SMOTE-based method for balanced spectral nondestructive detection of moldy apple core

Abstract

Moldy apple core is an internal fruit disease that poses a threat to consumer health. In this study, a synthetic minority over-sampling technology (SMOTE) based model of moldy apple core was proposed to solve the problem of poor model performance due to imbalanced sample distribution in the spectral nondestructive detection of moldy apple core. Two different methods, random under-sampling (RUS) and SMOTE, were used to balance the original samples. A hybrid strategy of competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) was used for spectral variable selection, and partial least squares-discriminant analysis (PLS-DA) classification model and least squares-support vector machine (LS-SVM) nonlinear classification model were constructed. The SMOTE-based method was better than the RUS method in solving the problem of imbalanced sample distribution. The constructed SMOTE-CARS-SPA-LS-SVM model could not only accurately identify the healthy and diseased apples with a disease degree of >10% but also improve the detection accuracy of slightly moldy apple core. In addition, the method has certain practical applicability in external verification. • A hybrid strategy for spectral variable selection. • Unbalanced sample distribution affects the performance of the model. • The SMOTE-based method is superior to the RUS method. • The SMOTE-based method can improve the detection accuracy of moldy apple.