Year classification of high-oleic peanut seeds based on hyperspectral hybrid bands selection method

Abstract

The storage years of seeds have a significant impact on high-oleic peanut seed vigor and quality. Therefore, it is essential to identify different storage-year seeds for planting, direct consumption, industrial processing, and marketing. In this study, hyperspectral images with 616 spectral bands (from visible light to near-infrared) were employed to classify different storage-year peanut seeds. To extract characteristic information for classification, we proposed a hybrid band selection (HBS) method based on the successive projection algorithm (SPA) by fusing the color-sensitive bands and moisture-sensitive bands. Then three classifiers, support vector machine (SVM), extreme learning machine (ELM), and K-nearest neighbors (KNN), were selected for storage-year classification. The experimental results demonstrated that the features extracted with the HBS method can obtain higher classification accuracy than other methods'. Specifically, the HBS-ELM model achieved the highest classification performance, with accuracy of 90.22%.