Hyperspectral imaging combined with NGO-RBFNN for maize variety identification

Abstract

Maize is a vital food crop with various varieties cultivated in the world. The market order is significantly threatened by the prevalence of counterfeit and substandard maize seeds. The development of non-destructive methods for accurately identifying maize varieties is necessary. Hyperspectral imaging technology was utilized to acquire spectral data. 540 maize seeds of 6 varieties were divided into training set and test set in a ratio of 2:1. Regions of interest (ROI) with embryo size of 8 × 8 pixels were designated. The average spectral information in the range of 949.43–1709.49 nm was intercepted to eliminate the random noise at both ends of the raw spectral data. Savitzky-Golay (SG) smoothing preprocessing was used on the effective band information, and max normalization (MN) preprocessing was performed on the basis of SG. The characteristic wavelengths were screened using Successive Projections Algorithm (SPA), Competitive Adaptive Reweighted Sampling (CARS) for single screening, and CARS-SPA and CARS + SPA for combined screening. Based on full bands (FB) and characteristic wavelengths, Radial Basis Function Neural Network (RBFNN), Back Propagation Neural Network (BPNN), Extreme Learning Machine (ELM), Random Forest (RF), Support Vector Machine (SVM) were developed. RBFNN were optimized by Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Northern Goshawk Optimization (NGO). The results showed that the (SG + MN)-(CARS + SPA)-NGO-RBFNN model had the best performance with an accuracy of 93.89 % in the test set. The research proved that hyperspectral imaging combined with NGO-RBFNN can effectively identify various maize varieties, which provides a theoretical foundation for the identification of maize varieties.