Non-destructive detection of chilling injury in kiwifruit using a dual-laser scanning system with a principal component analysis - back propagation neural network

Abstract

As a physiological disorder, chilling injury in kiwifruit may develop when the fruit are stored for long periods at a low storage temperature of 0–1°C. Presence of the disorder, inconsistent with marketing requirements for high-quality fruit, may lead to substantial financial and reputational losses. Thus, early detection or removal of chill-damaged fruit is desirable. This study demonstrates a novel dual-laser scanning system which has potential to be developed into a fast online system for the detection of chilling injury in Actinidia chinensis var. chinensis ‘Zesy002’ kiwifruit. The system consists of two laser modules at 730 and 880 nm wavelengths, a scanning mechanism and two detectors at partial (90°) and full (180°) light transmission. A sample of 231 kiwifruit was used to prove the concept, including 80 sound and 151 chill-damaged fruit of three different severity categories (slight, moderate and severe). A principal component analysis – back propagation neural network was used to classify fruit with 5-fold cross-validation. A comparison was made with standard visible-near infrared (Vis-NIR) interactance spectroscopy used to classify the same fruit using the same modelling algorithm. The dual-laser scanning system showed a slightly higher binary classification accuracy than the Vis-NIR spectroscopy, with an average accuracy of 95% for distinguishing sound and chill-damaged fruit. The classification error rate was 0% for severe damaged fruit. These experimental results demonstrate the potential of this dual-laser scanning system for the detection of chill-damaged fruit. The setup using only two wavelengths, its unique scanning operation and flexible system layout make it practical and attractive for future development for application on high-speed fruit graders.