Early identification of apple bitter pit using hyperspectral imaging technology

Abstract

Bitter pit is a physiological disorder that severely affects apple quality and consumer satisfaction. This study explored hyperspectral imaging (400–1000 nm) for the early detection of bitter pit in "Qin crisp" apples during storage. Standard Normal Variate (SNV), Multivariate Scatter Correction (MSC), Savitzky-Golay Smoothing Filter (SG), First Derivative (1st-D), and Second Derivative (2nd-D) preprocessing methods were applied. Random Frog (RF) and Genetic Algorithm (GA) were used to filter the characteristic wavelength spectral and texture information, and the Support Vector Machine (SVM) classification model was consequently established. The MSC-SVM model achieved spectral-based accuracies of 98.15 % (training) and 86.11 % (testing), while the variance-RF-SVM texture-based accuracies of training and testing sets were 98.55 % and 93.33 %, respectively. Hyperspectral imaging demonstrated potential for the early detection of bitter pit, providing technical and theoretical references for reducing loss and improving apple quality.