Detection of the amylose and amylopectin contents of rice by hyperspectral imaging combined with a CNN-AdaBoost model

Abstract

The amylose and amylopectin contrnts of rice directly influence the flavor and texture of liquor from which it is brewed. This study utilized hyperspectral imaging (HSI) along with an integrated learning model to rapidly and non-destructively analyze the amylose and amylopectin contents in rice. In this study, the characteristic wavelengths were extracted using the interval random frog (iRF) algorithm, the successive projection algorithm (SPA), and a combination of the two algorithms (iRF-SPA). Afterward, color features of the rice were extracted, and models were developed to predict amylose and amylopectin contents using full wavelengths, characteristic wavelengths, and fused data with color features. These models included partial least squares regression (PLSR), convolutional neural networks (CNN), and convolutional neural network-based ensemble learning (CNN-AdaBoost). The results showed that the CNN-AdaBoost model built using the fusion data of the feature wavelengths and the color features extracted by the iRF-SPA method was the best predictor of rice amylose and amylopectin contents, with prediction accuracies of ${R_p}^2$ of 0.9931 and 0.9889, respectively. The study showed that HSI combined with the CNN-AdaBoost model enables rapid, non-destructive analysis of amylose and amylopectin contents in rice, offering a practical basis for assessing the chemical composition of raw food materials.