Establishment of evaluation criterion based on starch dyeing method and implementation of optical and acoustic techniques for postharvest determination of “HongYang” kiwifruit ripeness

Abstract

Realizing online detection of ripeness is meaningful and profitable in existing postharvest sorting system. This paper aims to establish evaluation criterion of kiwifruit ripeness based on starch dyeing method and realize online detection of kiwifruit ripeness based on spectroscopy and acoustic vibration. Considering storage time, flavor, and edible status of kiwifruit, evaluation criterion of kiwifruit ripeness was established based on starch conversion rate obtained from starch staining images. Differences were found in quality attributes (i.e., SSC and firmness), microstructure, absorbance spectra, and frequency-domain signals of kiwifruit at different ripeness levels. Based on spectral and acoustic vibration data collected by self-designed online systems, ripeness determination models were established. Linear PLSDA model performed better in spectral classification, while nonlinear SVM model performed better in acoustic vibration classification. Deep 1D-CNN models were proposed to further improve performance of ripeness determination models. Further novelty of this work was that it externally tested robustness of ripeness determination models using independent sample set with biological variability. For new batch of samples, overall accuracy of ripeness determination model based on spectral data (Accuracy = 93.08%) was slightly better than that of ripeness determination model based on acoustic vibration data (Accuracy = 92.31%). External test results showed that recognition performance of deep 1D-CNN model based on spectral data and deep 1D-CNN model based on acoustic vibration data was satisfactory. Deep 1D-CNN models proposed in this study can be used for determination of kiwifruit at different ripeness levels and have strong practicability in commercial online sorting.