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

Abstract

The healthy development of maize seed industry plays a key role in the effective supply of agricultural products, ensuring national food security. Seed thermal damage has an important impact on crop yield, seed vitality and nutritional value which is of great significance to distinguish maize seed germination before sowing. In the study, 100 thermal-damaged and 100 normal maize seeds were selected to collect spectral data based on hyperspectral imaging system and divided the training set and test set according to 3:1. The spectral information in the range of 963.27-1698.75nm was used for the subsequent study. Multiplicative scatter correction (MSC) and standard normal transform (SNV) were used to pretreat the original spectral data and support vector machine (SVM) model was established. Competitive adaptive reweighted sampling (CARS) and uninformative variables elimination (UVE) were used to reduce the dimension of full spectral features in order to further simplify the prediction model. Genetic algorithm (GA) and dung beetle optimizer (DBO) were used to optimize the parameters penalty coefficient c and kernel function g of SVM model to make the prediction results more accurate. The results showed that SNV-CARS-DBO-SVM model had the best performance with 92.00% of the prediction accuracy and running time of 3.22 seconds which provided an idea for non-destructive, efficient and accurate prediction of germination of thermal-damaged maize seeds.