Non-destructive detection and grading of chilling injury-induced lignification of kiwifruit using X-ray computer tomography and machine learning

The postharvest quality of kiwifruit is primarily maintained through low-temperature storage. However, the detrimental effects of low-temperature storage on kiwifruit include the chilling injury-induced lignification (CIIL) of the subcutaneous tissue. Also bruising during storage and transportation inevitably occurs, leading to the loss of value of kiwifruit. In this study, X-ray computed tomography (CT) imaging combined with machine learning was used to identify internal defects, including CIIL and bruising, and to grade 'Jinyan' kiwifruit on the basis of CIIL. Feature parameters, such as the number of defects, coordination number, sphericity, length/width, and average volume of defects, were extracted from CT images and used to construct linear discriminant analysis (LDA), support vector machines (SVM) and random forest (RF) models that identified healthy, bruised, bruised and CIIL, and CIIL kiwifruit with mean area under the receiver operating characteristic curve (AUC) of 0.91, 0.93 and 0.94, respectively. The feature parameters associated with CIIL indices were arranged by Pearson correlation coefficient combined with heat map. The best RF model with an AUC of 0.93 was obtained using the first five highly correlated parameters, i.e., the mean gray value of the volume, the difference value, the volume of the lignified fraction (VLF), the volume of the defect, and the shape factor, while the optimal LDA and SVM models were obtained from the first four parameters with AUCs of 0.91 and 0.95, respectively. Additionally, the VLF increased during the shelf life of kiwifruit with severe CIIL, reflecting the intolerance to the storage conditions. This is probably due to the fact that the CIIL tissue absorbed water from surrounding normal tissue and produced larger volume of lower gray scale value similar to the lignified tissue. In contrast, the kiwifruit with mild or moderate CIIL had no significant changes in VLF and retained their commercial value. In conclusion, combining X-ray CT and machine learning represents a new way to identify and grade kiwifruit defects.