Measurement orientation compensation and comparison of transmission spectroscopy for online detection of moldy apple core

Abstract

Moldy core is an internal disease afflicting otherwise healthy apples. This paper reports the results of a study on the effect of fruit orientation on the online detection of moldy core in Fuji apples using Vis/NIR transmission spectroscopy. Using a self-designed online spectral measurement system, transmission spectra in the range 550–900 nm were collected from apples positioned in three orientations: vertical calyx-stem axis with stem upward (T1), horizontal calyx-stem axis (T2), and vertical calyx-stem axis with stem downward (T3). In order to evaluate the spectral stability of the system in different orientations, the area change rate (ACR) and signal-to-noise ratio (SNR) were introduced. The propagation of light inside the fruit was also analyzed. After excluding outliers, 244 apple samples were divided into a training set (113 healthy apples and 70 diseased apples) and a test set (38 healthy apples and 23 diseased apples) at a ratio of 3:1. After spectral pretreatment, a support vector machine (SVM) was used to develop compensation models for the detection of moldy core in apples at all orientations (global model) and each separate orientation (local models). The results show that T2 provides the best performance in our system, and that the measurement orientation significantly affects the recognition accuracy of moldy core in apples. In contrast, the global model is insensitive to changes in fruit orientation, and the recognition accuracy of the global SVM model is 100% in all three orientations. Thus, it is considered that the global SVM model is a promising candidate for the online detection of moldy core in apples using transmission spectroscopy, regardless of the spectral measurement orientation.