industrial applications, especially in fruit storage, where the fruit respiration rate indicates optimal storage conditions.