Identification of living and non-living watermelon seeds based on Hyperspectral Imaging Technology

Abstract

Seeds storage has been a global challenge for their conservation, and it is crucial to test their germination ability before the operation. Nondestructive detection (NDD) techniques are frequently used in testing and evaluating elements. In this work, the hyperspectral image technology is employed to distinguish the living and nonliving watermelon seeds. First, fifty collected watermelon seeds were kept at 45°C for 72 hours, and another fifty seeds were stored in dry bottle at 20 °C. Hyperspectral images of 100 samples were collected. Mean spectral data was calculated with the range of 400~ 1000nm from region of interest (ROI). Then, Savitzky-Golay (SG) and standard normalized variable (SNV) approaches were utilized to preprocess the spectral data, and principal component analysis (PCA) was used to select intervals to pick the top principal components (PCs). Moreover, a model based on support vector machine optimized by grey wolf algorithm (GWO-SVM) was introduced in this paper. Compared with normal SVM, the proposed scheme is tested and verified using the seed data with a classification rate improved from 87.5% to 97.5%. The overall results showed that nondestructive technic with SVM classification tool could be used in identification of watermelon seeds.