Identification of red jujube varieties based on hyperspectral imaging technology combined with CARS-IRIV and SSA-SVM

Abstract

To identify the varieties of red jujube rapidly and nondestructively, hyperspectral imaging (HSI) technology was applied in this article. Hyperspectral data of 480 samples with four different varieties were acquired in the range of 400.68–1001.61 nm. First, Savitzky–Golay and standard normal variable were utilized to process raw spectra. Afterward, a novel method combining competitive adaptive reweighted sampling and iterative retained information variable (CARS-IRIV) was proposed to select feature wavelengths. The support vector machine (SVM) modeling results indicated that CARS-IRIV had better information extraction performance and simplified the model. Finally, to further improve the accuracy, sparrow search algorithm (SSA) was adopted to optimize the parameters (c, g) of SVM. The results showed that SSA-SVM exhibited greater accuracy than other compared models, and the accuracy of training and test sets were 100% and 96.68%, respectively. It confirmed that HSI technology coupled with CARS-IRIV-SSA-SVM can effectively identify varieties of red jujube.

Practical Applications

The traditional ways of classifying red jujube varieties are destructive and laborious. Therefore, HSI technology was adopted to overcome the above shortcomings. In this article, the best identification performance was based on the CARS-IRIV-SSA-SVM model with an identification accuracy of 96.68%. This study is helpful for the identification of other agricultural product varieties by HSI technology.