Hyperspectral imaging combined with DBO-SVM for the germination prediction of thermally damaged seeds

Abstract

Healthy development of the maize seed industry plays a key role in the effective supply of agricultural products and ensures national food security. Thermal damage to seeds significantly affects crop yield, seed vitality and nutritional value, making it crucial to identify maize seed germination potential before sowing. In this study, 100 thermally damaged and 100 normal maize seeds were selected, and spectral data were collected using a hyperspectral imaging system. The samples were divided into training and test sets in a 3 : 1 ratio. Spectral information in the range of 963.27–1698.75 nm was used for subsequent studies. Multiplicative scatter correction (MSC) and standard normal transform (SNV) methods were used to pretreat the original spectral data, and a support vector machine (SVM) model was established. Competitive adaptive reweighted sampling (CARS) and uninformative variables elimination (UVE) methods were used to reduce the dimensions of the full spectral features to further simplify the prediction model. In addition, the genetic algorithm (GA) and dung beetle optimizer (DBO) were used to optimize the penalty coefficient c and kernel function g parameters of the SVM model to enhance the prediction accuracy. Results showed that the SNV-CARS-DBO-SVM model achieved the best performance with a prediction accuracy of 92.00% and a running time of 3.22 seconds, providing a strategy for the non-destructive, efficient and accurate prediction of the germination of thermally damaged maize seeds.